pysimplesql.py

"""
## DISCLAIMER: While **pysimplesql** works with and was inspired by the excellent
PySimpleGUI™ project, it has no affiliation.

## Rapidly build and deploy database applications in Python **pysimplesql** binds
PySimpleGUI to various databases for rapid, effortless database application
development. Makes a great replacement for MS Access or LibreOffice Base! Have the
full power and language features of Python while having the power and control of
managing your own codebase. **pysimplesql** not only allows for super simple
automatic control (not one single line of SQL needs written to use **pysimplesql**),
but also allows for very low level control for situations that warrant it.

----------------------------------------------------------------------------------------
NAMING CONVENTIONS USED THROUGHOUT THE SOURCE CODE
----------------------------------------------------------------------------------------
There is a lot of ambiguity with database terminology, as many terms are used
interchangeably in some circumstances, but not in others.  The Internet has post after
post debating this topic.  See one example here:
https://dba.stackexchange.com/questions/65609/column-vs-field-have-i-been-using-these-terms-incorrectly  # fmt: skip
To avoid confusion in the source code, specific naming conventions will be used whenever
possible.

Naming conventions can fall under 4 categories:
- referencing the database (variables, functions, etc. that relate to the database)
- referencing the `DataSet` (variables, functions, etc. that relate to the `DataSet`)
- referencing pysimplesql
- referencing PySimpleGUI

- Database related:
    driver - a `SQLDriver` derived class
    t, table, tables - the database table name(s)
    r, row, rows - A group of related data in a table
    c, col, cols, column, columns -  A set of values of a particular type
    q, query - An SQL query string
    domain - the data type of the data (INTEGER, TEXT, etc.)

 - `DataSet` related:
    r, row, rows, df - A row, or collection of rows from  querying the database
    c, col, cols, column, columns -  A set of values of a particular type
    Record - A collection of fields that make up a row
    field - the value found where a row intersects a column

- pysimplesql related
    frm - a `Form` object
    dataset, datasets - a `DataSet` object, or collection of `DataSet` objects
    data_key - the key (name) of a dataset object

- PySimpleGUI related
    win, window - A PySimpleGUI Window object
    element - a Window element
    element_key -  a window element key
----------------------------------------------------------------------------------------
"""  # noqa: E501

from __future__ import annotations  # docstrings

import asyncio
import contextlib
import functools
import logging
import math
import os.path
import queue
import threading  # threaded popup
from datetime import date, datetime
from time import sleep, time  # threaded popup
from typing import Callable, Dict, List, Optional, Tuple, Type, TypedDict, Union

import pandas as pd
import PySimpleGUI as sg

# Wrap optional imports so that pysimplesql can be imported as a single file if desired:
with contextlib.suppress(ModuleNotFoundError, ImportError):
    from .language_pack import *  # noqa F403

with contextlib.suppress(ModuleNotFoundError, ImportError):
    from .theme_pack import *  # noqa F403

try:
    from .reserved_sql_keywords import ADAPTERS as RESERVED
except (ModuleNotFoundError, ImportError):
    # Use common as minimum default
    # fmt: off
    RESERVED = {
        "common": [
            "SELECT", "INSERT", "DELETE", "UPDATE", "DROP", "CREATE", "ALTER", "WHERE",
            "FROM", "INNER", "JOIN", "AND", "OR", "LIKE", "ON", "IN", "SET", "BY",
            "GROUP", "ORDER", "LEFT", "OUTER", "IF", "END", "THEN", "LOOP", "AS",
            "ELSE", "FOR", "CASE", "WHEN", "MIN", "MAX", "DISTINCT",
        ]
    }
    # fmt: on

logger = logging.getLogger(__name__)

# ---------------------------
# Types for automatic mapping
# ---------------------------
TYPE_RECORD: int = 1
TYPE_SELECTOR: int = 2
TYPE_EVENT: int = 3

# -----------------
# Transform actions
# -----------------
TFORM_ENCODE: int = 1
TFORM_DECODE: int = 0

# -----------
# Event types
# -----------
# Custom events (requires 'function' dictionary key)
EVENT_FUNCTION: int = 0
# DataSet-level events (requires 'table' dictionary key)
EVENT_FIRST: int = 1
EVENT_PREVIOUS: int = 2
EVENT_NEXT: int = 3
EVENT_LAST: int = 4
EVENT_SEARCH: int = 5
EVENT_INSERT: int = 6
EVENT_DELETE: int = 7
EVENT_DUPLICATE: int = 13
EVENT_SAVE: int = 8
EVENT_QUICK_EDIT: int = 9
# Form-level events
EVENT_SEARCH_DB: int = 10
EVENT_SAVE_DB: int = 11
EVENT_EDIT_PROTECT_DB: int = 12

# ----------------
# GENERIC BITMASKS
# ----------------
# Can be used with other bitmask values
SHOW_MESSAGE: int = 4096

# ---------------------------
# PROMPT_SAVE RETURN BITMASKS
# ---------------------------
PROMPT_SAVE_DISCARDED: int = 1
PROMPT_SAVE_PROCEED: int = 2
PROMPT_SAVE_NONE: int = 4

# ---------------------------
# PROMPT_SAVE MODES
# ---------------------------
PROMPT_MODE: int = 1
AUTOSAVE_MODE: int = 2

# ---------------------------
# RECORD SAVE RETURN BITMASKS
# ---------------------------
SAVE_FAIL: int = 1  # Save failed due to callback
SAVE_SUCCESS: int = 2  # Save was successful
SAVE_NONE: int = 4  # There was nothing to save

# ----------------------
# SEARCH RETURN BITMASKS
# ----------------------
SEARCH_FAILED: int = 1  # No result was found
SEARCH_RETURNED: int = 2  # A result was found
SEARCH_ABORTED: int = 4  # The search was aborted, likely during a callback
SEARCH_ENDED: int = 8  # We have reached the end of the search

# ----------------------------
# DELETE RETURNS BITMASKS
# ----------------------------
DELETE_FAILED: int = 1  # No result was found
DELETE_RETURNED: int = 2  # A result was found
DELETE_ABORTED: int = 4  # The search was aborted, likely during a callback
DELETE_RECURSION_LIMIT_ERROR: int = 8  # We hit max nested levels

# Mysql sets this as 15 when using foreign key CASCADE DELETE
DELETE_CASCADE_RECURSION_LIMIT: int = 15

# -------
# Sorting
# -------
SORT_NONE = 0
SORT_ASC = 1
SORT_DESC = 2


# -------
# CLASSES
# -------
# TODO: Combine TableRow and ElementRow into one class for simplicity
class TableRow(list):

    """
    Convenience class used by Tables to associate a primary key with a row of data.

    Note: This is typically not used by the end user.
    """

    def __init__(self, pk: int, *args, **kwargs):
        self.pk = pk
        super().__init__(*args, **kwargs)

    def __str__(self):
        return str(self[:])

    def __int__(self):
        return self.pk

    def __repr__(self):
        # Add some extra information that could be useful for debugging
        return f"TableRow(pk={self.pk}): {super().__repr__()}"


class ElementRow:

    """
    Convenience class used by listboxes and comboboxes to associate a primary key with
    a row of data.

    Note: This is typically not used by the end user.
    """

    def __init__(self, pk: int, val: Union[str, int]) -> None:
        self.pk = pk
        self.val = val

    def __repr__(self):
        return str(self.val)

    def __str__(self):
        return str(self.val)

    def __int__(self):
        return self.pk

    def get_pk(self):
        # Return the primary key portion of the row
        return self.pk

    def get_val(self):
        # Return the value portion of the row
        return self.val

    def get_instance(self):
        # Return this instance of the row
        return self


class Relationship:

    """
    Used to track primary/foreign key relationships in the database.

    See the following for more information: `Form.add_relationship` and
    `Form.auto_add_relationships`.

    Note: This class is not typically used the end user,
    """

    # store our own instances
    instances = []

    @classmethod
    def get_relationships(cls, table: str) -> List[Relationship]:
        """
        Return the relationships for the passed-in table.

        :param table: The table to get relationships for
        :returns: A list of @Relationship objects
        """
        return [r for r in cls.instances if r.child_table == table]

    @classmethod
    def get_update_cascade_tables(cls, table: str) -> List[str]:
        """
        Return a unique list of the relationships for this table that should requery
        with this table.

        :param table: The table to get cascaded children for
        :returns: A unique list of table names
        """
        rel = [
            r.child_table
            for r in cls.instances
            if r.parent_table == table and r.on_update_cascade
        ]
        # make unique
        return list(set(rel))

    @classmethod
    def get_delete_cascade_tables(cls, table: str) -> List[str]:
        """
        Return a unique list of the relationships for this table that should be deleted
        with this table.

        :param table: The table to get cascaded children for
        :returns: A unique list of table names
        """
        rel = [
            r.child_table
            for r in cls.instances
            if r.parent_table == table and r.on_delete_cascade
        ]
        # make unique
        return list(set(rel))

    @classmethod
    def get_parent(cls, table: str) -> Union[str, None]:
        """
        Return the parent table for the passed-in table.

        :param table: The table (str) to get relationships for
        :returns: The name of the Parent table, or None if there is none
        """
        for r in cls.instances:
            if r.child_table == table and r.on_update_cascade:
                return r.parent_table
        return None

    @classmethod
    def parent_virtual(cls, table: str, frm: Form) -> Union[bool, None]:
        """
        Return True if current row of parent table is virtual.

        :param table: The table (str) to get relationships for
        :param frm: Form reference
        :returns: True if current row of parent table is virtual
        """
        for r in cls.instances:
            if r.child_table == table and r.on_update_cascade:
                try:
                    return frm[r.parent_table].row_is_virtual()
                except AttributeError:
                    return False
        return None

    @classmethod
    def get_update_cascade_fk_column(cls, table: str) -> Union[str, None]:
        """
        Return the cascade fk that filters for the passed-in table.

        :param table: The table name of the child
        :returns: The name of the cascade-fk, or None
        """
        for r in cls.instances:
            if r.child_table == table and r.on_update_cascade:
                return r.fk_column
        return None

    @classmethod
    def get_delete_cascade_fk_column(cls, table: str) -> Union[str, None]:
        """
        Return the cascade fk that filters for the passed-in table.

        :param table: The table name of the child
        :returns: The name of the cascade-fk, or None
        """
        for r in cls.instances:
            if r.child_table == table and r.on_delete_cascade:
                return r.fk_column
        return None

    def __init__(
        self,
        join_type: str,
        child_table: str,
        fk_column: Union[str, int],
        parent_table: str,
        pk_column: Union[str, int],
        update_cascade: bool,
        delete_cascade: bool,
        driver: SQLDriver,
        frm: Form,
    ) -> None:
        """
        Initialize a new Relationship instance.

        :param join_type: The join type. I.e. "LEFT JOIN", "INNER JOIN", etc.
        :param child_table: The table name of the child table
        :param fk_column: The child table's foreign key column
        :param parent_table: The table name of the parent table
        :param pk_column: The parent table's primary key column
        :param update_cascade: True if the child's fk_column ON UPDATE rule is 'CASCADE'
        :param delete_cascade: True if the child's fk_column ON DELETE rule is 'CASCADE'
        :param driver: A `SQLDriver` instance
        :param frm: A Form instance
        :returns: None
        """
        self.join_type = join_type
        self.child_table = child_table
        self.fk_column = fk_column
        self.parent_table = parent_table
        self.pk_column = pk_column
        self.update_cascade = update_cascade
        self.delete_cascade = delete_cascade
        self.driver = driver
        self.frm = frm
        Relationship.instances.append(self)

    @property
    def on_update_cascade(self):
        return bool(self.update_cascade and self.frm.update_cascade)

    @property
    def on_delete_cascade(self):
        return bool(self.delete_cascade and self.frm.delete_cascade)

    def __str__(self):
        """Return a join clause when cast to a string."""
        return self.driver.relationship_to_join_clause(self)

    def __repr__(self):
        """Return a more descriptive string for debugging."""
        return (
            f"Relationship ("
            f"\n\tjoin={self.join_type},"
            f"\n\tchild_table={self.child_table},"
            f"\n\tfk_column={self.fk_column},"
            f"\n\tparent_table={self.parent_table},"
            f"\n\tpk_column={self.pk_column}"
            f"\n)"
        )


class ElementMap(dict):

    """
    Map a PySimpleGUI element to a specific `DataSet` column.

    This is what makes the GUI automatically update to the contents of the database.
    This happens automatically when a PySimpleGUI Window is bound to a `Form` by using
    the bind parameter of `Form` creation, or by executing `Form.auto_map_elements()` as
    long as the Table.column naming convention is used, This method can be used to
    manually map any element to any `DataSet` column regardless of naming convention.
    """

    def __init__(
        self,
        element: sg.Element,
        dataset: DataSet,
        column: str,
        where_column: str = None,
        where_value: str = None,
    ) -> None:
        """
        Create a new ElementMap instance.

        :param element: A PySimpleGUI Element
        :param dataset: A `DataSet` object
        :param column: The name of the column to bind to the element
        :param where_column: Used for key, value shorthand
        :param where_value: Used for key, value shorthand
        :returns: None
        """
        super().__init__()
        self["element"] = element
        self["dataset"] = dataset
        self["table"] = dataset.table
        self["column"] = column
        self["where_column"] = where_column
        self["where_value"] = where_value

    def __getattr__(self, key: str):
        try:
            return self[key]
        except KeyError:
            raise KeyError(f"ElementMap has no key {key}.")

    def __setattr__(self, key, value):
        self[key] = value


class DataSet:

    """
    `DataSet` objects are used for an internal representation of database tables.

    `DataSet` instances are added by the following `Form` methods: `Form.add_table`,
    `Form.auto_add_tables`. A `DataSet` is synonymous for a SQL Table (though you can
    technically have multiple `DataSet` objects referencing the same table, with each
    `DataSet` object having its own sorting, where clause, etc.).
    Note: While users will interact with DataSet objects often in pysimplesql, they
    typically aren't created manually by the user.
    """

    instances = []  # Track our own instances

    def __init__(
        self,
        data_key: str,
        frm_reference: Form,
        table: str,
        pk_column: str,
        description_column: str,
        query: Optional[str] = "",
        order_clause: Optional[str] = "",
        filtered: bool = True,
        prompt_save: int = None,
        save_quiet: bool = None,
        duplicate_children: bool = None,
    ) -> None:
        """
        Initialize a new `DataSet` instance.

        :param data_key: The name you are assigning to this `DataSet` object (I.e.
            'people').
        :param frm_reference: This is a reference to the @ Form object, for convenience
        :param table: Name of the table
        :param pk_column: The name of the column containing the primary key for this
            table.
        :param description_column: The name of the column used for display to users
            (normally in a combobox or listbox).
        :param query: You can optionally set an initial query here. If none is provided,
            it will default to "SELECT * FROM {table}"
        :param order_clause: The sort order of the returned query. If none is provided
            it will default to "ORDER BY {description_column} ASC"
        :param filtered: (optional) If True, the relationships will be considered and an
            appropriate WHERE clause will be generated. False will display all records
            in the table.
        :param prompt_save: (optional) Default: Mode set in `Form`. Prompt to save
            changes when dirty records are present. There are two modes available,
            (if pysimplesql is imported as `ss`) use:
            `ss.PROMPT_MODE` to prompt to save when unsaved changes are present.
            `ss.AUTOSAVE_MODE` to automatically save when unsaved changes are present.
        :param save_quiet: (optional) Default: Set in `Form`. True to skip info popup on
            save. Error popups will still be shown.
        :param duplicate_children: (optional) Default: Set in `Form`. If record has
            children, prompt user to choose to duplicate current record, or both.
        :returns: None
        """
        DataSet.instances.append(self)
        self.driver = frm_reference.driver
        # No query was passed in, so we will generate a generic one
        if not query:
            query = self.driver.default_query(table)
        # No order was passed in, so we will generate generic one
        if not order_clause:
            order_clause = self.driver.default_order(description_column)

        self.key: str = data_key
        self.frm: Form = frm_reference
        self._current_index: int = 0
        self.table: str = table
        self.pk_column: str = pk_column
        self.description_column: str = description_column
        self.query: str = query
        self.order_clause: str = order_clause
        self.join_clause: str = ""
        self.where_clause: str = ""  # In addition to the generated where clause!
        self.dependents: list = []
        self.column_info: ColumnInfo  # ColumnInfo collection
        self.rows: Union[pd.DataFrame, None] = None
        self.search_order: List[str] = []
        self.selector: List[str] = []
        self.callbacks: CallbacksDict = {}
        self.transform: Optional[Callable[[pd.DataFrame, int], None]] = None
        self.filtered: bool = filtered
        if prompt_save is None:
            self._prompt_save = self.frm._prompt_save
        else:
            self._prompt_save: int = prompt_save
        if save_quiet is None:
            self.save_quiet = self.frm.save_quiet
        else:
            self.save_quiet: bool = save_quiet
        if duplicate_children is None:
            self.duplicate_children = self.frm.duplicate_children
        else:
            self.duplicate_children: bool = duplicate_children
        self._simple_transform: SimpleTransformsDict = {}

    # Override the [] operator to retrieve columns by key
    def __getitem__(self, key: str):
        return self.get_current(key)

    # Make current_index a property so that bounds can be respected
    @property
    def current_index(self):
        return self._current_index

    @current_index.setter
    # Keeps the current_index in bounds
    def current_index(self, val: int):
        if val > len(self.rows) - 1:
            self._current_index = len(self.rows) - 1
        elif val < 0:
            self._current_index = 0
        else:
            self._current_index = val

    @classmethod
    def purge_form(cls, frm: Form, reset_keygen: bool) -> None:
        """
        Purge the tracked instances related to frm.

        :param frm: the `Form` to purge `DataSet`` instances from
        :param reset_keygen: Reset the keygen after purging?
        :returns: None
        """
        new_instances = []
        selector_keys = []

        for dataset in DataSet.instances:
            if dataset.frm != frm:
                new_instances.append(dataset)
            else:
                logger.debug(
                    f"Removing DataSet {dataset.key} related to "
                    f"{frm.driver.__class__.__name__}"
                )
                # we need to get a list of elements to purge from the keygen
                for s in dataset.selector:
                    selector_keys.append(s["element"].key)

        # Reset the keygen for selectors and elements from this Form
        # This is probably a little hack-ish, perhaps I should relocate the keygen?
        if reset_keygen:
            for k in selector_keys:
                keygen.reset_key(k)
            keygen.reset_from_form(frm)
        # Update the internally tracked instances
        DataSet.instances = new_instances

    def set_prompt_save(self, mode: int) -> None:
        """
        Set the prompt to save action when navigating records.

        :param mode: a constant value. If pysimplesql is imported as `ss`, use:
            - `ss.PROMPT_MODE` to prompt to save when unsaved changes are present.
            - `ss.AUTOSAVE_MODE` to automatically save when unsaved changes are present.
        :returns: None
        """
        self._prompt_save = mode

    def set_search_order(self, order: List[str]) -> None:
        """
        Set the search order when using the search box.

        This is a list of column names to be searched, in order

        :param order: A list of column names to search
        :returns: None
        """
        self.search_order = order

    def set_callback(
        self, callback: str, fctn: Callable[[Form, sg.Window], bool]
    ) -> None:
        """
        Set DataSet callbacks. A runtime error will be thrown if the callback is not
        supported.

        The following callbacks are supported:
            before_save   called before a record is saved. The save will continue if the
                callback returns true, or the record will rollback if the callback
                returns false.
            after_save    called after a record is saved. The save will commit to the
                database if the callback returns true, else it will rollback the
                transaction
            before_update Alias for before_save
            after_update  Alias for after_save
            before_delete called before a record is deleted.  The delete will move
                forward if the callback returns true, else the transaction will rollback
            after_delete  called after a record is deleted. The delete will commit to
                the database if the callback returns true, else it will rollback the
                transaction
            before_duplicate called before a record is duplicate.  The duplicate will
                move forward if the callback returns true, else the transaction will
                rollback
            after_duplicate  called after a record is duplicate. The duplicate will
                commit to the database if the callback returns true, else it will
                rollback the transaction
            before_search called before searching.  The search will continue if the
                callback returns True
            after_search  called after a search has been performed.  The record change
                will undo if the callback returns False
            record_changed called after a record has changed (previous,next, etc.)

        :param callback: The name of the callback, from the list above
        :param fctn: The function to call.  Note, the function must take in two
            parameters, a `Form` instance, and a `PySimpleGUI.Window` instance, and
            return True or False
        :returns: None
        """
        logger.info(f"Callback {callback} being set on table {self.table}")
        supported = [
            "before_save",
            "after_save",
            "before_delete",
            "after_delete",
            "before_duplicate",
            "after_duplicate",
            "before_update",
            "after_update",  # Aliases for before/after_save
            "before_search",
            "after_search",
            "record_changed",
        ]
        if callback in supported:
            # handle our convenience aliases
            callback = "before_save" if callback == "before_update" else callback
            callback = "after_save" if callback == "after_update" else callback
            self.callbacks[callback] = fctn
        else:
            raise RuntimeError(f'Callback "{callback}" not supported.')

    def set_transform(self, fn: callable) -> None:
        """
        Set a transform on the data for this `DataSet`.

        Here you can set custom a custom transform to both decode data from the
        database and encode data written to the database. This allows you to have dates
        stored as timestamps in the database yet work with a human-readable format in
        the GUI and within PySimpleSQL. This transform happens only while PySimpleSQL
        actually reads from or writes to the database.

        :param fn: A callable function to preform encode/decode. This function should
        take three arguments: query, row (which will be populated by a dictionary of the
        row data), and an encode parameter (1 to encode, 0 to decode - see constants
        `TFORM_ENCODE` and `TFORM_DECODE`). Note that this transform works on one row at
        a time. See the example `journal_with_data_manipulation.py` for a usage example.
        :returns: None
        """
        self.transform = fn

    def set_query(self, query: str) -> None:
        """
        Set the query string for the `DataSet`.

        This is more for advanced users.  It defaults to "SELECT * FROM {table};" This
        can override the default

        :param query: The query string you would like to associate with the table
        :returns: None
        """
        logger.debug(f"Setting {self.table} query to {query}")
        self.query = query

    def set_join_clause(self, clause: str) -> None:
        """
        Set the `DataSet` object's join string.

        This is more for advanced users, as it will automatically generate from the
        database Relationships otherwise.

        :param clause: The join clause, such as "LEFT JOIN That on This.pk=That.fk"
        :returns: None
        """
        logger.debug(f"Setting {self.table} join clause to {clause}")
        self.join_clause = clause

    def set_where_clause(self, clause: str) -> None:
        """
        Set the `DataSet` object's where clause.

        This is ADDED TO the auto-generated where clause from Relationship data

        :param clause: The where clause, such as "WHERE pkThis=100"
        :returns: None
        """
        logger.debug(
            f"Setting {self.table} where clause to {clause} for DataSet {self.key}"
        )
        self.where_clause = clause

    def set_order_clause(self, clause: str) -> None:
        """
        Set the `DataSet` object's order clause.

        This is more for advanced users, as it will automatically generate from the
        database Relationships otherwise.

        :param clause: The order clause, such as "Order by name ASC"
        :returns: None
        """
        logger.debug(f"Setting {self.table} order clause to {clause}")
        self.order_clause = clause

    def update_column_info(self, column_info: ColumnInfo = None) -> None:
        """
        Generate column information for the `DataSet` object.

        This may need done, for example, when a manual query using joins is used. This
        is more for advanced users.

        :param column_info: (optional) A `ColumnInfo` instance. Defaults to being
            generated by the `SQLDriver`.
        :returns: None
        """
        # Now we need to set  new column names, as the query could have changed
        if column_info is not None:
            self.column_info = column_info
        else:
            self.column_info = self.driver.column_info(self.table)

    def set_description_column(self, column: str) -> None:
        """
        Set the `DataSet` object's description column.

        This is the column that will display in Listboxes, Comboboxes, Tables, etc.
        By default, this is initialized to either the 'description','name' or 'title'
        column, or the 2nd column of the table if none of those columns exist. This
        method allows you to specify a different column to use as the description for
        the record.

        :param column: The name of the column to use
        :returns: None
        """
        self.description_column = column

    def records_changed(self, column: str = None, recursive=True) -> bool:
        """
        Checks if records have been changed.

        This is done by comparing PySimpleGUI control values with the stored `DataSet`
        values.

        :param column: Limit the changed records search to just the supplied column name
        :param recursive: True to check related `DataSet` instances
        :returns: True or False on whether changed records were found
        """
        logger.debug(f'Checking if records have changed in table "{self.table}"...')

        # Virtual rows wills always be considered dirty
        if self.row_is_virtual():
            return True

        dirty = False
        # First check the current record to see if it's dirty
        for mapped in self.frm.element_map:
            # Compare the DB version to the GUI version
            if mapped.table == self.table:
                # if passed custom column name
                if column is not None and mapped.column != column:
                    continue

                # don't check if there aren't any rows. Fixes checkbox = '' when no
                # rows.
                if not len(self.frm[mapped.table].rows.index):
                    continue

                # Get the element value and cast it, so we can compare it to the
                # database version.
                element_val = self.column_info[mapped.column].cast(mapped.element.get())

                # Get the table value.  If this is a keyed element, we need figure out
                # the appropriate table column.
                table_val = None
                if mapped.where_column is not None:
                    for index, row in self.rows.itterrows():
                        if row[mapped.where_column] == mapped.where_value:
                            table_val = row[mapped.column]
                else:
                    table_val = self[mapped.column]

                if type(mapped.element) is sg.PySimpleGUI.Checkbox:
                    table_val = checkbox_to_bool(table_val)
                    element_val = checkbox_to_bool(element_val)

                # Sanitize things a bit due to empty values being slightly different in
                # the two cases.
                if table_val is None:
                    table_val = ""

                # Strip trailing whitespace from strings
                if type(table_val) is str:
                    table_val = table_val.rstrip()
                if type(element_val) is str:
                    element_val = element_val.rstrip()

                # Make the comparison
                # Temporary debug output
                # print(
                #    f"element: {element_val}({type(element_val)}),
                #    db: {table_val}({type(table_val)})"
                # )
                if element_val != table_val:
                    dirty = True
                    logger.debug("CHANGED RECORD FOUND!")
                    logger.debug(
                        f"\telement type: {type(element_val)} "
                        f"column_type: {type(table_val)}"
                    )
                    logger.debug(
                        f"\t{mapped.element.Key}:{element_val} != "
                        f"{mapped.column}:{table_val}"
                    )
                    return dirty
                dirty = False

        # handle recursive checking next
        if recursive:
            for rel in self.frm.relationships:
                if rel.parent_table == self.table and rel.on_update_cascade:
                    dirty = self.frm[rel.child_table].records_changed()
                    if dirty:
                        break
        return dirty

    def prompt_save(
        self, update_elements: bool = True
    ) -> Union[PROMPT_SAVE_PROCEED, PROMPT_SAVE_DISCARDED, PROMPT_SAVE_NONE]:
        """
        Prompts the user, asking if they want to save when changes are detected.

        This is called when the current record is about to change.

        :param update_elements: (optional) Passed to `Form.save_records()` ->
            `Form.save_records_recursive()` to update_elements. Additionally used to
            discard changes if user reply's 'No' to prompt.
        :returns: A prompt return value of one of the following: `PROMPT_PROCEED`,
            `PROMPT_DISCARDED`, or `PROMPT_NONE`.
        """
        # Return False if there is nothing to check or _prompt_save is False
        # TODO: children too?
        if (
            self.current_index is None
            or len(self.rows.index) == 0
            or not self._prompt_save
        ):
            return PROMPT_SAVE_NONE

        # See if any rows are virtual
        vrows = len(
            [row for idx, row in self.rows.iterrows() if self.row_is_virtual(idx)]
        )
        # Check if any records have changed
        changed = self.records_changed() or vrows
        if changed:
            if self._prompt_save == AUTOSAVE_MODE:
                save_changes = "yes"
            else:
                save_changes = self.frm.popup.yes_no(
                    lang.dataset_prompt_save_title, lang.dataset_prompt_save
                )
            if save_changes == "yes":
                # save this record's cascaded relationships, last to first
                if (
                    self.frm.save_records(
                        table=self.table, update_elements=update_elements
                    )
                    & SAVE_FAIL
                ):
                    return PROMPT_SAVE_DISCARDED
                return PROMPT_SAVE_PROCEED
            # if no
            self.purge_virtual()
            if vrows and update_elements:
                self.frm.update_elements(self.key)
            return PROMPT_SAVE_DISCARDED
        # if no changes
        return PROMPT_SAVE_NONE

    def requery(
        self,
        select_first: bool = True,
        filtered: bool = True,
        update_elements: bool = True,
        requery_dependents: bool = True,
    ) -> None:
        """
        Requeries the table.

        The `DataSet` object maintains an internal representation of
        the actual database table. The requery method will query the actual database and
        sync the `DataSet` object to it.

        :param select_first: (optional) If True, the first record will be selected after
            the requery.
        :param filtered: (optional) If True, the relationships will be considered and an
            appropriate WHERE clause will be generated. If False all records in the
            table will be fetched.
        :param update_elements: (optional) Passed to `DataSet.first()` to
            update_elements. Note that the select_first parameter must equal True to use
            this parameter.
        :param requery_dependents: (optional) passed to `DataSet.first()` to
            requery_dependents. Note that the select_first parameter must = True to use
            this parameter.
        :returns: None
        """
        join = ""
        where = ""

        if not self.filtered:
            filtered = False

        if filtered:
            # Stop requery short if parent has no records or current row is virtual
            parent_table = Relationship.get_parent(self.table)
            if parent_table and (
                not len(self.frm[parent_table].rows.index)
                or Relationship.parent_virtual(self.table, self.frm)
            ):
                self.rows = pd.DataFrame(columns=self.rows.columns)  # purge rows
                if update_elements:
                    self.frm.update_elements(self.key)
                if requery_dependents:
                    self.requery_dependents(update_elements=update_elements)
                return

            # else, get join/where clause like normal
            join = self.driver.generate_join_clause(self)
            where = self.driver.generate_where_clause(self)

        query = self.query + " " + join + " " + where + " " + self.order_clause
        # We want to store our sort settings before we wipe out the current DataFrame
        try:
            sort_settings = self.store_sort_settings()
        except (AttributeError, KeyError):
            sort_settings = [None, SORT_NONE]  # default for first query

        rows = self.driver.execute(query)
        self.rows = rows

        if len(self.rows.index):
            if "sort_order" not in self.rows.attrs:
                # Store the sort order as a dictionary in the attrs of the DataFrame
                sort_order = self.rows[self.pk_column].to_list()
                self.rows.attrs["sort_order"] = {self.pk_column: sort_order}
            # now we can restore the sort order
            self.load_sort_settings(sort_settings)
            self.sort(self.table)

        # Perform transform one row at a time
        if self.transform is not None:
            self.rows = self.rows.apply(
                lambda row: self.transform(self, row, TFORM_DECODE) or row, axis=1
            )

        # Strip trailing white space, as this is what sg[element].get() does, so we
        # can have an equal comparison. Not the prettiest solution.  Will look into
        # this more on the PySimpleGUI end and make a follow-up ticket.

        # TODO: Is the [:,:] still needed now that we are working with DateFrames?
        self.rows.loc[:, :] = self.rows.applymap(
            lambda x: x.rstrip() if isinstance(x, str) else x
        )

        if select_first:
            self.first(
                update_elements=update_elements,
                requery_dependents=requery_dependents,
                skip_prompt_save=True,  # already saved
            )

    def requery_dependents(
        self, child: bool = False, update_elements: bool = True
    ) -> None:
        """
        Requery parent `DataSet` instances as defined by the relationships of the table.

        :param child: (optional) If True, will requery self. Default False; used to skip
            requery when called by parent.
        :param update_elements: (optional) passed to `DataSet.requery()` ->
            `DataSet.first()` to update_elements.
        :returns: None
        """
        if child:
            # dependents=False: no recursive dependent requery
            self.requery(update_elements=update_elements, requery_dependents=False)

        for rel in self.frm.relationships:
            if rel.parent_table == self.table and rel.on_update_cascade:
                logger.debug(
                    f"Requerying dependent table {self.frm[rel.child_table].table}"
                )
                self.frm[rel.child_table].requery_dependents(
                    child=True, update_elements=update_elements
                )

    def first(
        self,
        update_elements: bool = True,
        requery_dependents: bool = True,
        skip_prompt_save: bool = False,
    ) -> None:
        """
        Move to the first record of the table.

        Only one entry in the table is ever considered "Selected"  This is one of
        several functions that influences which record is currently selected. See
        `DataSet.first()`, `DataSet.previous()`, `DataSet.next()`, `DataSet.last()`,
        `DataSet.search()`, `DataSet.set_by_pk()`, `DataSet.set_by_index()`.

        :param update_elements: (optional) Update the GUI elements after switching
            records.
        :param requery_dependents: (optional) Requery dependents after switching records
        :param skip_prompt_save: (optional) True to skip prompting to save dirty records
        :returns: None
        """
        logger.debug(f"Moving to the first record of table {self.table}")
        if skip_prompt_save is False:
            # don't update self/dependents if we are going to below anyway
            self.prompt_save(update_elements=False)

        self.current_index = 0
        if update_elements:
            self.frm.update_elements(self.key)
        if requery_dependents:
            self.requery_dependents(update_elements=update_elements)
        # callback
        if "record_changed" in self.callbacks:
            self.callbacks["record_changed"](self.frm, self.frm.window)

    def last(
        self,
        update_elements: bool = True,
        requery_dependents: bool = True,
        skip_prompt_save: bool = False,
    ):
        """
        Move to the last record of the table.

        Only one entry in the table is ever considered "Selected"  This is one of
        several functions that influences which record is currently selected. See
        `DataSet.first()`, `DataSet.previous()`, `DataSet.next()`, `DataSet.last()`,
        `DataSet.search()`, `DataSet.set_by_pk()`, `DataSet.set_by_index()`.

        :param update_elements: (optional) Update the GUI elements after switching
            records.
        :param requery_dependents: (optional) Requery dependents after switching records
        :param skip_prompt_save: (optional) True to skip prompting to save dirty records
        :returns: None
        """
        logger.debug(f"Moving to the last record of table {self.table}")
        if skip_prompt_save is False:
            # don't update self/dependents if we are going to below anyway
            self.prompt_save(update_elements=False)

        self.current_index = len(self.rows.index) - 1
        if update_elements:
            self.frm.update_elements(self.key)
        if requery_dependents:
            self.requery_dependents()
        # callback
        if "record_changed" in self.callbacks:
            self.callbacks["record_changed"](self.frm, self.frm.window)

    def next(
        self,
        update_elements: bool = True,
        requery_dependents: bool = True,
        skip_prompt_save: bool = False,
    ):
        """
        Move to the next record of the table.

        Only one entry in the table is ever considered "Selected"  This is one of
        several functions that influences which record is currently selected. See
        `DataSet.first()`, `DataSet.previous()`, `DataSet.next()`, `DataSet.last()`,
        `DataSet.search()`, `DataSet.set_by_pk()`, `DataSet.set_by_index()`.

        :param update_elements: (optional) Update the GUI elements after switching
            records.
        :param requery_dependents: (optional) Requery dependents after switching records
        :param skip_prompt_save: (optional) True to skip prompting to save dirty records
        :returns: None
        """
        if self.current_index < len(self.rows.index) - 1:
            logger.debug(f"Moving to the next record of table {self.table}")
            if skip_prompt_save is False:
                # don't update self/dependents if we are going to below anyway
                self.prompt_save(update_elements=False)

            self.current_index += 1
            if update_elements:
                self.frm.update_elements(self.key)
            if requery_dependents:
                self.requery_dependents()
            # callback
            if "record_changed" in self.callbacks:
                self.callbacks["record_changed"](self.frm, self.frm.window)

    def previous(
        self,
        update_elements: bool = True,
        requery_dependents: bool = True,
        skip_prompt_save: bool = False,
    ):
        """
        Move to the previous record of the table.

        Only one entry in the table is ever considered "Selected"  This is one of
        several functions that influences which record is currently selected. See
        `DataSet.first()`, `DataSet.previous()`, `DataSet.next()`, `DataSet.last()`,
        `DataSet.search()`, `DataSet.set_by_pk()`, `DataSet.set_by_index()`.

        :param update_elements: (optional) Update the GUI elements after switching
            records.
        :param requery_dependents: (optional) Requery dependents after switching records
        :param skip_prompt_save: (optional) True to skip prompting to save dirty records
        :returns: None
        """
        if self.current_index > 0:
            logger.debug(f"Moving to the previous record of table {self.table}")
            if skip_prompt_save is False:
                # don't update self/dependents if we are going to below anyway
                self.prompt_save(update_elements=False)

            self.current_index -= 1
            if update_elements:
                self.frm.update_elements(self.key)
            if requery_dependents:
                self.requery_dependents()
            # callback
            if "record_changed" in self.callbacks:
                self.callbacks["record_changed"](self.frm, self.frm.window)

    def search(
        self,
        search_string: str,
        update_elements: bool = True,
        requery_dependents: bool = True,
        skip_prompt_save: bool = False,
    ) -> Union[SEARCH_FAILED, SEARCH_RETURNED, SEARCH_ABORTED]:
        """
        Move to the next record in the `DataSet` that contains `search_string`.

        Successive calls will search from the current position, and wrap around back to
        the beginning. The search order from `DataSet.set_search_order()` will be used.
        If the search order is not set by the user, it will default to the description
        column (see `DataSet.set_description_column()`). Only one entry in the table is
        ever considered "Selected"  This is one of several functions that influences
        which record is currently selected. See `DataSet.first()`, `DataSet.previous()`,
        `DataSet.next()`, `DataSet.last()`, `DataSet.search()`, `DataSet.set_by_pk()`,
        `DataSet.set_by_index()`.

        :param search_string: The search string to look for
        :param update_elements: (optional) Update the GUI elements after switching
            records.
        :param requery_dependents: (optional) Requery dependents after switching records
        :param skip_prompt_save: (optional) True to skip prompting to save dirty records
        :returns: One of the following search values: `SEARCH_FAILED`,
            `SEARCH_RETURNED`, `SEARCH_ABORTED`.
        """
        # See if the string is an element name
        # TODO this is a bit of an ugly hack, but it works
        if search_string in self.frm.window.key_dict:
            search_string = self.frm.window[search_string].get()
        if not search_string:
            return SEARCH_ABORTED

        logger.debug(
            f'Searching for a record of table {self.table} "'
            f'with search string "{search_string}"'
        )
        # callback
        if "before_search" in self.callbacks and not self.callbacks["before_search"](
            self.frm, self.frm.window
        ):
            return SEARCH_ABORTED

        # TODO: Should this be before the before_search callback?
        if skip_prompt_save is False:
            # don't update self/dependents if we are going to below anyway
            self.prompt_save(update_elements=False)

        # First lets make a search order.. TODO: remove this hard coded garbage
        if len(self.rows.index):
            logger.debug(f"DEBUG: {self.search_order} {self.rows.columns[0]}")
        for field in self.search_order:
            # Perform a search for str, from the current position to the end and back by
            # creating a list of all indexes
            for i in list(range(self.current_index + 1, len(self.rows.index))) + list(
                range(0, self.current_index)
            ):
                if (
                    field in list(self.rows.columns)
                    and search_string.lower() in str(self.rows.iloc[i][field]).lower()
                ):
                    old_index = self.current_index
                    self.current_index = i
                    if update_elements:
                        self.frm.update_elements(self.key)
                    if requery_dependents:
                        self.requery_dependents()

                    # callback
                    if "after_search" in self.callbacks and not self.callbacks[
                        "after_search"
                    ](self.frm, self.frm.window):
                        self.current_index = old_index
                        self.frm.update_elements(self.key)
                        self.requery_dependents()
                        return SEARCH_ABORTED

                    # callback
                    if "record_changed" in self.callbacks:
                        self.callbacks["record_changed"](self.frm, self.frm.window)

                    return SEARCH_RETURNED
        return SEARCH_FAILED
        # If we have made it here, then it was not found!
        # sg.Popup('Search term "'+str+'" not found!')
        # TODO: Play sound?

    def set_by_index(
        self,
        index: int,
        update_elements: bool = True,
        requery_dependents: bool = True,
        skip_prompt_save: bool = False,
        omit_elements: List[str] = None,
    ) -> None:
        """
        Move to the record of the table located at the specified index in DataSet.

        Only one entry in the table is ever considered "Selected"  This is one of
        several functions that influences which record is currently selected. See
        `DataSet.first()`, `DataSet.previous()`, `DataSet.next()`, `DataSet.last()`,
        `DataSet.search()`, `DataSet.set_by_pk()`, `DataSet.set_by_index()`.

        :param index: The index of the record to move to.
        :param update_elements: (optional) Update the GUI elements after switching
            records.
        :param requery_dependents: (optional) Requery dependents after switching records
        :param skip_prompt_save: (optional) True to skip prompting to save dirty records
        :param omit_elements: (optional) A list of elements to omit from updating
        :returns: None
        """
        logger.debug(f"Moving to the record at index {index} on {self.table}")
        if omit_elements is None:
            omit_elements = []

        if skip_prompt_save is False:
            # see if sg.Table has potential changes
            if len(omit_elements) and self.records_changed(recursive=False):
                # most likely will need to update, either to
                # discard virtual or update after save
                omit_elements = []
            # don't update self/dependents if we are going to below anyway
            self.prompt_save(update_elements=False)

        self.current_index = index
        if update_elements:
            self.frm.update_elements(self.table, omit_elements=omit_elements)
        if requery_dependents:
            self.requery_dependents()

    def set_by_pk(
        self,
        pk: int,
        update_elements: bool = True,
        requery_dependents: bool = True,
        skip_prompt_save: bool = False,
        omit_elements: list[str] = None,
    ) -> None:
        """
        Move to the record with this primary key.

        This is useful when modifying a record (such as renaming).  The primary key can
        be stored, the record re-named, and then the current record selection updated
        regardless of the new sort order.
        Only one entry in the table is ever considered "Selected"  This is one of
        several functions that influences which record is currently selected. See
        `DataSet.first()`, `DataSet.previous()`, `DataSet.next()`, `DataSet.last()`,
        `DataSet.search()`, `DataSet.set_by_pk()`, `DataSet.set_by_index()`.

        :param pk: The record to move to containing the primary key
        :param update_elements: (optional) Update the GUI elements after switching
            records.
        :param requery_dependents: (optional) Requery dependents after switching records
        :param skip_prompt_save: (optional) True to skip prompting to save dirty records
        :param omit_elements: (optional) A list of elements to omit from updating
        :returns: None
        """
        logger.debug(f"Setting table {self.table} record by primary key {pk}")
        if omit_elements is None:
            omit_elements = []

        if skip_prompt_save is False:
            # see if sg.Table has potential changes
            if len(omit_elements) and self.records_changed(recursive=False):
                # most likely will need to update, either to
                # discard virtual or update after save
                omit_elements = []
            # don't update self/dependents if we are going to below anyway
            self.prompt_save(update_elements=False)

        # find current index of pk in resorted rows (not in-place)
        self.current_index = (
            self.rows.sort_index().index[self.rows[self.pk_column] == pk].tolist()[0]
        )
        if update_elements:
            self.frm.update_elements(self.table, omit_elements=omit_elements)
        if requery_dependents:
            self.requery_dependents()

    def get_current(
        self, column: str, default: Union[str, int] = ""
    ) -> Union[str, int]:
        """
        Get the current value for the supplied column.

        You can also use indexing of the `Form` object to get the current value of a
        column I.e. frm[{DataSet}].[{column}].

        :param column: The column you want to get the value from
        :param default: A value to return if the record is null
        :returns: The value of the column requested
        """
        logger.debug(f"Getting current record for {self.table}.{column}")
        if len(self.rows.index):
            if self.get_current_row()[column]:
                return self.get_current_row()[column]
            return default
        return default

    def set_current(self, column: str, value: Union[str, int]) -> None:
        """
        Set the current value for the supplied column.

        You can also use indexing of the `Form` object to set the current value of a
        column. I.e. frm[{DataSet}].[{column}] = 'New value'.

        :param column: The column you want to set the value for
        :param value: A value to set the current record's column to
        :returns: None
        """
        logger.debug(f"Setting current record for {self.table}.{column} = {value}")
        self.get_current_row()[column] = value

    def get_keyed_value(
        self, value_column: str, key_column: str, key_value: Union[str, int]
    ) -> Union[str, int, None]:
        """
        Return `value_column` where` key_column`=`key_value`.

        Useful for datastores with key/value pairs.

        :param value_column: The column to fetch the value from
        :param key_column: The column in which to search for the value
        :param key_value: The value to search for
        :returns: Returns the value found in `value_column`
        """
        for i, r in self.rows.itterrows():
            if r[key_column] == key_value:
                return r[value_column]
        return None

    def get_current_pk(self) -> int:
        """
        Get the primary key of the currently selected record.

        :returns: the primary key
        """
        return self.get_current(self.pk_column)

    def get_current_row(self) -> Union[pd.Series, None]:
        """
        Get the row for the currently selected record of this table.

        :returns: A pandas Series object
        """
        if not self.rows.empty:
            # force the current_index to be in bounds!
            # For child reparenting
            self.current_index = self.current_index

            return self.rows.iloc[self.current_index]
        return None

    def add_selector(
        self,
        element: sg.Element,
        data_key: str,
        where_column: str = None,
        where_value: str = None,
    ) -> None:
        """
        Use an element such as a listbox, combobox or a table as a selector item for
        this table.

        Note: This is not typically used by the end user, as this is called from the
        `selector()` convenience function.

        :param element: the PySimpleGUI element used as a selector element
        :param data_key: the `DataSet` item this selector will operate on
        :param where_column: (optional)
        :param where_value: (optional)
        :returns: None
        """
        if type(element) not in [
            sg.PySimpleGUI.Listbox,
            sg.PySimpleGUI.Slider,
            sg.Combo,
            sg.Table,
        ]:
            raise RuntimeError(
                f"add_selector() error: {element} is not a supported element."
            )

        logger.debug(f"Adding {element.Key} as a selector for the {self.table} table.")
        d = {
            "element": element,
            "data_key": data_key,
            "where_column": where_column,
            "where_value": where_value,
        }
        self.selector.append(d)

    def insert_record(
        self, values: Dict[str : Union[str, int]] = None, skip_prompt_save: bool = False
    ) -> None:
        """
        Insert a new record virtually in the `DataSet` object.

        If values are passed, it will initially set those columns to the values (I.e.
        {'name': 'New Record', 'note': ''}), otherwise they will be fetched from the
        database if present.

        :param values: column:value pairs
        :param skip_prompt_save: Skip prompting the user to save dirty records before
            the insert.
        :returns: None
        """
        # todo: you don't add a record if there isn't a parent!!!
        # todo: this is currently filtered out by enabling of the element, but it should
        #  be filtered here too!
        # todo: bring back the values parameter?
        if skip_prompt_save is False:
            # don't update self/dependents if we are going to below anyway
            self.prompt_save(update_elements=False)

        # Don't insert if parent has no records or is virtual
        parent_table = Relationship.get_parent(self.table)
        if (
            parent_table
            and not len(self.frm[parent_table].rows)
            or Relationship.parent_virtual(self.table, self.frm)
        ):
            logger.debug(f"{parent_table=} is empty or current row is virtual")
            return

        # Get a new dict for a new row with default values already filled in
        new_values = self.column_info.default_row_dict(self)

        # If the values parameter was passed in, overwrite any values in the dict
        if values is not None:
            for k, v in values.items():
                if k in new_values:
                    new_values[k] = v

        # Make sure we take into account the foreign key relationships...
        for r in self.frm.relationships:
            if self.table == r.child_table and r.on_update_cascade:
                new_values[r.fk_column] = self.frm[r.parent_table].get_current_pk()

        # Update the pk to match the expected pk the driver would generate on insert.
        new_values[self.pk_column] = self.driver.next_pk(self.table, self.pk_column)

        # Insert the new values using DataSet.insert_row(),
        # marking the new row as virtual
        self.insert_row(new_values)

        # and move to the new record
        self.set_by_pk(
            new_values[self.pk_column],
            update_elements=True,
            requery_dependents=True,
            skip_prompt_save=True,  # already saved
        )

    def save_record(
        self, display_message: bool = None, update_elements: bool = True
    ) -> int:
        """
        Save the currently selected record.

        Saves any changes made via the GUI back to the database.  The
        before_save and after_save `DataSet.callbacks` will call your
        own functions for error checking if needed!.

        :param display_message: Displays a message "Updates saved successfully",
            otherwise is silent on success.
        :param update_elements: Update the GUI elements after saving
        :returns: SAVE_NONE, SAVE_FAIL or SAVE_SUCCESS masked with SHOW_MESSAGE
        """
        logger.debug(f"Saving records for table {self.table}...")

        if display_message is None:
            display_message = not self.save_quiet

        # Ensure that there is actually something to save
        if not len(self.rows):
            self.frm.popup.info(
                lang.dataset_save_empty, display_message=display_message
            )
            return SAVE_NONE + SHOW_MESSAGE

        # callback
        if "before_save" in self.callbacks and self.callbacks["before_save"]() is False:
            logger.debug("We are not saving!")
            if update_elements:
                self.frm.update_elements(self.key)
            if display_message:
                self.frm.popup.ok(
                    lang.dataset_save_callback_false_title,
                    lang.dataset_save_callback_false,
                )
            return SAVE_FAIL + SHOW_MESSAGE

        # Check right away to see if any records have changed, no need to proceed any
        # further than we have to.
        if not self.records_changed(recursive=False) and self.frm.force_save is False:
            self.frm.popup.info(lang.dataset_save_none, display_message=display_message)
            return SAVE_NONE + SHOW_MESSAGE

        # Work with a copy of the original row and transform it if needed
        # Note that while saving, we are working with just the current row of data,
        # unless it's 'keyed' via ?/=
        current_row = self.get_current_row().copy()

        # Track the keyed queries we have to run.  Set to None, so we can tell later if
        # there were keyed elements
        keyed_queries: Optional[
            List
        ] = None  # {'column':column, 'changed_row': row, 'where_clause': where_clause}

        # Propagate GUI data back to the stored current_row
        for mapped in [m for m in self.frm.element_map if m.dataset == self]:
            # convert the data into the correct type using the domain in ColumnInfo
            element_val = self.column_info[mapped.column].cast(mapped.element.get())

            # Looked for keyed elements first
            if mapped.where_column is not None:
                if keyed_queries is None:
                    # Make the list here so != None if keyed elements
                    keyed_queries = []
                for row in self.rows:
                    if (
                        row[mapped.where_column] == mapped.where_value
                        and row[mapped.column] != element_val
                    ):
                        # This record has changed.  We will save it
                        row[mapped.column] = element_val  # propagate the value
                        changed = {mapped.column: element_val}
                        where_col = self.driver.quote_column(mapped.where_column)
                        where_val = self.driver.quote_value(mapped.where_value)
                        where_clause = f"WHERE {where_col} = {where_val}"
                        keyed_queries.append(
                            {
                                "column": mapped.column,
                                "changed_row": changed,
                                "where_clause": where_clause,
                            }
                        )
            else:
                current_row[mapped.column] = element_val

        changed_row = dict(current_row.items())
        cascade_fk_changed = False
        # check to see if cascading-fk has changed before we update database
        cascade_fk_column = Relationship.get_update_cascade_fk_column(self.table)
        if cascade_fk_column:
            # check if fk
            for mapped in self.frm.element_map:
                if mapped.dataset == self and mapped.column == cascade_fk_column:
                    cascade_fk_changed = self.records_changed(
                        column=cascade_fk_column, recursive=False
                    )

        # Update the database from the stored rows
        if self.transform is not None:
            self.transform(self, changed_row, TFORM_ENCODE)

        # Save or Insert the record as needed
        if keyed_queries is not None:
            # Now execute all the saved queries from earlier
            for q in keyed_queries:
                # Update the database from the stored rows
                if self.transform is not None:
                    self.transform(self, q["changed_row"], TFORM_ENCODE)
                result = self.driver.save_record(
                    self, q["changed_row"], q["where_clause"]
                )
                if result.exception is not None:
                    self.frm.popup.ok(
                        lang.dataset_save_keyed_fail_title,
                        lang.dataset_save_keyed_fail.format_map(
                            LangFormat(exception=result.exception)
                        ),
                    )
                    self.driver.rollback()
                    return SAVE_FAIL  # Do not show the message in this case
        else:
            if self.row_is_virtual():
                result = self.driver.insert_record(
                    self.table, self.get_current_pk(), self.pk_column, changed_row
                )
            else:
                result = self.driver.save_record(self, changed_row)

            if result.attrs["exception"] is not None:
                self.frm.popup.ok(
                    lang.dataset_save_fail_title,
                    lang.dataset_save_fail.format_map(
                        LangFormat(exception=result.exception)
                    ),
                )
                self.driver.rollback()
                return SAVE_FAIL  # Do not show the message in this case

            # Store the pk, so we can move to it later - use the value returned in the
            # attrs if possible. The expected pk may have changed from autoincrement
            # and/or concurrent access.
            pk = (
                result.attrs["lastrowid"]
                if result.attrs["lastrowid"] is not None
                else self.get_current_pk()
            )
            current_row[self.pk_column] = pk

            # then update the current row data
            self.rows.iloc[self.current_index] = current_row

            # If child changes parent, move index back and requery/requery_dependents
            if (
                cascade_fk_changed and not self.row_is_virtual()
            ):  # Virtual rows already requery, and have no dependents.
                self.frm[self.table].requery(select_first=False)  # keep spot in table
                self.frm[self.table].requery_dependents()

            # Lets refresh our data
            if self.row_is_virtual():
                # Requery so that the new row honors the order clause
                self.requery(select_first=False, update_elements=False)
                if update_elements:
                    # Then move to the record
                    self.set_by_pk(
                        pk,
                        skip_prompt_save=True,
                        requery_dependents=False,
                    )
                    # only need to reset the Insert button
                    self.frm.update_elements(edit_protect_only=True)

        # callback
        if "after_save" in self.callbacks and not self.callbacks["after_save"](
            self.frm, self.frm.window
        ):
            self.driver.rollback()
            return SAVE_FAIL + SHOW_MESSAGE

        # If we made it here, we can commit the changes, since the save and insert above
        # do not commit or rollback
        self.driver.commit()

        # Sort so the saved row honors the current order.
        self.sort(self.table)

        if update_elements:
            self.frm.update_elements(self.key)
        logger.debug("Record Saved!")
        self.frm.popup.info(lang.dataset_save_success, display_message=display_message)

        return SAVE_SUCCESS + SHOW_MESSAGE

    def save_record_recursive(
        self,
        results: SaveResultsDict,
        display_message=False,
        check_prompt_save: bool = False,
        update_elements: bool = True,
    ) -> SaveResultsDict:
        """
        Recursively save changes, taking into account the relationships of the tables.

        :param results: Used in Form.save_records to collect DataSet.save_record
            returns. Pass an empty dict to get list of {table : result}
        :param display_message: Passed to DataSet.save_record. Displays a message
            that updates were saved successfully, otherwise is silent on success.
        :param check_prompt_save: Used when called from Form.prompt_save. Updates
            elements without saving if individual `DataSet._prompt_save()` is False.
        :returns: dict of {table : results}
        """
        for rel in self.frm.relationships:
            if rel.parent_table == self.table and rel.on_update_cascade:
                self.frm[rel.child_table].save_record_recursive(
                    results=results,
                    display_message=display_message,
                    check_prompt_save=check_prompt_save,
                    update_elements=update_elements,
                )
        # if dataset-level doesn't allow prompt_save
        if check_prompt_save and self._prompt_save is False:
            if update_elements:
                self.frm.update_elements(self.key)
            results[self.table] = PROMPT_SAVE_NONE
            return results
        # otherwise, proceed
        result = self.save_record(
            display_message=display_message, update_elements=update_elements
        )
        results[self.table] = result
        return results

    def delete_record(
        self, cascade: bool = True
    ):  # TODO: check return type, we return True below
        """
        Delete the currently selected record.

        The before_delete and after_delete callbacks are run during this process
        to give some control over the process.

        :param cascade: Delete child records (as defined by `Relationship`s that were
            set up) before deleting this record.
        :returns: None
        """
        # Ensure that there is actually something to delete
        if not len(self.rows):
            return None

        # callback
        if "before_delete" in self.callbacks and not self.callbacks["before_delete"](
            self.frm, self.frm.window
        ):
            return None

        children = []
        if cascade:
            children = Relationship.get_delete_cascade_tables(self.table)

        msg_children = ", ".join(children)
        if len(children):
            msg = lang.delete_cascade.format_map(LangFormat(children=msg_children))
        else:
            msg = lang.delete_single
        answer = self.frm.popup.yes_no(lang.delete_title, msg)
        if answer == "no":
            return True

        if self.row_is_virtual():
            self.rows.purge_virtual()
            self.frm.update_elements(self.key)
            # only need to reset the Insert button
            self.frm.update_elements(edit_protect_only=True)
            return None

        # Delete child records first!
        result = self.driver.delete_record(self, True)

        if not isinstance(result, pd.DataFrame):
            if result == DELETE_RECURSION_LIMIT_ERROR:
                self.frm.popup.ok(
                    lang.delete_failed_title,
                    lang.delete_failed.format_map(
                        LangFormat(exception=lang.delete_recursion_limit_error)
                    ),
                )
            elif result.exception is not None:
                self.frm.popup.ok(
                    lang.delete_failed_title,
                    lang.delete_failed.format_map(
                        LangFormat(exception=result.exception)
                    ),
                )

        # callback
        if "after_delete" in self.callbacks:
            if not self.callbacks["after_delete"](self.frm, self.frm.window):
                self.driver.rollback()
            else:
                self.driver.commit()
        else:
            self.driver.commit()

        self.requery(select_first=False)
        self.frm.update_elements(self.key)
        self.requery_dependents()
        return None

    def duplicate_record(
        self, children: bool = None
    ) -> Union[bool, None]:  # TODO check return type, returns True within
        """
        Duplicate the currently selected record.

        The before_duplicate and after_duplicate callbacks are run during this
        process to give some control over the process.

        :param children: Duplicate child records (as defined by `Relationship`s that
            were set up) before duplicating this record.
        :returns: None
        """
        # Ensure that there is actually something to duplicate
        if not len(self.rows.index) or self.row_is_virtual():
            return None

        # callback
        if "before_duplicate" in self.callbacks and not self.callbacks[
            "before_duplicate"
        ](self.frm, self.frm.window):
            return None

        if children is None:
            children = self.duplicate_children

        child_list = []
        if children:
            child_list = Relationship.get_update_cascade_tables(self.table)

        msg_children = ", ".join(child_list)
        msg = lang.duplicate_child.format_map(
            LangFormat(children=msg_children)
        ).splitlines()
        layout = [[sg.T(line, font="bold")] for line in msg]
        if len(child_list):
            answer = sg.Window(
                lang.duplicate_child_title,
                [
                    layout,
                    [
                        sg.Button(
                            button_text=lang.duplicate_child_button_dupparent,
                            key="parent",
                            use_ttk_buttons=themepack.use_ttk_buttons,
                            pad=themepack.popup_button_pad,
                        )
                    ],
                    [
                        sg.Button(
                            button_text=lang.duplicate_child_button_dupboth,
                            key="cascade",
                            use_ttk_buttons=themepack.use_ttk_buttons,
                            pad=themepack.popup_button_pad,
                        )
                    ],
                    [
                        sg.Button(
                            button_text=lang.button_cancel,
                            key="cancel",
                            use_ttk_buttons=themepack.use_ttk_buttons,
                            pad=themepack.popup_button_pad,
                        )
                    ],
                ],
                keep_on_top=True,
                modal=True,
                ttk_theme=themepack.ttk_theme,
            ).read(close=True)
            if answer[0] == "parent":
                children = False
            elif answer[0] in ["cancel", None]:
                return True
        else:
            msg = lang.duplicate_single
            answer = self.frm.popup.yes_no(lang.duplicate_single_title, msg)
            if answer == "no":
                return True
        # Store our current pk, so we can move to it if the duplication fails
        pk = self.get_current_pk()

        # Have the driver duplicate the record
        result = self.driver.duplicate_record(self, children)
        if result.attrs["exception"]:
            self.driver.rollback()
            self.frm.popup.ok(
                lang.duplicate_failed_title,
                lang.duplicate_failed.format_map(
                    LangFormat(exception=result.exception)
                ),
            )
        else:
            pk = result.attrs["lastrowid"]

        # callback
        if "after_duplicate" in self.callbacks:
            if not self.callbacks["after_duplicate"](self.frm, self.frm.window):
                self.driver.rollback()
            else:
                self.driver.commit()
        else:
            self.driver.commit()
        self.driver.commit()

        # requery and move to new pk
        self.requery(select_first=False)
        self.set_by_pk(pk, skip_prompt_save=True)
        return None

    def get_description_for_pk(self, pk: int) -> Union[str, int, None]:
        """
        Get the description from the `DataSet` on the matching pk.

        Return the desctription from `DataSet.description_column` for the row where the
        `DataSet.pk_column` = `pk`.

        :param pk: The primary key from which to find the description for
        :returns: The value found in the description column, or None if nothing is found
        """
        for index, row in self.rows.iterrows():
            if row[self.pk_column] == pk:
                return row[self.description_column]
        return None

    def row_is_virtual(self, index: int = None) -> bool:
        """
        Check whether the row at `index` is virtual

        :param index: The index to check. If none is passed, then the current index will
            be used.
        :returns: True or False based on whether the row is virtual
        """
        if index is None:
            index = self.current_index
        if self.rows is not None and len(self.rows.index):
            return self.rows.attrs["virtual"][index]
        return False

    def table_values(
        self, columns: List[str] = None, mark_virtual: bool = False
    ) -> List[TableRow]:
        """
        Create a values list of `TableRows`s for use in a PySimpleGUI Table element.

        :param columns: A list of column names to create table values for.
            Defaults to getting them from the `DataSet.rows` DataFrame.
        :param mark_virtual: Place a marker next to virtual records
        :returns: A list of `TableRow`s suitable for using with PySimpleGUI Table
            element values.
        """

        values = []
        try:
            all_columns = list(self.rows.columns)
        except IndexError:
            all_columns = []

        columns = all_columns if columns is None else columns

        pk_column = self.column_info.pk_column()

        for index, row in self.rows.iterrows():
            if mark_virtual:
                lst = (
                    [themepack.marker_virtual] if self.row_is_virtual(index) else [" "]
                )
            else:
                lst = []

            rels = Relationship.get_relationships(self.table)
            pk = None
            for col in all_columns:
                # Is this the primary key column?
                if col == pk_column:
                    pk = row[col]
                # Skip this column if we aren't supposed to grab it
                if col not in columns:
                    continue
                # Get this column info, including fk descriptions
                found = False
                for rel in rels:
                    if col == rel.fk_column:
                        lst.append(
                            self.frm[rel.parent_table].get_description_for_pk(row[col])
                        )
                        found = True
                        break
                if not found:
                    lst.append(row[col])
            values.append(TableRow(pk, lst))

        return values

    def get_related_table_for_column(self, column: str) -> str:
        """
        Get parent table name as it relates to this column.

        :param column: The column name to get related table information for
        :returns: The name of the related table, or the current table if none are found
        """
        rels = Relationship.get_relationships(self.table)
        for rel in rels:
            if column == rel.fk_column:
                return rel.parent_table
        return self.table  # None could be found, return our own table instead

    def quick_editor(
        self,
        pk_update_funct: callable = None,
        funct_param: any = None,
        skip_prompt_save: bool = False,
    ) -> None:
        """
        The quick editor is a dynamic PySimpleGUI Window for quick editing of tables.
        This is very useful for putting a button next to a combobox or listbox so that
        the available values can be added/edited/deleted easily.
        Note: This is not typically used by the end user, as it can be configured from
        the `field()` convenience function.

        :param pk_update_funct: (optional) A function to call to determine the pk to
            select by default when the quick editor loads.
        :param funct_param: (optional) A parameter to pass to the `pk_update_funct`
        :param skip_prompt_save: (Optional) True to skip prompting to save dirty records
        :returns: None
        """

        if skip_prompt_save is False:
            self.frm.prompt_save()
        # Reset the keygen to keep consistent naming
        logger.info("Creating Quick Editor window")
        keygen.reset()
        data_key = self.key
        layout = []
        headings = self.column_info.names()
        visible = [1] * len(headings)
        visible[0] = 0
        col_width = int(55 / (len(headings) - 1))
        for i in range(0, len(headings)):
            headings[i] = headings[i].ljust(col_width, " ")

        layout.append(
            [
                selector(
                    data_key,
                    sg.Table,
                    key=f"{data_key}:quick_editor",
                    num_rows=10,
                    headings=headings,
                    visible_column_map=visible,
                )
            ]
        )
        layout.append([actions(data_key, edit_protect=False)])
        layout.append([sg.Text("")])
        layout.append([sg.HorizontalSeparator()])
        for col in self.column_info.names():
            column = f"{data_key}.{col}"
            if col != self.pk_column:
                layout.append([field(column)])

        quick_win = sg.Window(
            lang.quick_edit_title.format_map(LangFormat(data_key=data_key)),
            layout,
            keep_on_top=True,
            modal=True,
            finalize=True,
            ttk_theme=themepack.ttk_theme,  # Must, otherwise will redraw window
        )
        quick_frm = Form(self.frm.driver, bind_window=quick_win)

        # Select the current entry to start with
        if pk_update_funct is not None:
            if funct_param is None:
                quick_frm[data_key].set_by_pk(pk_update_funct())
            else:
                quick_frm[data_key].set_by_pk(pk_update_funct(funct_param))

        while True:
            event, values = quick_win.read()

            if quick_frm.process_events(event, values):
                logger.debug(
                    f"PySimpleSQL Quick Editor event handler handled the event {event}!"
                )
            if event in [sg.WIN_CLOSED, "Exit"]:
                break

            logger.debug(f"This event ({event}) is not yet handled.")
        quick_win.close()
        self.requery()
        self.frm.update_elements()

    def add_simple_transform(self, transforms: SimpleTransformsDict) -> None:
        """
        Merge a dictionary of transforms into the `DataSet._simple_transform`
        dictionary.

        Example:
        -------
        {'entry_date' : {
            'decode' : lambda row,col: datetime.utcfromtimestamp(int(row[col])).strftime('%m/%d/%y'), # fmt: skip
            'encode' : lambda row,col: datetime.strptime(row[col], '%m/%d/%y').replace(tzinfo=timezone.utc).timestamp(), # fmt: skip
        }}
        :param transforms: A dict of dicts containing either 'encode' or 'decode' along
            with a callable to do the transform. See example above
        :returns: None
        """  # noqa: E501
        for k, v in transforms.items():
            if not callable(v):
                RuntimeError(f"Transform for {k} must be callable!")
            self._simple_transform[k] = v

    def purge_virtual(self) -> None:
        """
        Purge virtual rows from the DataFrame.

        :returns: None
        """
        # remove the rows where virtual is True in place, along with the corresponding
        # virtual attribute
        idx_to_remove = [idx for idx, v in self.rows.attrs["virtual"].items() if v]
        self.rows.drop(index=idx_to_remove, inplace=True)
        for idx in idx_to_remove:
            del self.rows.attrs["virtual"][idx]

    def sort_by_column(self, column: str, table: str, reverse=False) -> None:
        """
        Sort the DataFrame by column. Using the mapped relationships of the database,
        foreign keys will automatically sort based on the parent table's description
        column, rather than the foreign key number.

        :param column: The name of the column to sort the DataFrame by
        :param table: The name of the table the column belongs to
        :param reverse: Reverse the sort; False = ASC, True = DESC
        :returns: None
        """
        # Target sorting by this DataFrame

        # We don't want to sort by foreign keys directly - we want to sort by the
        # description column of the foreign table that the foreign key references
        tmp_column = None
        rels = Relationship.get_relationships(table)
        for rel in rels:
            if column == rel.fk_column:
                # Create a mapping dictionary from the parent DataFrame
                df_parent = self.frm[rel.parent_table].rows
                desc_parent = self.frm[rel.parent_table].description_column
                mapping = dict(zip(df_parent[rel.pk_column], df_parent[desc_parent]))

                # Create a temporary column in self.rows for the fk data
                tmp_column = f"temp_{rel.parent_table}.{rel.pk_column}"
                self.rows[tmp_column] = self.rows[rel.fk_column].map(mapping)
                column = tmp_column
                break
        try:
            self.rows.sort_values(
                column,
                ascending=not reverse,
                inplace=True,
            )
        except KeyError:
            logger.debug(f"DataFrame could not sort by column {column}. KeyError.")
        finally:
            # Drop the temporary description column (if it exists)
            if tmp_column is not None:
                self.rows.drop(columns=tmp_column, inplace=True, errors="ignore")

    def sort_by_index(self, index: int, table: str, reverse=False):
        """
        Sort the self.rows DataFrame by column index Using the mapped relationships of
        the database, foreign keys will automatically sort based on the parent table's
        description column, rather than the foreign key number.

        :param index: The index of the column to sort the DateFrame by
        :param table: The name of the table the column belongs to
        :param reverse: Reverse the sort; False = ASC, True = DESC
        :returns: None
        """
        column = self.rows.columns[index]
        self.sort_by_column(column, table, reverse)

    def store_sort_settings(self) -> list:
        """
        Store the current sort settingg. Sort settings are just the sort column and
        reverse setting. Sort order can be restored with
        `DataSet.load_sort_settings()`.

        :returns: A list containing the sort_column and the sort_reverse
        """
        return [self.rows.attrs["sort_column"], self.rows.attrs["sort_reverse"]]

    def load_sort_settings(self, sort_settings: list) -> None:
        """
        Load a previously stored sort setting. Sort settings are just the sort columm
        and reverse setting.

        :param sort_settings: A list as returned by `DataSet.store_sort_settings()`
        """
        self.rows.attrs["sort_column"] = sort_settings[0]
        self.rows.attrs["sort_reverse"] = sort_settings[1]

    def sort_reset(self) -> None:
        """
        Reset the sort order to the original order as defined by the DataFram index

        :returns: None
        """
        # Restore the original sort order
        self.rows.sort_index(inplace=True)

    def sort(self, table: str, update_elements: bool = True, sort_order=None) -> None:
        """
        Sort according to the internal sort_column and sort_reverse variables. This is a
        good way to re-sort without changing the sort_cycle.

        :param table: The table associated with this DataSet.  Passed along to
            `DataSet.sort_by_column()`
        :param update_elements: Update associated selectors and navigation buttons, and
            table header sort marker.
        :param sort_order: Passed to `Dataset.update_headings`. A SORT_* constant
            (SORT_NONE, SORT_ASC, SORT_DESC). Note that the update_elements parameter
            must = True to use this parameter.
        :returns: None
        """
        pk = self.get_current_pk()
        if self.rows.attrs["sort_column"] is None:
            logger.debug("Sort column is None.  Resetting sort.")
            self.sort_reset()
        else:
            logger.debug(f"Sorting by column {self.rows.attrs['sort_column']}")
            self.sort_by_column(
                self.rows.attrs["sort_column"], table, self.rows.attrs["sort_reverse"]
            )
        self.set_by_pk(
            pk,
            update_elements=False,
            requery_dependents=False,
            skip_prompt_save=True,
        )
        if update_elements and len(self.rows.index):
            self.frm.update_selectors(self.table)
            self.frm.update_elements(self.table, edit_protect_only=True)
            self.update_headings(self.rows.attrs["sort_column"], sort_order)

    def sort_cycle(self, column: str, table: str, update_elements: bool = True) -> int:
        """
        Cycle between original sort order of the DataFrame, ASC by column, and DESC by
        column with each call.

        :param column: The column name to cycle the sort on
        :param table: The table that the column belongs to
        :param update_elements: Passed to `Dataset.sort` to update update associated
            selectors and navigation buttons, and table header sort marker.
        :returns: A sort constant; SORT_NONE, SORT_ASC, or SORT_DESC
        """
        if column != self.rows.attrs["sort_column"]:
            self.rows.attrs["sort_column"] = column
            self.rows.attrs["sort_reverse"] = False
            self.sort(table, update_elements=update_elements, sort_order=SORT_ASC)
            return SORT_ASC
        if not self.rows.attrs["sort_reverse"]:
            self.rows.attrs["sort_reverse"] = True
            self.sort(table, update_elements=update_elements, sort_order=SORT_DESC)
            return SORT_DESC
        self.rows.attrs["sort_reverse"] = False
        self.rows.attrs["sort_column"] = None
        self.sort(table, update_elements=update_elements, sort_order=SORT_NONE)
        return SORT_NONE

    def update_headings(self, column, sort_order):
        for e in self.selector:
            element = e["element"]
            if element.metadata["TableHeading"]:
                element.metadata["TableHeading"].update_headings(
                    element, column, sort_order
                )

    def insert_row(self, row: dict, idx: int = None) -> None:
        """
        Insert a new virtual row into the DataFrame. Virtual rows are ones that exist
        in memory, but not in the database. When a save action is performed, virtual
        rows will be added into the database.

        :param row: A dict representation of a row of data
        :param idx: The index where the row should be inserted (default to last index)
        :returns: None
        """
        row_series = pd.Series(row)
        attrs = self.rows.attrs.copy()
        self.rows = pd.concat([self.rows, row_series.to_frame().T], ignore_index=True)
        self.rows.attrs = attrs

        # I don't have the idx parameter working yet
        if idx is None:
            idx = len(self.rows.index)
        idx_label = self.rows.index.max() if len(self.rows.index) > 0 else 0
        self.rows.attrs["virtual"].loc[idx_label] = True


class Form:

    """
    `Form` class.

    Maintains an internal version of the actual database
    `DataSet` objects can be accessed by key, I.e. frm['data_key'].
    """

    instances = []  # Track our instances
    relationships = []  # Track our relationships

    def __init__(
        self,
        driver: SQLDriver,
        bind_window: sg.Window = None,
        prefix_data_keys: str = "",
        parent: Form = None,
        filter: str = None,
        select_first: bool = True,
        prompt_save: int = PROMPT_MODE,
        save_quiet: bool = False,
        update_cascade: bool = True,
        delete_cascade: bool = True,
        duplicate_children: bool = True,
        description_column_names: List[str] = None,
    ) -> None:
        """
        Initialize a new `Form` instance.

        :param driver: Supported `SQLDriver`. See `Sqlite()`, `Mysql()`, `Postgres()`
        :param bind_window: Bind this window to the `Form`
        :param prefix_data_keys: (optional) prefix auto generated data_key names with
            this value. Example 'data_'
        :param parent: (optional)Parent `Form` to base dataset off of
        :param filter: (optional) Only import elements with the same filter set.
            Typically set with `field()`, but can also be set manually as a dict with
            the key 'filter' set in the element's metadata
        :param select_first: (optional) Default:True. For each top-level parent, selects
            first row, populating children as well.
        :param prompt_save: (optional) Default:PROMPT_MODE. Prompt to save changes when
            dirty records are present.
            Two modes available, (if pysimplesql is imported as `ss`) use:
            - `ss.PROMPT_MODE` to prompt to save when unsaved changes are present.
            - `ss.AUTOSAVE_MODE` to automatically save when unsaved changes are present.
        :param save_quiet: (optional) Default:False. True to skip info popup on save.
            Error popups will still be shown.
        :param update_cascade: (optional) Default:True. Requery and filter child table
            on selected parent primary key. (ON UPDATE CASCADE in SQL)
        :param delete_cascade: (optional) Default:True. Delete the dependent child
            records if the parent table record is deleted. (ON UPDATE DELETE in SQL)
        :param duplicate_children: (optional) Default:True. If record has children,
            prompt user to choose to duplicate current record, or both.
        :param description_column_names: (optional) A list of names to use for the
            DataSet object's description column, displayed in Listboxes, Comboboxes, and
            Tables instead of the primary key. The first matching column of the table is
            given priority. If no match is found, the second column is used. Default
            list: ['description', 'name', 'title'].
        :returns: None
        """
        win_pb = ProgressBar(lang.startup_form)
        win_pb.update(lang.startup_init, 0)
        Form.instances.append(self)

        self.driver: SQLDriver = driver
        self.filter: str = filter
        self.parent: Form = parent  # TODO: This doesn't seem to really be used yet
        self.window: Optional[sg.Window] = None
        self._edit_protect: bool = False
        self.datasets: Dict[str, DataSet] = {}
        self.element_map: List[ElementMap] = []
        self.popup = None
        """
        The element map dict is set up as below:

        .. literalinclude:: ../doc_examples/element_map.1.py
        :language: python
        :caption: Example code
        """
        self.event_map = []  # Array of dicts, {'event':, 'function':, 'table':}
        self.relationships: List[Relationship] = []
        self.callbacks: CallbacksDict = {}
        self._prompt_save: int = prompt_save
        self.save_quiet: bool = save_quiet
        self.force_save: bool = False
        self.update_cascade: bool = update_cascade
        self.delete_cascade: bool = delete_cascade
        self.duplicate_children: int = duplicate_children
        if description_column_names is None:
            self.description_column_names = ["description", "name", "title"]
        else:
            self.description_column_names = description_column_names

        # Add our default datasets and relationships
        win_pb.update(lang.startup_datasets, 25)
        self.auto_add_datasets(prefix_data_keys)
        win_pb.update(lang.startup_relationships, 50)
        self.auto_add_relationships()
        self.requery_all(
            select_first=select_first, update_elements=False, requery_dependents=True
        )
        if bind_window is not None:
            win_pb.update(lang.startup_binding, 75)
            self.window = bind_window
            self.popup = Popup()
            self.bind(self.window)
        win_pb.close()

    def __del__(self):
        self.close()

    # Override the [] operator to retrieve dataset by key
    def __getitem__(self, key: str) -> DataSet:
        try:
            return self.datasets[key]
        except KeyError:
            raise RuntimeError(
                f"The DataSet for `{key}` does not exist. This can be caused because "
                f"the database does not exist, the database user does not have the "
                f"proper permissions set, or any number of db configuration issues."
            )

    def close(self, reset_keygen: bool = True):
        """
        Safely close out the `Form`.

        :param reset_keygen: True to reset the keygen for this `Form`
        """
        # First delete the dataset associated
        DataSet.purge_form(self, reset_keygen)
        self.driver.close()

    def bind(self, win: sg.Window) -> None:
        """
        Bind the PySimpleGUI Window to the Form for the purpose of GUI element, event
        and relationship mapping. This can happen automatically on `Form` creation with
        the bind parameter and is not typically called by the end user. This function
        literally just groups all the auto_* methods.  See `Form.auto_add_tables()`,
        `Form.auto_add_relationships()`, `Form.auto_map_elements()`,
        `Form.auto_map_events()`.

        :param win: The PySimpleGUI window
        :returns:  None
        """
        logger.info("Binding Window to Form")
        self.window = win
        self.popup = Popup()
        self.auto_map_elements(win)
        self.auto_map_events(win)
        self.update_elements()
        logger.debug("Binding finished!")

    def execute(self, query: str) -> pd.DataFrame:
        """
        Convenience function to pass along to `SQLDriver.execute()`.

        :param query: The query to execute
        :returns: A pandas DataFrame object with attrs set for lastrowid and exception
        """
        return self.driver.execute(query)

    def commit(self) -> None:
        """
        Convenience function to pass along to `SQLDriver.commit()`.

        :returns: None
        """
        self.driver.commit()

    def set_callback(
        self, callback_name: str, fctn: Callable[[Form, sg.Window], Union[None, bool]]
    ) -> None:
        """
        Set `Form` callbacks. A runtime error will be raised if the callback is not
        supported. The following callbacks are supported: update_elements Called after
        elements are updated via `Form.update_elements()`. This allows for other GUI
        manipulation on each update of the GUI edit_enable Called before editing mode is
        enabled. This can be useful for asking for a password for example edit_disable
        Called after the editing mode is disabled.

            {element_name} Called while updating MAPPED element.  This overrides the
            default element update implementation. Note that the {element_name} callback
            function needs to return a value to pass to Win[element].update()

        :param callback_name: The name of the callback, from the list above
        :param fctn: The function to call.  Note, the function must take in two
            parameters, a Form instance, and a PySimpleGUI.Window instance
        :returns: None
        """
        logger.info(f"Callback {callback_name} being set on Form")
        supported = ["update_elements", "edit_enable", "edit_disable"]

        # Add in mapped elements
        for mapped in self.element_map:
            supported.append(mapped.element.key)

        # Add in other window elements
        for element in self.window.key_dict:
            supported.append(element)

        if callback_name in supported:
            self.callbacks[callback_name] = fctn
        else:
            raise RuntimeError(
                f'Callback "{callback_name}" not supported. callback: {callback_name} '
                f"supported: {supported}"
            )

    def add_dataset(
        self,
        data_key: str,
        table: str,
        pk_column: str,
        description_column: str,
        query: str = "",
        order_clause: str = "",
    ) -> None:
        """
        Manually add a `DataSet` object to the `Form` When you attach to a database,
        PySimpleSQL isn't aware of what it contains until this command is run Note that
        `Form.auto_add_datasets()` does this automatically, which is called when a
        `Form` is created.

        :param data_key: The key to give this `DataSet`.  Use frm['data_key'] to access
            it.
        :param table: The name of the table in the database
        :param pk_column: The primary key column of the table in the database
        :param description_column: The column to be used to display to users in
            listboxes, comboboxes, etc.
        :param query: The initial query for the table.  Auto generates "SELECT * FROM
            {table}" if none is passed
        :param order_clause: The initial sort order for the query
        :returns: None
        """
        self.datasets.update(
            {
                data_key: DataSet(
                    data_key,
                    self,
                    table,
                    pk_column,
                    description_column,
                    query,
                    order_clause,
                )
            }
        )
        # set a default sort order
        self[data_key].set_search_order([description_column])

    def add_relationship(
        self,
        join: str,
        child_table: str,
        fk_column: str,
        parent_table: str,
        pk_column: str,
        update_cascade: bool,
        delete_cascade: bool,
    ) -> None:
        """
        Add a foreign key relationship between two dataset of the database When you
        attach a database, PySimpleSQL isn't aware of the relationships contained until
        dataset are added via `Form.add_data`, and the relationship of various tables is
        set with this function. Note that `Form.auto_add_relationships()` will do this
        automatically from the schema of the database, which also happens automatically
        when a `Form` is created.

        :param join: The join type of the relationship ('LEFT JOIN', 'INNER JOIN',
            'RIGHT JOIN')
        :param child_table: The child table containing the foreign key
        :param fk_column: The foreign key column of the child table
        :param parent_table: The parent table containing the primary key
        :param pk_column: The primary key column of the parent table
        :param update_cascade: Requery and filter child table results on selected parent
            primary key (ON UPDATE CASCADE in SQL)
        :param delete_cascade: Delete the dependent child records if the parent table
            record is deleted (ON UPDATE DELETE in SQL)
        :returns: None
        """
        self.relationships.append(
            Relationship(
                join,
                child_table,
                fk_column,
                parent_table,
                pk_column,
                update_cascade,
                delete_cascade,
                self.driver,
                self,
            )
        )

    def set_fk_column_cascade(
        self,
        child_table: str,
        fk_column: str,
        update_cascade: bool = None,
        delete_cascade: bool = None,
    ) -> None:
        """
        Set a foreign key's update_cascade and delete_cascade behavior.

        `Form.auto_add_relationships()` does this automatically from the database
        schema.

        :param child_table: Child table with the foreign key.
        :param fk_column: Foreign key column of the child table.
        :param update_cascade: True to requery and filter child table on selected parent
            primary key.
        :param delete_cascade: True to delete dependent child records if parent record
            is deleted.
        :returns: None
        """
        for rel in self.relationships:
            if rel.child_table == child_table and rel.fk_column == fk_column:
                logger.info(f"Updating {fk_column=} relationship.")
                if update_cascade is not None:
                    rel.update_cascade = update_cascade
                if delete_cascade is not None:
                    rel.delete_cascade = delete_cascade

    def auto_add_datasets(self, prefix_data_keys: str = "") -> None:
        """
        Automatically add `DataSet` objects from the database by looping through the
        tables available and creating a `DataSet` object for each. Each dataset key is
        an optional prefix plus the name of the table. When you attach to a sqlite
        database, PySimpleSQL isn't aware of what it contains until this command is run.
        This is called automatically when a `Form ` is created. Note that
        `Form.add_table()` can do this manually on a per-table basis.

        :param prefix_data_keys: Adds a prefix to the auto-generated `DataSet` keys
        :returns: None
        """
        logger.info(
            "Automatically generating dataset for each table in the sqlite database"
        )
        # Clear any current dataset so successive calls won't double the entries
        self.datasets = {}
        tables = self.driver.get_tables()
        for table in tables:
            column_info = self.driver.column_info(table)

            # auto generate description column.  Default it to the 2nd column,
            # but can be overwritten below
            description_column = column_info.col_name(1)
            for col in column_info.names():
                if col in self.description_column_names:
                    description_column = col
                    break

            # Get our pk column
            pk_column = self.driver.pk_column(table)

            data_key = prefix_data_keys + table
            logger.debug(
                f'Adding DataSet "{data_key}" on table {table} to Form with primary '
                f"key {pk_column} and description of {description_column}"
            )
            self.add_dataset(data_key, table, pk_column, description_column)
            self.datasets[data_key].column_info = column_info

    # Make sure to send a list of table names to requery if you want
    # dependent dataset to requery automatically
    def auto_add_relationships(self) -> None:
        """
        Automatically add a foreign key relationship between tables of the database.
        This is done by foreign key constraints within the database.  Automatically
        requery the child table if the parent table changes (ON UPDATE CASCADE in sql is
        set) When you attach a database, PySimpleSQL isn't aware of the relationships
        contained until tables are added and the relationship of various tables is set.
        This happens automatically during `Form` creation. Note that
        `Form.add_relationship()` can do this manually.

        :returns: None
        """
        logger.info("Automatically adding foreign key relationships")
        # Clear any current rels so that successive calls will not double the entries
        self.relationships = []  # clear any relationships already stored
        relationships = self.driver.relationships()
        for r in relationships:
            logger.debug(
                f'Adding relationship {r["from_table"]}.{r["from_column"]} = '
                f'{r["to_table"]}.{r["to_column"]}'
            )
            self.add_relationship(
                "LEFT JOIN",
                r["from_table"],
                r["from_column"],
                r["to_table"],
                r["to_column"],
                r["update_cascade"],
                r["delete_cascade"],
            )

    # Map an element to a DataSet.
    # Optionally a where_column and a where_value.  This is useful for key,value pairs!
    def map_element(
        self,
        element: sg.Element,
        dataset: DataSet,
        column: str,
        where_column: str = None,
        where_value: str = None,
    ) -> None:
        """
        Map a PySimpleGUI element to a specific `DataSet` column.  This is what makes
        the GUI automatically update to the contents of the database.  This happens
        automatically when a PySimpleGUI Window is bound to a `Form` by using the bind
        parameter of `Form` creation, or by executing `Form.auto_map_elements()` as long
        as the element metadata is configured properly. This method can be used to
        manually map any element to any `DataSet` column regardless of metadata
        configuration.

        :param element: A PySimpleGUI Element
        :param dataset: A `DataSet` object
        :param column: The name of the column to bind to the element
        :param where_column: Used for ke, value shorthand TODO: expand on this
        :param where_value: Used for ey, value shorthand TODO: expand on this
        :returns: None
        """
        logger.debug(f"Mapping element {element.key}")
        self.element_map.append(
            ElementMap(element, dataset, column, where_column, where_value)
        )

    def auto_map_elements(self, win: sg.Window, keys: List[str] = None) -> None:
        """
        Automatically map PySimpleGUI Elements to `DataSet` columns. A special naming
        convention has to be used for automatic mapping to happen.  Note that
        `Form.map_element()` can be used to manually map an Element to a column.
        Automatic mapping relies on a special naming convention as well as certain data
        in the Element's metadata. The convenience functions `field()`, `selector()`,
        and `actions()` do this automatically and should be used in almost all cases to
        make elements that conform to this standard, but this information will allow you
        to do this manually if needed. For individual fields, Element keys must be named
        'Table.column'. Additionally, the metadata must contain a dict with the key of
        'type' set to `TYPE_RECORD`. For selectors, the key can be named whatever you
        want, but the metadata must contain a dict with the key of 'type' set to
        TPE_SELECTOR.

        :param win: A PySimpleGUI Window
        :param keys: (optional) Limit the auto mapping to this list of Element keys
        :returns: None
        """
        logger.info("Automapping elements")
        # Clear previously mapped elements so successive calls won't produce duplicates
        self.element_map = []
        for key in win.key_dict:
            element = win[key]

            # Skip this element if there is no metadata present
            if type(element.metadata) is not dict:
                continue

            # Process the filter to ensure this element should be mapped to this Form
            if element.metadata["filter"] == self.filter:
                element.metadata["Form"] = self

            # Skip this element if it's an event
            if element.metadata["type"] == TYPE_EVENT:
                continue

            if element.metadata["Form"] != self:
                continue
            # If we passed in a custom list of elements
            if keys is not None and key not in keys:
                continue

            # Map Record Element
            if element.metadata["type"] == TYPE_RECORD:
                # Does this record imply a where clause (indicated by ?)
                # If so, we can strip out the information we need
                data_key = element.metadata["data_key"]
                field = element.metadata["field"]
                if "?" in field:
                    table_info, where_info = field.split("?")
                else:
                    table_info = field
                    where_info = None
                try:
                    table, col = table_info.split(".")
                except ValueError:
                    table, col = table_info, None

                if where_info is None:
                    where_column = where_value = None
                else:
                    where_column, where_value = where_info.split("=")

                # make sure we don't use reserved keywords that could end up in a query
                for keyword in [table, col, where_column, where_value]:
                    if keyword is not None and keyword:
                        self.driver.check_keyword(keyword)

                # DataSet objects are named after the tables they represent
                # (with an optional prefix)
                # TODO: How to handle the prefix?
                # TODO: check in DataSet.table
                if table in self.datasets and col in self[table].column_info:
                    # Map this element to DataSet.column
                    self.map_element(
                        element, self[table], col, where_column, where_value
                    )

            # Map Selector Element
            elif element.metadata["type"] == TYPE_SELECTOR:
                k = element.metadata["table"]
                if k is None:
                    continue
                if element.metadata["Form"] != self:
                    continue
                if "?" in k:
                    table_info, where_info = k.split("?")
                    where_column, where_value = where_info.split("=")
                else:
                    table_info = k
                    where_column = where_value = None
                data_key = table_info

                if data_key in self.datasets:
                    self[data_key].add_selector(
                        element, data_key, where_column, where_value
                    )

                    # Enable sorting if TableHeading  is present
                    if type(element) is sg.Table and "TableHeading" in element.metadata:
                        table_heading: TableHeadings = element.metadata["TableHeading"]
                        # We need a whole chain of things to happen
                        # when a heading is clicked on:
                        # 1 Run the ResultRow.sort_cycle() with the correct column name
                        # 2 Run TableHeading.update_headings() with the:
                        #   Table element, sort_column, sort_reverse
                        # 3 Run update_elements() to see the changes
                        table_heading.enable_sorting(
                            element,
                            _SortCallbackWrapper(self, data_key),
                        )

                else:
                    logger.debug(f"Can not add selector {str(element)}")

    def set_element_clauses(
        self, element: sg.Element, where_clause: str = None, order_clause: str = None
    ) -> None:
        """
        Set the where and/or order clauses for the specified element in the element map.

        :param element: A PySimpleGUI Element
        :param where_clause: (optional) The where clause to set
        :param order_clause: (optional) The order clause to set
        :returns: None
        """
        for mapped in self.element_map:
            if mapped.element == element:
                mapped.where_clause = where_clause
                mapped.order_clause = order_clause

    def map_event(
        self, event: str, fctn: Callable[[None], None], table: str = None
    ) -> None:
        """
        Manually map a PySimpleGUI event (returned by Window.read()) to a callable. The
        callable will execute when the event is detected by `Form.process_events()`.
        Most users will not have to manually map any events, as `Form.auto_map_events()`
        will create most needed events when a PySimpleGUI Window is bound to a `Form` by
        using the bind parameter of `Form` creation, or by executing
        `Form.auto_map_elements()`.

        :param event: The event to watch for, as returned by PySimpleGUI Window.read()
            (an element name for example)
        :param fctn: The callable to run when the event is detected. It should take no
            parameters and have no return value
        :param table: (optional) currently not used
        :returns: None
        """
        dic = {"event": event, "function": fctn, "table": table}
        logger.debug(f"Mapping event {event} to function {fctn}")
        self.event_map.append(dic)

    def replace_event(
        self, event: str, fctn: Callable[[None], None], table: str = None
    ) -> None:
        """
        Replace an event that was manually mapped with `Form.auto_map_events()` or
        `Form.map_event()`. The callable will execute.

        :param event: The event to watch for, as returned by PySimpleGUI Window.read()
            (an element name for example)
        :param fctn: The callable to run when the event is detected. It should take no
            parameters and have no return value
        :param table: (optional) currently not used
        :returns: None
        """
        for e in self.event_map:
            if e["event"] == event:
                e["function"] = fctn
                e["table"] = table if table is not None else e["table"]

    def auto_map_events(self, win: sg.Window) -> None:
        """
        Automatically map events. pysimplesql relies on certain events to function
        properly. This method maps all the record navigation (previous, next, etc.) and
        database actions (insert, delete, save, etc.).  Note that the event mapper is
        very general-purpose, and you can add your own event triggers to the mapper
        using `Form.map_event()`, or even replace one of the auto-generated ones if you
        have specific needs by using `Form.replace_event()`.

        :param win: A PySimpleGUI Window
        :returns: None
        """
        logger.info("Automapping events")
        # Clear mapped events to ensure successive calls won't produce duplicates
        self.event_map = []

        for key in win.key_dict:
            # key = str(key)  # sometimes end up with an integer element 0?TODO:Research
            element = win[key]
            # Skip this element if there is no metadata present
            if type(element.metadata) is not dict:
                logger.debug(f"Skipping mapping of {key}")
                continue
            if element.metadata["Form"] != self:
                continue
            if element.metadata["type"] == TYPE_EVENT:
                event_type = element.metadata["event_type"]
                table = element.metadata["table"]
                column = element.metadata["column"]
                function = element.metadata["function"]
                funct = None

                data_key = table
                data_key = data_key if data_key in self.datasets else None
                if event_type == EVENT_FIRST:
                    if data_key:
                        funct = self[data_key].first
                elif event_type == EVENT_PREVIOUS:
                    if data_key:
                        funct = self[data_key].previous
                elif event_type == EVENT_NEXT:
                    if data_key:
                        funct = self[data_key].next
                elif event_type == EVENT_LAST:
                    if data_key:
                        funct = self[data_key].last
                elif event_type == EVENT_SAVE:
                    if data_key:
                        funct = self[data_key].save_record
                elif event_type == EVENT_INSERT:
                    if data_key:
                        funct = self[data_key].insert_record
                elif event_type == EVENT_DELETE:
                    if data_key:
                        funct = self[data_key].delete_record
                elif event_type == EVENT_DUPLICATE:
                    if data_key:
                        funct = self[data_key].duplicate_record
                elif event_type == EVENT_EDIT_PROTECT_DB:
                    self.edit_protect()  # Enable it!
                    funct = self.edit_protect
                elif event_type == EVENT_SAVE_DB:
                    funct = self.save_records
                elif event_type == EVENT_SEARCH:
                    # Build the search box name
                    search_element, command = key.split(":")
                    search_box = f"{search_element}:search_input"
                    if data_key:
                        funct = functools.partial(self[data_key].search, search_box)
                # elif event_type==EVENT_SEARCH_DB:
                elif event_type == EVENT_QUICK_EDIT:
                    referring_table = table
                    table = self[table].get_related_table_for_column(column)
                    funct = functools.partial(
                        self[table].quick_editor,
                        self[referring_table].get_current,
                        column,
                    )
                elif event_type == EVENT_FUNCTION:
                    funct = function
                else:
                    logger.debug(f"Unsupported event_type: {event_type}")

                if funct is not None:
                    self.map_event(key, funct, data_key)

    def edit_protect(self) -> None:
        """
        The edit protect system allows records to be protected from accidental editing
        by disabling the insert, delete, duplicate and save buttons on the GUI.  A
        button to toggle the edit protect mode can easily be added by using the
        `actions()` convenience function.

        :returns: None
        """
        logger.debug("Toggling edit protect mode.")
        # Callbacks
        if (
            self._edit_protect
            and "edit_enable" in self.callbacks
            and not self.callbacks["edit_enable"](self, self.window)
        ):
            return
        if "edit_disable" in self.callbacks and not self.callbacks["edit_disable"](
            self, self.window
        ):
            return

        self._edit_protect = not self._edit_protect
        self.update_elements(edit_protect_only=True)

    def get_edit_protect(self) -> bool:
        """
        Get the current edit protect state.

        :returns: True if edit protect is enabled, False if not enabled
        """
        return self._edit_protect

    def prompt_save(self) -> PromptSaveValue:
        """
        Prompt to save if any GUI changes are found the affect any table on this form.
        The helps prevent data entry loss when performing an action that changes the
        current record of a `DataSet`.

        :returns: One of the prompt constant values: PROMPT_SAVE_PROCEED,
            PROMPT_SAVE_DISCARDED, PROMPT_SAVE_NONE
        """
        user_prompted = False  # Has the user been prompted yet?
        for data_key in self.datasets:
            if not self[data_key]._prompt_save:
                continue

            if self[data_key].records_changed(recursive=False) and not user_prompted:
                # only show popup once, regardless of how many dataset have changed
                user_prompted = True
                if self._prompt_save == AUTOSAVE_MODE:
                    save_changes = "yes"
                else:
                    save_changes = self.popup.yes_no(
                        lang.form_prompt_save_title, lang.form_prompt_save
                    )
                if save_changes != "yes":
                    # update the elements to erase any GUI changes,
                    # since we are choosing not to save
                    for data_key_ in self.datasets:
                        self[data_key_].rows.purge_virtual()
                    self.update_elements()
                    # We did have a change, regardless if the user chose not to save
                    return PROMPT_SAVE_DISCARDED
                break
        if user_prompted:
            self.save_records(check_prompt_save=True)
        return PROMPT_SAVE_PROCEED if user_prompted else PROMPT_SAVE_NONE

    def set_force_save(self, force: bool = False) -> None:
        """
        Force save without checking for changes first, so even an unchanged record will
        be written back to the database.

        :param force: True to force unchanged records to save.
        :returns: None
        """
        self.force_save = force

    def save_records(
        self,
        table: str = None,
        cascade_only: bool = False,
        check_prompt_save: bool = False,
        update_elements: bool = True,
    ) -> Union[SAVE_SUCCESS, SAVE_FAIL, SAVE_NONE]:
        """
        Save records of all `DataSet` objects` associated with this `Form`.

        :param table: Name of table to save, as well as any cascaded relationships.
            Used in `DataSet.prompt_save()`
        :param cascade_only: Save only tables with cascaded relationships. Default
            False.
        :param check_prompt_save: Passed to `DataSet.save_record_recursive` to check if
            individual `DataSet` has prompt_save enabled. Used when
            `DataSet.save_records()` is called from `Form.prompt_save()`.
        :param update_elements: (optional) Passed to `Form.save_record_recursive()`
        :returns: result - can be used with RETURN BITMASKS
        """
        if check_prompt_save:
            logger.debug("Saving records in all datasets that allow prompt_save...")
        else:
            logger.debug("Saving records in all datasets...")

        display_message = not self.save_quiet

        result = 0
        show_message = True
        failed_tables = []

        if table:
            tables = [table]  # if passed single table
        # for cascade_only, build list of top-level dataset that have children
        elif cascade_only:
            tables = [
                dataset.table
                for dataset in self.datasets.values()
                if len(Relationship.get_update_cascade_tables(dataset.table))
                and Relationship.get_parent(dataset.table) is None
            ]
        # default behavior, build list of top-level dataset (ones without a parent)
        else:
            tables = [
                dataset.table
                for dataset in self.datasets.values()
                if Relationship.get_parent(dataset.table) is None
            ]

        # call save_record_recursive on tables, which saves from last to first.
        result_list = []
        for q in tables:
            res = self[q].save_record_recursive(
                results={},
                display_message=False,
                check_prompt_save=check_prompt_save,
                update_elements=update_elements,
            )
            result_list.append(res)

        # flatten list of result dicts
        results = {k: v for d in result_list for k, v in d.items()}
        logger.debug(f"Form.save_records - results of tables - {results}")

        # get tables that failed
        for t, res in results.items():
            if not res & SHOW_MESSAGE:
                show_message = (
                    False  # Only one instance of not showing the message hides all
                )
            if res & SAVE_FAIL:
                failed_tables.append(t)
            result |= res

        # Build a descriptive message, since the save spans many tables potentially
        msg = ""
        msg_tables = ", ".join(failed_tables)
        if result & SAVE_FAIL:
            if result & SAVE_SUCCESS:
                msg = lang.form_save_partial
            msg += lang.form_save_problem.format_map(LangFormat(tables=msg_tables))
            if show_message:
                self.popup.ok(lang.form_save_problem_title, msg)
            return result
        msg = lang.form_save_success if result & SAVE_SUCCESS else lang.form_save_none
        if show_message:
            self.popup.info(msg, display_message=display_message)
        return result

    def set_prompt_save(self, mode: int) -> None:
        """
        Set the prompt to save action when navigating records for all `DataSet` objects
        associated with this `Form`.

        :param mode: a constant value. If pysimplesql is imported as `ss`, use:
                    `ss.PROMPT_MODE` to prompt to save when unsaved changes are present.
                    `ss.AUTOSAVE_MODE` to autosave when unsaved changes are present.
        :returns: None
        """
        self._prompt_save = mode
        for data_key in self.datasets:
            self[data_key].set_prompt_save(mode)

    def update_elements(
        self,
        target_data_key: str = None,
        edit_protect_only: bool = False,
        omit_elements: List[str] = None,
    ) -> None:
        """
        Updated the GUI elements to reflect values from the database for this `Form`
        instance only. Not to be confused with the main `update_elements()`, which
        updates GUI elements for all `Form` instances. This method also executes
        `update_selectors()`, which updates selector elements.

        :param target_data_key: (optional) dataset key to update elements for, otherwise
            updates elements for all datasets
        :param edit_protect_only: (optional) If true, only update items affected by
            edit_protect
        :param omit_elements: A list of elements to omit updating
        :returns: None
        """
        if omit_elements is None:
            omit_elements = []

        msg = "edit protect" if edit_protect_only else "PySimpleGUI"
        logger.debug(f"update_elements(): Updating {msg} elements")
        win = self.window
        # Disable/Enable action elements based on edit_protect or other situations

        for data_key in self.datasets:
            if target_data_key is not None and data_key != target_data_key:
                continue

            # disable mapped elements for this table if
            # there are no records in this table or edit protect mode
            disable = len(self[data_key].rows.index) == 0 or self._edit_protect
            self.update_element_states(data_key, disable)

            for m in (m for m in self.event_map if m["table"] == self[data_key].table):
                # Disable delete and mapped elements for this table if there are no
                # records in this table or edit protect mode
                if ":table_delete" in m["event"]:
                    disable = len(self[data_key].rows.index) == 0 or self._edit_protect
                    win[m["event"]].update(disabled=disable)

                # Disable duplicate if no rows, edit protect, or current row virtual
                elif ":table_duplicate" in m["event"]:
                    disable = (
                        len(self[data_key].rows.index) == 0
                        or self._edit_protect
                        or self[data_key]
                        .get_current_row()
                        .attrs.get("virtual", False)
                        .iloc[0]
                    )
                    win[m["event"]].update(disabled=disable)

                # Disable first/prev if only 1 row, or first row
                elif ":table_first" in m["event"] or ":table_previous" in m["event"]:
                    disable = (
                        len(self[data_key].rows.index) < 2
                        or self[data_key].current_index == 0
                    )
                    win[m["event"]].update(disabled=disable)

                # Disable next/last if only 1 row, or last row
                elif ":table_next" in m["event"] or ":table_last" in m["event"]:
                    disable = len(self[data_key].rows.index) < 2 or (
                        self[data_key].current_index
                        == len(self[data_key].rows.index) - 1
                    )
                    win[m["event"]].update(disabled=disable)

                # Disable insert on children with no parent/virtual parent records or
                # edit protect mode
                elif ":table_insert" in m["event"]:
                    parent = Relationship.get_parent(data_key)
                    if parent is not None:
                        disable = bool(
                            len(self[parent].rows.index) == 0
                            or self._edit_protect
                            or Relationship.parent_virtual(data_key, self)
                        )
                    else:
                        disable = self._edit_protect
                    win[m["event"]].update(disabled=disable)

                # Disable db_save when needed
                elif ":db_save" in m["event"] or ":save_table" in m["event"]:
                    disable = len(self[data_key].rows.index) == 0 or self._edit_protect
                    win[m["event"]].update(disabled=disable)

                # Enable/Disable quick edit buttons
                elif ":quick_edit" in m["event"]:
                    win[m["event"]].update(disabled=disable)
        if edit_protect_only:
            return

        # Render GUI Elements
        # d= dictionary (the element map dictionary)
        for mapped in self.element_map:
            # If the optional target_data_key parameter was passed, we will only update
            # elements bound to that table
            if (
                target_data_key is not None
                and mapped.table != self[target_data_key].table
            ):
                continue

            # skip updating this element if requested
            if mapped.element in omit_elements:
                continue

            if type(mapped.element) is not sg.Text:  # don't show markers for sg.Text
                # Show the Required Record marker if the column has notnull set and
                # this is a virtual row
                marker_key = mapped.element.key + ":marker"
                try:
                    if mapped.dataset.row_is_virtual():
                        # get the column name from the key
                        col = mapped.column
                        # get notnull from the column info
                        if (
                            col in mapped.dataset.column_info.names()
                            and mapped.dataset.column_info[col].notnull
                        ):
                            self.window[marker_key].update(
                                visible=True,
                                text_color=themepack.marker_required_color,
                            )
                    else:
                        self.window[marker_key].update(visible=False)
                        if self.window is not None:
                            self.window[marker_key].update(visible=False)
                except AttributeError:
                    self.window[marker_key].update(visible=False)

            updated_val = None
            # If there is a callback for this element, use it
            if mapped.element.key in self.callbacks:
                self.callbacks[mapped.element.key]()

            elif mapped.where_column is not None:
                # We are looking for a key,value pair or similar.
                # Sift through and see what to put
                updated_val = mapped.dataset.get_keyed_value(
                    mapped.column, mapped.where_column, mapped.where_value
                )
                if type(mapped.element) in [
                    sg.PySimpleGUI.CBox
                ]:  # TODO, may need to add more??
                    updated_val = checkbox_to_bool(updated_val)

            elif type(mapped.element) is sg.PySimpleGUI.Combo:
                # Update elements with foreign dataset first
                # This will basically only be things like comboboxes
                # TODO: move this to only compute if something else changes?
                # Find the relationship to determine which table to get data from
                target_table = None
                # TODO this should be get_relationships_for_data?
                rels = Relationship.get_relationships(mapped.dataset.table)
                for rel in rels:
                    if rel.fk_column == mapped.column:
                        target_table = self[rel.parent_table]
                        pk_column = target_table.pk_column
                        description = target_table.description_column
                        break

                if target_table is None:
                    logger.info(
                        f"Error! Could not find related data for element "
                        f"{mapped.element.key} bound to DataSet "
                        f"key {mapped.table}, column: {mapped.column}"
                    )

                    # we don't want to update the list in this case, as it was most
                    # likely supplied and not tied to data
                    updated_val = mapped.dataset[mapped.column]

                # Populate the combobox entries
                else:
                    lst = []
                    for index, row in target_table.rows.iterrows():
                        lst.append(ElementRow(row[pk_column], row[description]))

                    # Map the value to the combobox, by getting the description_column
                    # and using it to set the value
                    for index, row in target_table.rows.iterrows():
                        if row[target_table.pk_column] == mapped.dataset[rel.fk_column]:
                            for entry in lst:
                                if entry.get_pk() == mapped.dataset[rel.fk_column]:
                                    updated_val = entry
                                    break
                            break
                    mapped.element.update(values=lst)
            elif type(mapped.element) is sg.PySimpleGUI.Table:
                # Tables use an array of arrays for values.  Note that the headings
                # can't be changed.
                values = mapped.dataset.table_values()
                # Select the current one
                pk = mapped.dataset.get_current_pk()

                if len(values):
                    # set index to pk
                    index = [[v[0] for v in values].index(pk)]
                    # calculate pk percentage position
                    pk_position = index[0] / len(values)
                else:  # if empty
                    index = []
                    pk_position = 0

                # Update table, and set vertical scroll bar to follow selected element
                update_table_element(
                    self.window, mapped.element, values, index, pk_position
                )
                continue

            elif type(mapped.element) in [
                sg.PySimpleGUI.InputText,
                sg.PySimpleGUI.Multiline,
                sg.PySimpleGUI.Text,
            ]:
                # Update the element in the GUI
                # For text objects, lets clear it first...

                # HACK for sqlite query not making needed keys! This will clear
                mapped.element.update("")

                updated_val = mapped.dataset[mapped.column]

            elif type(mapped.element) is sg.PySimpleGUI.Checkbox:
                updated_val = checkbox_to_bool(mapped.dataset[mapped.column])
            elif type(mapped.element) is sg.PySimpleGUI.Image:
                val = mapped.dataset[mapped.column]

                try:
                    val = eval(val)
                except:  # noqa: E722
                    # treat it as a filename
                    mapped.element.update(val)
                else:
                    # update the bytes data
                    mapped.element.update(data=val)
                # Prevent the update from triggering below, since we are doing it here
                updated_val = None
            else:
                sg.popup(f"Unknown element type {type(mapped.element)}")

            # Finally, we will update the actual GUI element!
            if updated_val is not None:
                mapped.element.update(updated_val)

        self.update_selectors(target_data_key, omit_elements)

        # Run callbacks
        if "update_elements" in self.callbacks:
            # Running user update function
            logger.info("Running the update_elements callback...")
            self.callbacks["update_elements"](self, self.window)

    def update_selectors(
        self, target_data_key: str = None, omit_elements: List[str] = None
    ) -> None:
        """
        Updated the GUI elements to reflect values from the database for this `Form`
        instance only. Not to be confused with the main `update_elements()`, which
        updates GUI elements for all `Form` instances.

        :param target_data_key: (optional) dataset key to update elements for, otherwise
            updates elements for all datasets
        :param omit_elements: A list of elements to omit updating
        :returns: None
        """
        if omit_elements is None:
            omit_elements = []

        # ---------
        # SELECTORS
        # ---------
        # We can update the selector elements
        # We do it down here because it's not a mapped element...
        # Check for selector events
        for data_key, dataset in self.datasets.items():
            if target_data_key is not None and target_data_key != data_key:
                continue

            if len(dataset.selector):
                for e in dataset.selector:
                    logger.debug("update_elements: SELECTOR FOUND")
                    # skip updating this element if requested
                    if e["element"] in omit_elements:
                        continue

                    element: sg.Element = e["element"]
                    logger.debug(f"{type(element)}")
                    pk_column = dataset.pk_column
                    description_column = dataset.description_column
                    if element.key in self.callbacks:
                        self.callbacks[element.key]()

                    if (
                        type(element) == sg.PySimpleGUI.Listbox
                        or type(element) == sg.PySimpleGUI.Combo
                    ):
                        logger.debug("update_elements: List/Combo selector found...")
                        lst = []
                        for r in dataset.rows:
                            if e["where_column"] is not None:
                                if str(r[e["where_column"]]) == str(
                                    e["where_value"]
                                ):  # TODO: Kind of a hackish way to check for equality.
                                    lst.append(
                                        ElementRow(r[pk_column], r[description_column])
                                    )
                                else:
                                    pass
                            else:
                                lst.append(
                                    ElementRow(r[pk_column], r[description_column])
                                )

                        element.update(values=lst, set_to_index=dataset.current_index)

                        # set vertical scroll bar to follow selected element
                        # (for listboxes only)
                        if type(element) == sg.PySimpleGUI.Listbox:
                            try:
                                element.set_vscroll_position(
                                    dataset.current_index / len(lst)
                                )
                            except ZeroDivisionError:
                                element.set_vscroll_position(0)

                    elif type(element) == sg.PySimpleGUI.Slider:
                        # Re-range the element depending on the number of records
                        l = len(dataset.rows)  # noqa: E741
                        element.update(value=dataset._current_index + 1, range=(1, l))

                    elif type(element) is sg.PySimpleGUI.Table:
                        logger.debug("update_elements: Table selector found...")
                        # Populate entries
                        try:
                            columns = element.metadata["TableHeading"].columns()
                        except KeyError:
                            columns = None  # default to all columns

                        values = dataset.table_values(columns, mark_virtual=True)

                        # Get the primary key to select.
                        # Use the list above instead of getting it directly
                        # from the table, as the data has yet to be updated
                        pk = dataset.get_current_pk()

                        found = False
                        if len(values):
                            # set to index by pk
                            index = [[v.pk for v in values].index(pk)]
                            # calculate pk percentage position
                            pk_position = index[0] / len(values)
                            found = True
                        else:  # if empty
                            index = []
                            pk_position = 0

                        logger.debug(f"Selector:: index:{index} found:{found}")

                        # Update table, and set vertical scroll bar to follow
                        update_table_element(
                            self.window, element, values, index, pk_position
                        )

    def requery_all(
        self,
        select_first: bool = True,
        filtered: bool = True,
        update_elements: bool = True,
        requery_dependents: bool = True,
    ) -> None:
        """
        Requeries all `DataSet` objects associated with this `Form`. This effectively
        re-loads the data from the database into `DataSet` objects.

        :param select_first: passed to `DataSet.requery()` -> `DataSet.first()`. If
            True, the first record will be selected after the requery
        :param filtered: passed to `DataSet.requery()`. If True, the relationships will
            be considered and an appropriate WHERE clause will be generated. False will
            display all records from the table.
        :param update_elements: passed to `DataSet.requery()` -> `DataSet.first()` to
            `Form.update_elements()`. Note that the select_first parameter must = True
            to use this parameter.
        :param requery_dependents: passed to `DataSet.requery()` -> `DataSet.first()` to
            `Form.requery_dependents()`. Note that the select_first parameter
            must = True to use this parameter.
        :returns: None
        """
        # TODO: It would make sense to reorder these, and put filtered first
        # then select_first/update/dependents
        logger.info("Requerying all datasets")
        for data_key in self.datasets:
            if Relationship.get_parent(data_key) is None:
                self[data_key].requery(
                    select_first=select_first,
                    filtered=filtered,
                    update_elements=update_elements,
                    requery_dependents=requery_dependents,
                )

    def process_events(self, event: str, values: list) -> bool:
        """
        Process mapped events for this specific `Form` instance.

        Not to be confused with the main `process_events()`, which processes events for
        ALL `Form` instances. This should be called once per iteration in your event
        loop. Note: Events handled are responsible for requerying and updating elements
        as needed.

        :param event: The event returned by PySimpleGUI.read()
        :param values: the values returned by PySimpleGUI.read()
        :returns: True if an event was handled, False otherwise
        """
        if self.window is None:
            logger.info(
                "***** Form appears to be unbound. "
                "Do you have frm.bind(win) in your code? *****"
            )
            return False
        if event:
            for e in self.event_map:
                if e["event"] == event:
                    logger.debug(f"Executing event {event} via event mapping.")
                    e["function"]()
                    logger.debug("Done processing event!")
                    return True

            # Check for  selector events
            for _data_key, dataset in self.datasets.items():
                if len(dataset.selector):
                    for e in dataset.selector:
                        element: sg.Element = e["element"]
                        if element.key == event and len(dataset.rows) > 0:
                            changed = False  # assume that a change will not take place
                            if type(element) == sg.PySimpleGUI.Listbox:
                                row = values[element.Key][0]
                                dataset.set_by_pk(row.get_pk())
                                changed = True
                            elif type(element) == sg.PySimpleGUI.Slider:
                                dataset.set_by_index(int(values[event]) - 1)
                                changed = True
                            elif type(element) == sg.PySimpleGUI.Combo:
                                row = values[event]
                                dataset.set_by_pk(row.get_pk())
                                changed = True
                            elif type(element) is sg.PySimpleGUI.Table:
                                index = values[event][0]
                                pk = self.window[event].Values[index].pk

                                # no need to update the selector!
                                dataset.set_by_pk(pk, True, omit_elements=[element])

                                changed = True
                            if changed and "record_changed" in dataset.callbacks:
                                dataset.callbacks["record_changed"](self, self.window)
                            return changed
        return False

    def update_element_states(
        self, table: str, disable: bool = None, visible: bool = None
    ) -> None:
        """
        Disable/enable and/or show/hide all elements associated with a table.

        :param table: table name associated with elements to disable/enable
        :param disable: True/False to disable/enable element(s), None for no change
        :param visible: True/False to make elements visible or not, None for no change
        :returns: None
        """
        for mapped in self.element_map:
            if mapped.table != table:
                continue
            element = mapped.element
            if type(element) in [
                sg.PySimpleGUI.InputText,
                sg.PySimpleGUI.MLine,
                sg.PySimpleGUI.Combo,
                sg.PySimpleGUI.Checkbox,
            ]:
                # if element.Key in self.window.key_dict.keys():
                logger.debug(
                    f"Updating element {element.Key} to disabled: "
                    f"{disable}, visible: {visible}"
                )
                if disable is not None:
                    element.update(disabled=disable)
                if visible is not None:
                    element.update(visible=visible)


# =====================================================================================
# MAIN PYSIMPLESQL UTILITY FUNCTIONS
# =====================================================================================
# These functions exist as utilities to the pysimplesql module
# This is a dummy class for documenting utility functions
class Utility:

    """
    Utility functions are a collection of functions and classes that directly improve on
    aspects of the pysimplesql module.

    See the documentation for the following utility functions:
    `process_events()`, `update_elements()`, `bind()`, `simple_transform()`, `KeyGen()`,

    Note: This is a dummy class that exists purely to enhance documentation and has no
    use to the end user.
    """

    pass


def process_events(event: str, values: list) -> bool:
    """
    Process mapped events for ALL Form instances.

    Not to be confused with `Form.process_events()`, which processes events for
    individual `Form` instances. This should be called once per iteration in your event
    loop. Note: Events handled are responsible for requerying and updating elements as
    needed.

    :param event: The event returned by PySimpleGUI.read()
    :param values: the values returned by PySimpleGUI.read()
    :returns: True if an event was handled, False otherwise
    """
    handled = False
    for i in Form.instances:
        if i.process_events(event, values):
            handled = True
    return handled


def update_elements(data_key: str = None, edit_protect_only: bool = False) -> None:
    """
    Updated the GUI elements to reflect values from the database for ALL Form instances.
    Not to be confused with `Form.update_elements()`, which updates GUI elements for
    individual `Form` instances.

    :param data_key: (optional) key of `DataSet` to update elements for, otherwise
        updates elements for all datasets.
    :param edit_protect_only: (optional) If true, only update items affected by
        edit_protect.
    :returns: None
    """
    for i in Form.instances:
        i.update_elements(data_key, edit_protect_only)


def bind(win: sg.Window) -> None:
    """
    Bind ALL forms to window. Not to be confused with `Form.bind()`, which binds
    specific forms to the window.

    :param win: The PySimpleGUI window to bind all forms to
    :returns: None
    """
    for i in Form.instances:
        i.bind(win)


def simple_transform(dataset: DataSet, row, encode):
    """
    Convenience transform function that makes it easier to add transforms to your
    records.
    """
    for col, function in dataset._simple_transform.items():
        if col in row:
            msg = f"Transforming {col} from {row[col]}"
            if encode == TFORM_DECODE:
                row[col] = function["decode"](row, col)
            else:
                row[col] = function["encode"](row, col)
            logger.debug(f"{msg} to {row[col]}")


def update_table_element(
    window: sg.Window,
    element: Type[sg.Table],
    values: List[TableRow],
    select_rows: List[int],
    vscroll_position: float = None,
) -> None:
    """
    Updates a PySimpleGUI sg.Table with new data and suppresses extra events emitted.

    Call this function instead of simply calling update() on a sg.Table element.
    The reason is that updating the selection or values will in turn fire more
    changed events, adding up to an endless loop of events.

    :param window: A PySimpleGUI Window containing the sg.Table element to be updated.
    :param element: The sg.Table element to be updated.
    :param values: A list of table rows to update the sg.Table with.
    :param select_rows: List of rows to select as if user did.
    :param vscroll_position: From 0 to 1.0, the percentage from the top to move
        scrollbar to.

    :returns: None
    """
    # Disable handling for "<<TreeviewSelect>>" event
    element.Widget.unbind("<<TreeviewSelect>>")
    # update element
    element.update(values=values, select_rows=select_rows)
    # set vertical scroll bar to follow selected element
    # call even for 0.0, so that a 'reset sort' repositions vscroll to top.
    if vscroll_position is not None:
        element.set_vscroll_position(vscroll_position)
    window.refresh()  # Event handled and bypassed
    # Enable handling for "<<TreeviewSelect>>" event
    element.widget.bind("<<TreeviewSelect>>", element._treeview_selected)


def checkbox_to_bool(value):
    """
    Allows a variety of checkbox values to still return True or False.

    :param value: Value to convert into True or False
    :returns: bool
    """
    return str(value).lower() in ["y", "yes", "t", "true", "1"]


class Popup:

    """
    Popup helper class.

    Has popup functions for internal use. Stores last info popup as last_info
    """

    def __init__(self):
        """
        Create a new Popup instance
        :returns: None.
        """
        self.last_info_msg = ""
        self.popup_info = None

    def ok(self, title, msg):
        """
        Internal use only.

        Creates sg.Window with LanguagePack OK button
        """
        msg = msg.splitlines()
        layout = [[sg.T(line, font="bold")] for line in msg]
        layout.append(
            sg.Button(
                button_text=lang.button_ok,
                key="ok",
                use_ttk_buttons=themepack.use_ttk_buttons,
                pad=themepack.popup_button_pad,
            )
        )
        popup_win = sg.Window(
            title,
            layout=[layout],
            keep_on_top=True,
            modal=True,
            finalize=True,
            ttk_theme=themepack.ttk_theme,
            element_justification="center",
        )

        while True:
            event, values = popup_win.read()
            if event in [sg.WIN_CLOSED, "Exit", "ok"]:
                break
        popup_win.close()

    def yes_no(self, title, msg):
        """
        Internal use only.

        Creates sg.Window with LanguagePack Yes/No button
        """
        msg = msg.splitlines()
        layout = [[sg.T(line, font="bold")] for line in msg]
        layout.append(
            sg.Button(
                button_text=lang.button_yes,
                key="yes",
                use_ttk_buttons=themepack.use_ttk_buttons,
                pad=themepack.popup_button_pad,
            )
        )
        layout.append(
            sg.Button(
                button_text=lang.button_no,
                key="no",
                use_ttk_buttons=themepack.use_ttk_buttons,
                pad=themepack.popup_button_pad,
            )
        )
        popup_win = sg.Window(
            title,
            layout=[layout],
            keep_on_top=True,
            modal=True,
            finalize=True,
            ttk_theme=themepack.ttk_theme,
            element_justification="center",
        )

        while True:
            event, values = popup_win.read()
            if event in [sg.WIN_CLOSED, "Exit", "no", "yes"]:
                result = event
                break
        popup_win.close()
        return result

    def info(
        self, msg: str, display_message: bool = True, auto_close_seconds: int = None
    ):
        """
        Creates sg.Window with no buttons to display passed in message string, and
        writes message to to self.last_info. Uses title as defined in
        lang.info_popup_title. By default auto-closes in seconds as defined in
        themepack.popup_info_auto_close_seconds.

        :param msg: String to display as message
        :param display_message: (optional) By default True. False only writes
            [title,msg] to self.last_info.
        :param auto_close_seconds: (optional) Gets value from
            themepack.info_popup_auto_close_seconds by default.
        :returns: None
        """
        """
        Internal use only.

        Creates sg.Window with no buttons, auto-closing after seconds as defined in
        themepack
        """
        title = lang.info_popup_title
        if auto_close_seconds is None:
            auto_close_seconds = themepack.popup_info_auto_close_seconds
        self.last_info_msg = msg
        if display_message:
            msg = msg.splitlines()
            layout = [sg.T(line, font="bold") for line in msg]
            self.popup_info = sg.Window(
                title=title,
                layout=[layout],
                no_titlebar=False,
                keep_on_top=True,
                finalize=True,
                alpha_channel=themepack.popup_info_alpha_channel,
                element_justification="center",
                ttk_theme=themepack.ttk_theme,
            )
            threading.Thread(
                target=self.auto_close,
                args=(self.popup_info, auto_close_seconds),
                daemon=True,
            ).start()

    def auto_close(self, window: sg.Window, seconds: int):
        """
        Use in a thread to automatically close the passed in sg.Window.

        :param window: sg.Window object to close
        :param seconds: Seconds to keep window open
        :returns: None
        """
        step = 1
        while step <= seconds:
            sleep(1)
            step += 1
        self.close(window)

    def close(self, window):
        window.close()


class ProgressBar:
    def __init__(self, title: str, max_value: int = 100, hide_delay: int = 100):
        """
        Creates a progress bar window with a message label and a progress bar.

        The progress bar is updated by calling the `update` method to update the
        progress in incremental steps until the `close` method is called.

        :param title: Title of the window
        :param max_value: Maximum value of the progress bar
        :param hide_delay: Delay in milliseconds before displaying the Window
        :returns: None
        """
        self.win = None
        self.title = title
        self.layout = [
            [sg.Text("", key="message", size=(50, 2))],
            [
                sg.ProgressBar(
                    max_value,
                    orientation="h",
                    size=(30, 20),
                    key="bar",
                    style=themepack.ttk_theme,
                )
            ],
        ]

        self.max = max
        self.hide_delay = hide_delay
        self.start_time = time() * 1000
        self.update_queue = queue.Queue()  # Thread safe
        self.animate_thread = None
        self._stop_event = threading.Event()  # Added stop event
        self.last_phrase_time = None
        self.phrase_index = 0

    def update(self, message: str, current_count: int):
        """
        Updates the progress bar with the current progress message and value.
        :param message: Message to display
        :param current_count: Current value of the progress bar
        :returns: None
        """
        if time() * 1000 - self.start_time < self.hide_delay:
            return

        if self.win is None:
            self._create_window()

        self.win["message"].update(message)
        self.win["bar"].update(current_count=current_count)

    def close(self):
        """
        Closes the progress bar window.

        :returns: None
        """
        if self.win is not None:
            self.win.close()

    def _create_window(self):
        self.win = sg.Window(
            self.title,
            layout=self.layout,
            keep_on_top=True,
            finalize=True,
            ttk_theme=themepack.ttk_theme,
        )


class ProgressAnimate:
    def __init__(self, title: str, config: dict = None):
        """
        Creates an animated progress bar with a message label.

        The progress bar will animate indefinitely, until the process passed in to the
        `run` method finishes.

        The config for the animated progress bar contains oscillators for the bar
        divider and colors, a list of phrases to be displayed, and the number of seconds
        to elapse between phrases.  This is all specified in the config dict
        as follows:
        my_oscillators = {
            # oscillators for the bar divider and colors
            "bar": {"value_start": 0, "value_range": 100, "period": 3, "offset": 0},
            "red": {"value_start": 0, "value_range": 255, "period": 2, "offset": 0},
            "green": {"value_start": 0, "value_range": 255, "period": 3, "offset": 120},
            "blue": {"value_start": 0, "value_range": 255, "period": 4, "offset": 240},

            # phrases to display and the number of seconds to elapse between phrases
            "phrases": [
                "Loading...", "Please be patient...", "This may take a while...",
                "Almost done...", "Almost there...", "Just a little longer...",
                "Please wait...", "Still working...",
            ],
            "phrase_delay": 2
        }
        Defaults are used for any keys that are not specified in the dictionary.

        :param title: Title of the window
        :param config: Dictionary of configuration options as listed above
        :returns: None
        """
        default_config = {
            # oscillators for the bar divider and colors
            "bar": {"value_start": 0, "value_range": 100, "period": 3, "offset": 0},
            "red": {"value_start": 0, "value_range": 255, "period": 2, "offset": 0},
            "green": {"value_start": 0, "value_range": 255, "period": 3, "offset": 120},
            "blue": {"value_start": 0, "value_range": 255, "period": 4, "offset": 240},
            # phrases to display and the number of seconds to elapse between phrases
            "phrases": lang.animate_phrases,
            "phrase_delay": 5,
        }
        if config is None:
            config = {}

        if type(config) is not dict:
            raise ValueError("config must be a dictionary")

        if set(config.keys()) - set(default_config.keys()):
            raise NotImplementedError(
                f"config may only contain keys: {default_config.keys()}"
            )

        for k in ["bar", "red", "green", "blue"]:
            if k in config and not all(isinstance(v, (int, float)) for v in config[k]):
                raise ValueError(f"values for {k} component must all be numeric")
            required_keys = {"value_start", "value_range", "period", "offset"}
            if k in config and not required_keys.issubset(set(config.keys())):
                raise ValueError(f"{k} must contain all of {required_keys}")

        if "phrases" in config:
            if type(config["phrases"]) is not list:
                raise ValueError("phrases must be a list")
            if not all(isinstance(v, str) for v in config["phrases"]):
                raise ValueError("phrases must be a list of strings")

        if "phrase_delay" in config and not all(
            isinstance(v, (int, float)) for v in config["phrase_delay"]
        ):  # noqa SIM102
            raise ValueError("phrase_delay must be numeric")

        self.config = {**default_config, **config}

        self.title = title
        self.win: sg.Window = None
        self.layout = [
            [sg.Text("", key="message", size=(50, 2))],
            [
                sg.ProgressBar(
                    100,
                    orientation="h",
                    size=(30, 20),
                    key="bar",
                    style=themepack.ttk_theme,
                )
            ],
        ]
        self.last_phrase_time = None
        self.phrase_index = 0
        self.completed = asyncio.Event()

    def run(self, fn: callable, *args, **kwargs):
        """
        Runs the function in a separate co-routine, while animating the progress bar in
        another.
        """
        if not callable(fn):
            raise ValueError("fn must be a callable")

        return asyncio.run(self._dispatch(fn, *args, **kwargs))

    def close(self):
        self.win = None

    async def _gui(self):
        if self.win is None:
            self.win = sg.Window(
                self.title,
                layout=self.layout,
                keep_on_top=True,
                finalize=True,
                ttk_theme=themepack.ttk_theme,
            )

        current_count = 0
        while not self.completed.is_set():
            current_count += 1
            self._animate(self.config)
            await asyncio.sleep(0.05)
        self.win.close()

    async def run_process(self, fn: callable, *args, **kwargs):
        loop = asyncio.get_running_loop()
        try:
            result = await loop.run_in_executor(
                None, functools.partial(fn, *args, **kwargs)
            )
            return result
        except Exception as e:  # noqa: BLE001
            print(f"\nAn error occurred in the process: {e}")
            raise e  # Pass the exception along to the caller
        finally:
            self.completed.set()

    async def _dispatch(self, fn: callable, *args, **kwargs):
        # Dispatch to the multiple asyncio co-processes
        gui_task = asyncio.create_task(self._gui())
        result = await self.run_process(fn, *args, **kwargs)
        await gui_task
        return result  # noqa RET504

    def _animate(self, config: dict = None):
        def _oscillate_params(oscillator):
            return (
                oscillator["value_start"],
                oscillator["value_range"],
                oscillator["period"],
                oscillator["offset"],
            )

        # oscillate the bar back and forth
        count = self._oscillate(*_oscillate_params(config["bar"]))

        # oscillate red color channel
        cr = self._oscillate(*_oscillate_params(config["red"]))

        # oscillate green color channel
        cg = self._oscillate(*_oscillate_params(config["blue"]))

        # oscillate blue color channel
        cb = self._oscillate(*_oscillate_params(config["green"]))

        color_1 = f"#{cr:02x}{cg:02x}{cb:02x}"
        color_2 = f"#{255-cg:02x}{255-cb:02x}{255-cr:02x}"
        message = self._animated_message(config["phrases"], config["phrase_delay"])

        self.win["message"].update(message)
        self.win["bar"].update(current_count=count, bar_color=(color_1, color_2))

    @staticmethod
    def _oscillate(value_start: int, value_range: int, period: float, offset: int):
        millis = int(round(time() * 1000))
        t = (millis % (period * 1000)) / (period * 1000)
        angle = t * 2 * math.pi + math.radians(offset)
        sin_value = math.sin(angle)
        return int((sin_value + 1) * value_range / 2 + value_start)

    def _animated_message(self, phrases: list, phrase_delay: float):
        # Cycle through the messages at the specified interval
        current_time = time()
        if (
            self.last_phrase_time is None
            or current_time - self.last_phrase_time > phrase_delay
        ):
            current_message = phrases[self.phrase_index]
            self.phrase_index = (self.phrase_index + 1) % len(phrases)
            self.last_phrase_time = current_time
        else:
            current_message = phrases[(self.phrase_index - 1)]
        return current_message


class LangFormat(dict):

    """
    This is a convenience class used by LanguagePack format_map calls, allowing users to
    not include expected variables.

    Note: This is typically not used by the end user.
    """

    def __missing__(self, key):
        return None


class KeyGen:

    """
    The keygen system provides a mechanism to generate unique keys for use as
    PySimpleGUI element keys.

    This is needed because many auto-generated items will have the same name.  If for
    example you had two save buttons on the screen at the same time, they must have
    unique names.  The keygen will append a separator and an incremental number to keys
    that would otherwise be duplicates. A global KeyGen instance is created
    automatically, see `keygen` for info.
    """

    def __init__(self, separator="!"):
        """
        Create a new KeyGen instance.

        :param separator: The default separator that goes between the key and the
            incremental number
        :returns: None
        """
        self._keygen = {}
        self._separator = separator

    def get(self, key: str, separator: str = None) -> str:
        """
        Get a generated key from the `KeyGen`.

        :param key: The key from which to generate the new key.  If the key has not been
            used before, then it will be returned unmodified.  For each successive call
            with the same key, it will be appended with the separator character and an
            incremental number.  For example, if the key 'button' was passed to
            `KeyGen.get()` 3 times in a row, then the keys 'button', 'button:1', and
            'button:2' would be returned respectively.
        :param separator: (optional) override the default separator wth this separator
        :returns: None
        """
        if separator is None:
            separator = self._separator

        # Generate a unique key by attaching a sequential integer to the end
        if key not in self._keygen:
            self._keygen[key] = 0
        return_key = key
        if self._keygen[key] > 0:
            # only modify the key if it is a duplicate!
            return_key += f"{separator}{str(self._keygen[key])}"
        logger.debug(f"Key generated: {return_key}")
        self._keygen[key] += 1
        return return_key

    def reset_key(self, key: str) -> None:
        """
        Reset the generation sequence for the supplied key.

        :param key: The base key to reset te sequence for
        """
        with contextlib.suppress(KeyError):
            del self._keygen[key]

    def reset(self) -> None:
        """
        Reset the entire `KeyGen` and remove all keys.

        :returns: None
        """
        self._keygen = {}

    def reset_from_form(self, frm: Form) -> None:
        """
        Reset keys from the keygen that were from mapped PySimpleGUI elements of that
        `Form`.

        :param frm: The `Form` from which to get the list of mapped elements
        :returns: None
        """
        # reset keys related to form
        for mapped in frm.element_map:
            self.reset_key(mapped.element.key)


# create a global KeyGen instance
keygen = KeyGen(separator=":")
"""
This is a global keygen instance for general purpose use.

See `KeyGen` for more info
"""

# Convenience dicts for example database connection
postgres_examples = {
    "host": "tommy2.heliohost.org",
    "user": "pysimplesql_user",
    "password": "pysimplesql",
    "database": "pysimplesql_examples",
}

mysql_examples = {
    "host": "tommy2.heliohost.org",
    "user": "pysimplesql_user",
    "password": "pysimplesql",
    "database": "pysimplesql_examples",
}


# -------------------------------------------------------------------------------------
# CONVENIENCE FUNCTIONS
# -------------------------------------------------------------------------------------
# Convenience functions aide in building PySimpleGUI interfaces
# that work well with pysimplesql.
# TODO: How to save Form in metadata?  Perhaps give forms names and reference them?
#       For example - give forms names!  and reference them by name string
#       They could even be converted later to a real form during form creation?


# This is a dummy class for documenting convenience functions
class Convenience:

    """
    Convenience functions are a collection of functions and classes that aide in
    building PySimpleGUI layouts that conform to pysimplesql standards so that your
    database application is up and running quickly, and with all the great automatic
    functionality pysimplesql has to offer. See the documentation for the following
    convenience functions: `field()`, `selector()`, `actions()`, `TableHeadings`.

    Note: This is a dummy class that exists purely to enhance documentation and has no
    use to the end user.
    """

    pass


def field(
    field: str,
    element: Type[sg.Element] = sg.I,
    size: Tuple[int, int] = None,
    label: str = "",
    no_label: bool = False,
    label_above: bool = False,
    quick_editor: bool = True,
    filter=None,
    key=None,
    use_ttk_buttons=None,
    pad=None,
    **kwargs,
) -> sg.Column:
    """
    Convenience function for adding PySimpleGUI elements to the Window, so they are
    properly configured for pysimplesql. The automatic functionality of pysimplesql
    relies on accompanying metadata so that the `Form.auto_add_elements()` can pick them
    up. This convenience function will create a text label, along with an element with
    the above metadata already set up for you. Note: The element key will default to the
    record name if none is supplied. See `set_label_size()`, `set_element_size()` and
    `set_mline_size()` for setting default sizes of these elements.

    :param field: The database record in the form of table.column I.e. 'Journal.entry'
    :param element: (optional) The element type desired (defaults to PySimpleGUI.Input)
    :param size: Overrides the default element size that was set with
        `set_element_size()` for this element only.
    :param label: The text/label will automatically be generated from the column name.
        If a different text/label is desired, it can be specified here.
    :param no_label: Do not automatically generate a label for this element
    :param label_above: Place the label above the element instead of to the left.
    :param quick_editor: For records that reference another table, place a quick edit
        button next to the element
    :param filter: Can be used to reference different `Form`s in the same layout. Use a
        matching filter when creating the `Form` with the filter parameter.
    :param key: (optional) The key to give this element. See note above about the
        default auto generated key.
    :param kwargs: Any additional arguments will be passed to the PySimpleGUI element.
    :returns: Element(s) to be used in the creation of PySimpleGUI layouts.  Note that
        this function actually creates multiple Elements wrapped in a PySimpleGUI
        Column, but can be treated as a single Element.
    """
    # TODO: See what the metadata does after initial setup is complete - needed anymore?

    if use_ttk_buttons is None:
        use_ttk_buttons = themepack.use_ttk_buttons
    if pad is None:
        pad = themepack.quick_editor_button_pad

    # if Record imply a where clause (indicated by ?) If so, strip out the info we need
    if "?" in field:
        table_info, where_info = field.split("?")
        label_text = (
            where_info.split("=")[1].replace("fk", "").replace("_", " ").capitalize()
            + ":"
        )
    else:
        table_info = field
        label_text = (
            table_info.split(".")[1].replace("fk", "").replace("_", " ").capitalize()
            + ":"
        )
    table, column = table_info.split(".")

    key = field if key is None else key
    key = keygen.get(key)

    if "values" in kwargs:
        first_param = kwargs["values"]
        del kwargs["values"]  # make sure we don't put it in twice
    else:
        first_param = ""

    if element.__name__ == "Multiline":
        layout_element = element(
            first_param,
            key=key,
            size=size or themepack.default_mline_size,
            metadata={
                "type": TYPE_RECORD,
                "Form": None,
                "filter": filter,
                "field": field,
                "data_key": key,
            },
            **kwargs,
        )
    else:
        layout_element = element(
            first_param,
            key=key,
            size=size or themepack.default_element_size,
            metadata={
                "type": TYPE_RECORD,
                "Form": None,
                "filter": filter,
                "field": field,
                "data_key": key,
            },
            **kwargs,
        )
    layout_label = sg.T(
        label if label else label_text,
        size=themepack.default_label_size,
        key=f"{key}:label",
    )
    # Marker for required (notnull) records
    layout_marker = sg.Column(
        [
            [
                sg.T(
                    themepack.marker_required,
                    key=f"{key}:marker",
                    text_color=sg.theme_background_color(),
                    visible=True,
                )
            ]
        ],
        pad=(0, 0),
    )
    if element.__name__ == "Text":  # don't show markers for sg.Text
        if no_label:
            layout = [[layout_element]]
        elif label_above:
            layout = [[layout_label], [layout_element]]
        else:
            layout = [[layout_label, layout_element]]
    else:
        if no_label:
            layout = [[layout_marker, layout_element]]
        elif label_above:
            layout = [[layout_label], [layout_marker, layout_element]]
        else:
            layout = [[layout_label, layout_marker, layout_element]]
    # Add the quick editor button where appropriate
    if element == sg.Combo and quick_editor:
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_QUICK_EDIT,
            "table": table,
            "column": column,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.quick_edit) is bytes:
            layout[-1].append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}.quick_edit"),
                    size=(1, 1),
                    image_data=themepack.quick_edit,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                )
            )
        else:
            layout[-1].append(
                sg.B(
                    themepack.quick_edit,
                    key=keygen.get(f"{key}.quick_edit"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                )
            )
    # return layout
    # TODO: Does this actually need wrapped in a sg.Col???
    return sg.Col(layout=layout, pad=(0, 0))


def actions(
    table: str,
    key=None,
    default: bool = True,
    edit_protect: bool = None,
    navigation: bool = None,
    insert: bool = None,
    delete: bool = None,
    duplicate: bool = None,
    save: bool = None,
    search: bool = None,
    search_size: Tuple[int, int] = (30, 1),
    bind_return_key: bool = True,
    filter: str = None,
    use_ttk_buttons: bool = None,
    pad=None,
    **kwargs,
) -> sg.Column:
    """
    Allows for easily adding record navigation and record action elements to the
    PySimpleGUI window The navigation elements are generated automatically (first,
    previous, next, last and search).  The action elements can be customized by
    selecting which ones you want generated from the parameters available.  This allows
    full control over what is available to the user of your database application. Check
    out `ThemePacks` to give any of these autogenerated controls a custom look!.

    Note: By default, the base element keys generated for PySimpleGUI will be
    `table:action` using the name of the table passed in the table parameter plus the
    action strings below separated by a colon: (I.e. Journal:table_insert) edit_protect,
    db_save, table_first, table_previous, table_next, table_last, table_duplicate,
    table_insert, table_delete, search_input, search_button. If you supply a key with
    the key parameter, then these additional strings will be appended to that key. Also
    note that these autogenerated keys also pass through the `KeyGen`, so it's possible
    that these keys could be table_last:action!1, table_last:action!2, etc.

    :param table: The table name that this "element" will provide actions for
    :param key: (optional) The base key to give the generated elements
    :param default: Default edit_protect, navigation, insert, delete, save and search to
        either true or false (defaults to True) The individual keyword arguments will
        trump the default parameter.  This allows for starting with all actions
        defaulting to False, then individual ones can be enabled with True - or the
        opposite by defaulting them all to True, and disabling the ones not needed with
        False.
    :param edit_protect: An edit protection mode to prevent accidental changes in the
        database. It is a button that toggles the ability on and off to prevent
        accidental changes in the database by enabling/disabling the insert, edit,
        duplicate, delete and save buttons.
    :param navigation: The standard << < > >> (First, previous, next, last) buttons for
        navigation
    :param insert: Button to insert new records
    :param delete: Button to delete current record
    :param duplicate: Button to duplicate current record
    :param save: Button to save record.  Note that the save button feature saves changes
        made to any table, therefore only one save button is needed per window.
    :param search: A search Input element. Size can be specified with the `search_size`
        parameter
    :param search_size: The size of the search input element
    :param bind_return_key: Bind the return key to the search button. Defaults to true.
    :param filter: Can be used to reference different `Form`s in the same layout.  Use a
        matching filter when creating the `Form` with the filter parameter.
    :param pad: The padding to use for the generated elements.
    :returns: An element to be used in the creation of PySimpleGUI layouts.  Note that
        this is technically multiple elements wrapped in a PySimpleGUI.Column, but acts
        as one element for the purpose of layout building.
    """

    if use_ttk_buttons is None:
        use_ttk_buttons = themepack.use_ttk_buttons
    if pad is None:
        pad = themepack.action_button_pad

    edit_protect = default if edit_protect is None else edit_protect
    navigation = default if navigation is None else navigation
    insert = default if insert is None else insert
    delete = default if delete is None else delete
    duplicate = default if duplicate is None else duplicate
    save = default if save is None else save
    search = default if search is None else search
    key = f"{table}:" if key is None else f"{key}:"

    layout = []

    # Form-level events
    if edit_protect:
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_EDIT_PROTECT_DB,
            "table": None,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.edit_protect) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}edit_protect"),
                    size=(1, 1),
                    image_data=themepack.edit_protect,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.edit_protect,
                    key=keygen.get(f"{key}edit_protect"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
    if save:
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_SAVE_DB,
            "table": None,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.save) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}db_save"),
                    image_data=themepack.save,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(themepack.save, key=keygen.get(f"{key}db_save"), metadata=meta)
            )

    # DataSet-level events
    if navigation:
        # first
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_FIRST,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.first) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}table_first"),
                    size=(1, 1),
                    image_data=themepack.first,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.first,
                    key=keygen.get(f"{key}table_first"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        # previous
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_PREVIOUS,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.previous) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}table_previous"),
                    size=(1, 1),
                    image_data=themepack.previous,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.previous,
                    key=keygen.get(f"{key}table_previous"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        # next
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_NEXT,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.next) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}table_next"),
                    size=(1, 1),
                    image_data=themepack.next,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.next,
                    key=keygen.get(f"{key}table_next"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        # last
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_LAST,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.last) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}table_last"),
                    size=(1, 1),
                    image_data=themepack.last,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.last,
                    key=keygen.get(f"{key}table_last"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
    if duplicate:
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_DUPLICATE,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.duplicate) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}table_duplicate"),
                    size=(1, 1),
                    image_data=themepack.duplicate,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.duplicate,
                    key=keygen.get(f"{key}table_duplicate"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
    if insert:
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_INSERT,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.insert) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}table_insert"),
                    size=(1, 1),
                    image_data=themepack.insert,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.insert,
                    key=keygen.get(f"{key}table_insert"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
    if delete:
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_DELETE,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.delete) is bytes:
            layout.append(
                sg.B(
                    "",
                    key=keygen.get(f"{key}table_delete"),
                    size=(1, 1),
                    image_data=themepack.delete,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
        else:
            layout.append(
                sg.B(
                    themepack.delete,
                    key=keygen.get(f"{key}table_delete"),
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                )
            )
    if search:
        meta = {
            "type": TYPE_EVENT,
            "event_type": EVENT_SEARCH,
            "table": table,
            "column": None,
            "function": None,
            "Form": None,
            "filter": filter,
        }
        if type(themepack.search) is bytes:
            layout += [
                sg.Input("", key=keygen.get(f"{key}search_input"), size=search_size),
                sg.B(
                    "",
                    key=keygen.get(f"{key}search_button"),
                    bind_return_key=bind_return_key,
                    size=(1, 1),
                    image_data=themepack.search,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                ),
            ]
        else:
            layout += [
                sg.Input("", key=keygen.get(f"{key}search_input"), size=search_size),
                sg.B(
                    themepack.search,
                    key=keygen.get(f"{key}search_button"),
                    bind_return_key=bind_return_key,
                    metadata=meta,
                    use_ttk_buttons=use_ttk_buttons,
                    pad=pad,
                    **kwargs,
                ),
            ]
    return sg.Col(layout=[layout], pad=(0, 0))


def selector(
    table: str,
    element: Type[sg.Element] = sg.LBox,
    size: Tuple[int, int] = None,
    filter: str = None,
    key: str = None,
    **kwargs,
) -> sg.Element:
    """
    Selectors in pysimplesql are special elements that allow the user to change records
    in the database application. For example, Listboxes, Comboboxes and Tables all
    provide a convenient way to users to choose which record they want to select. This
    convenience function makes creating selectors very quick and as easy as using a
    normal PySimpleGUI element.

    :param table: The table name in the database that this selector will act on
    :param element: The type of element you would like to use as a selector (defaults to
        a Listbox)
    :param size: The desired size of this selector element
    :param filter: Can be used to reference different `Form`s in the same layout. Use a
        matching filter when creating the `Form` with the filter parameter.
    :param key: (optional) The key to give to this selector. If no key is provided, it
        will default to table:selector using the table specified in the table parameter.
        This is also passed through the keygen, so if selectors all use the default
        name, they will be made unique. ie: Journal:selector!1, Journal:selector!2, etc.
    :param kwargs: Any additional arguments supplied will be passed on to the
        PySimpleGUI element.
    """

    key = f"{table}:selector" if key is None else key
    key = keygen.get(key)

    meta = {"type": TYPE_SELECTOR, "table": table, "Form": None, "filter": filter}
    if element == sg.Listbox:
        layout = element(
            values=(),
            size=size or themepack.default_element_size,
            key=key,
            select_mode=sg.LISTBOX_SELECT_MODE_SINGLE,
            enable_events=True,
            metadata=meta,
        )
    elif element == sg.Slider:
        layout = element(
            enable_events=True,
            size=size or themepack.default_element_size,
            orientation="h",
            disable_number_display=True,
            key=key,
            metadata=meta,
        )
    elif element == sg.Combo:
        w = themepack.default_element_size[0]
        layout = element(
            values=(),
            size=size or (w, 10),
            readonly=True,
            enable_events=True,
            key=key,
            auto_size_text=False,
            metadata=meta,
        )
    elif element == sg.Table:
        # Check if the headings arg is a Table heading...
        if kwargs["headings"].__class__.__name__ == "TableHeadings":
            # Overwrite the kwargs from the TableHeading info
            kwargs["visible_column_map"] = kwargs["headings"].visible_map()
            kwargs["col_widths"] = kwargs["headings"].width_map()
            kwargs["auto_size_columns"] = False  # let the col_widths handle it
            # Store the TableHeadings object in metadata
            # to complete setup on auto_add_elements()
            meta["TableHeading"] = kwargs["headings"]
        else:
            required_kwargs = ["headings", "visible_column_map", "num_rows"]
            for kwarg in required_kwargs:
                if kwarg not in kwargs:
                    raise RuntimeError(
                        f"DataSet selectors must use the {kwarg} keyword argument."
                    )

        # Create other kwargs that are required
        kwargs["enable_events"] = True
        kwargs["select_mode"] = sg.TABLE_SELECT_MODE_BROWSE
        kwargs["justification"] = "left"

        # Create a narrow column for displaying a * character for virtual rows.
        # This will have to be the 2nd column right after the pk
        kwargs["headings"].insert(0, "")
        kwargs["visible_column_map"].insert(0, 1)
        if "col_widths" in kwargs:
            kwargs["col_widths"].insert(0, 2)

        # Make an empty list of values
        vals = [[""] * len(kwargs["headings"])]

        # Change the headings parameter to be a list so
        # the heading doesn't display dicts when it first loads
        # The TableHeadings instance is already stored in metadata
        if kwargs["headings"].__class__.__name__ == "TableHeadings":
            kwargs["headings"] = kwargs["headings"].heading_names()

        layout = element(values=vals, key=key, metadata=meta, **kwargs)
    else:
        raise RuntimeError(f'Element type "{element}" not supported as a selector.')

    return layout


class TableHeadings(list):

    """
    This is a convenience class used to build table headings for PySimpleGUI.

    In addition, `TableHeading` objects can sort columns in ascending or descending
    order by clicking on the column in the heading in the PySimpleGUI Table element if
    the sort_enable parameter is set to True.
    """

    # store our instances
    instances = []

    def __init__(self, sort_enable: bool = True) -> None:
        """
        Create a new TableHeadings object.

        :param sort_enable: True to enable sorting by heading column
        :returns: None
        """
        self._sort_enable = sort_enable
        self._width_map = []
        self._visible_map = []

        # Store this instance in the master list of instances
        TableHeadings.instances.append(self)

    def add_column(
        self, column: str, heading_column: str, width: int, visible: bool = True
    ) -> None:
        """
        Add a new heading column to this TableHeading object.  Columns are added in the
        order that this method is called. Note that the primary key column does not need
        to be included, as primary keys are stored internally in the `TableRow` class.

        :param heading_column: The name of this columns heading (title)
        :param column: The name of the column in the database the heading column is for
        :param width: The width for this column to display within the Table element
        :param visible: True if the column is visible.  Typically, the only hidden
            column would be the primary key column if any. This is also useful if the
            `DataSet.rows` DataFrame has information that you don't want to display.
        :returns: None
        """
        self.append({"heading": heading_column, "column": column})
        self._width_map.append(width)
        self._visible_map.append(visible)

    def heading_names(self) -> List[str]:
        """
        Return a list of heading_names for use with the headings parameter of
        PySimpleGUI.Table.

        :returns: a list of heading names
        """
        return [c["heading"] for c in self]

    def columns(self):
        """
        Return a list of column names.

        :returns: a list of column names
        """
        return [c["column"] for c in self if c["column"] is not None]

    def visible_map(self) -> List[Union[bool, int]]:
        """
        Convenience method for creating PySimpleGUI tables.

        :returns: a list of visible columns for use with th PySimpleGUI Table
            visible_column_map parameter
        """
        return list(self._visible_map)

    def width_map(self) -> List[int]:
        """
        Convenience method for creating PySimpleGUI tables.

        :returns: a list column widths for use with th PySimpleGUI Table col_widths
            parameter
        """
        return list(self._width_map)

    def update_headings(
        self, element: sg.Table, sort_column=None, sort_order: int = None
    ) -> None:
        """
        Perform the actual update to the PySimpleGUI Table heading.
        Note: Not typically called by the end user.

        :param element: The PySimpleGUI Table element
        :param sort_column: The column to show the sort direction indicators on
        :param sort_order: A SORT_* constant (SORT_NONE, SORT_ASC, SORT_DESC)
        :returns: None
        """

        # Load in our marker characters.  We will use them to both display the
        # sort direction and to detect current direction
        try:
            asc = themepack.marker_sort_asc
        except AttributeError:
            asc = "\u25BC"
        try:
            desc = themepack.marker_sort_desc
        except AttributeError:
            desc = "\u25B2"

        for i, x in zip(range(len(self)), self):
            # Clear the direction markers
            x["heading"] = x["heading"].replace(asc, "").replace(desc, "")
            if (
                x["column"] == sort_column
                and sort_column is not None
                and sort_order != SORT_NONE
            ):
                x["heading"] += asc if sort_order == SORT_ASC else desc
            element.Widget.heading(i, text=x["heading"], anchor="w")

    def enable_sorting(self, element: sg.Table, fn: callable) -> None:
        """
        Enable the sorting callbacks for each column index.
        Note: Not typically used by the end user. Called from `Form.auto_map_elements()`

        :param element: The PySimpleGUI Table element associated with this TableHeading
        :param fn: A callback functions to run when a heading is clicked. The callback
            should take one column parameter.
        :returns: None
        """
        if self._sort_enable:
            for i in range(len(self)):
                if self[i]["column"] is not None:
                    element.widget.heading(
                        i, command=functools.partial(fn, self[i]["column"])
                    )
        self.update_headings(element)

    def insert(self, idx, heading_column: str, column: str = None, *args, **kwargs):
        super().insert(idx, {"heading": heading_column, "column": column})


class _SortCallbackWrapper:

    """Internal class used when sg.Table column headers are clicked."""

    def __init__(self, frm_reference: Form, data_key: str):
        """
        Create a new _SortCallbackWrapper object.

        :param frm_reference: `Form` object
        :param data_key: `DataSet` key
        :returns: None
        """
        self.frm: Form = frm_reference
        self.data_key = data_key

    def __call__(self, column):
        self.frm[self.data_key].sort_cycle(column, self.data_key, update_elements=True)


# ======================================================================================
# THEMEPACKS
# ======================================================================================
# Change the look and feel of your database application all in one place.
class ThemePack:

    """
    ThemePacks are user-definable objects that allow for the look and feel of database
    applications built with PySimpleGUI + pysimplesql.  This includes everything from
    icons, the ttk themes, to sounds. Pysimplesql comes with 3 pre-made ThemePacks:
    default (aka ss_small), ss_large and ss_text. Creating your own is easy as well! In
    fact, a ThemePack can be as simple as one line if you just want to change one aspect
    of the default ThemePack. Example:
        my_tp = {'search': 'Click here to search'} # I want a different search button.

    Once a ThemePack is created, it's very easy to use.  Here is a very simple example
    of using a ThemePack:
        ss.themepack(my_tp_dict_variable)
        # make a search button, using the 'search' key from the ThemePack
        sg.Button(ss.themepack.search, key='search_button')
    """

    default = {
        # Theme to use with ttk widgets.
        # -------------------------------
        # Choices (on Windows) include:
        # 'default', 'winnative', 'clam', 'alt', 'classic', 'vista', 'xpnative'
        "ttk_theme": "default",
        # Defaults for actions() buttons & popups
        # ----------------------------------------
        "use_ttk_buttons": True,
        "quick_editor_button_pad": (3, 0),
        "action_button_pad": (3, 0),
        "popup_button_pad": (5, 5),
        # Action buttons
        # ----------------------------------------
        # fmt: off
        'edit_protect': b'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',  # noqa E501
        'quick_edit': b'iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAGJ3pUWHRSYXcgcHJvZmlsZSB0eXBlIGV4aWYAAHjarVdZsuQmEPznFD4CVSwFx2GN8A18fCeiUG/zZtoRfnrdQoCKpDJJaDP++Xuav/DH7L3xQVLMMVr8+ewzFxSS3X/5+ibrr299sKfwUm/uBkaVw93tRynav6A+PF44Y1B9rTdJWzhpILoDX39ujbzK/Rkk6nnXk9dAeexCzEmeoVYN1LTjBUU//oa1b+vZvFQIstQDBnLMw5Gz13faCNz+FHwSvlGPftZllJ0jc92iBkNCXqZ3J9A+J+glyadk3rN/l96Sz0Xr3Vsuo+YIhV82UHird/cw/DywuxHxa0MaVj6mo585e5pz7NkVH5HRqIq6kk0nDDpWpNxdr0Vcgk9AWa4r40q22AbKu2224mqUicHKNOSpU6FJ47o3aoDoebDgztzYXXXJCWduYImcXxdNFjDWwSC7xsOAM+/4xkLXuPkar1HCyJ3QlQnBCK/8eJnfNf6Xy8zZVorIpjtXwMVL14CxmFvf6AVCaCpv4UrwuZR++6SfJVWPbivNCRMstu4QNdBDW+7i2aFfwH0vITLSNQBShLEDwJADAzaSCxTJCrMQIY8JBBUgZ+e5ggEKgTtAssfSYCOMJYOx8Y7Q1ZcDR17V8CYQEVx0Am6wpkCW9wH6EZ+goRJc8CGEGCQkE3Io0UUfQ4xR4jK5Ik68BIkikiRLSS75FFJMklLKqWTODh4YcsySU865FDYFAxXEKuhfUFO5uuprqLFKTTXX0iCf5ltosUlLLbfSubsOm+ixS0899zLIDDjF8COMOGSkkUeZ0Np0088w45SZZp7lZk1Z/bj+A2ukrPHF1OonN2uoNSInBC07CYszMMaewLgsBiBoXpzZRN7zYm5xZjNjUQQGyLC4MZ0WY6DQD+Iw6ebuwdxXvJmQvuKN/8ScWdT9H8wZUPfJ2y9Y62ufaxdjexWunFqH1Yf2kYrhVNamVr66TynlKlOengN5/LcEGP4KxHWInT2n0cr1xiiwKpqr29qb9N20X8QeqQ3otEeYEQ7Zhv8Wzwe+GvfAM1dnenTIwYWrtgGOx36Irqbh40boXZ/c+kIE7qMbO5TnvkHCis3bIDg8XHF6chNb7J6V/eJuroIbTVENSTP6svMDvy+0XHshmR5tTeD9qwlyrVEs7X5E0/jiNv4MvwpXtAz1F4VY69XV55qzhkiIP1hDlCaIj5JZ+dfAn3fpUV9AbzzYncCMhbdhYrPaWRmmYguAmve8cpu2VdHBGCsm00U61EoTqyfs9zP14vf0cU5C6rcg13kE60uVNti9of4BbOgHbANYYzUJt84cKNukAodmqmTNMBLk9wvSoRSXe1bEZubhaYjSBE35JHSTNtBx5x2ScjsdEf1fUJcVyvwAex7YEbB1cTTvdw+mEx6nIIVviHQJ0ZZpSHCJoUsI0lEhYL7DteDKESzAt+ULu6dtZnabpu1Pes7vunUgfbfDXfDQqtO8IsuKgszGA2KVNktdJxhEa1Snj8jMR05JjkhNsSKauQ6XcXDArCKssNX4G60e+mGIXczhuFvvd3icEarivBezf8WCwg2XdgGn2q0RbEJasLQXHza31s6oiYH0trbDzzxSb9ZIoDMVGM4YpMRikr2pC1xHeS2cmjunis2g5N5QYkJnSR43KwREPRx4/hOeeeAcVTsi2zNAMAp7Yl363YQDk8p7DLa6uvlCYF4pP5z4Uwib+pK8Tgp7+4hBZYUj1vBtJ/u35j530Vs15+bF6eLBjymhtucH0MVI9aq82poT5TAm/Lx8T522rV9Km1ZWnYRiE1Z/3WxjfDfCF3vQfK+6RjQQeir12E0Rqg8tgBp1y1axTSVtkpyJuko2azhjb61AfnL4TaDOvsnvpztN6X350aqrGoxP4zEXbQkZvzwUUIIyovDRCk4dDe6x9/413X6sYeak4u7rwX23S5on2+n9eHQ+/jdDP63l1n05sPPJSvTdbOsW6nCMWxTw4kCqieHKAqnnDpwUZ+Yft+wPTyz3+rv97qRR3MOS0m2C1by7oDu7dcR2FV6PSH8+RHwiuhNST0LKAXLOMtTqw5eiOWV3V9LZYb4V0nU3v1QYzoHmX+RGJBpl98L8AAABhGlDQ1BJQ0MgcHJvZmlsZQAAeJx9kT1Iw0AcxV9TpUVaBO0gopChOlkQFXHUKhShQqgVWnUwufQLmjQkLS6OgmvBwY/FqoOLs64OroIg+AHi5Oik6CIl/i8ptIjx4Lgf7+497t4BQqPMNKtrHND0qplKxMVMdlUMvCKIPoQxDFFmljEnSUl4jq97+Ph6F+NZ3uf+HGE1ZzHAJxLPMsOsEm8QT29WDc77xBFWlFXic+Ixky5I/Mh1xeU3zgWHBZ4ZMdOpeeIIsVjoYKWDWdHUiKeIo6qmU76QcVnlvMVZK9dY6578haGcvrLMdZpDSGARS5AgQkENJZRRRYxWnRQLKdqPe/gHHb9ELoVcJTByLKACDbLjB/+D391a+ckJNykUB7pfbPtjBAjsAs26bX8f23bzBPA/A1d6219pADOfpNfbWvQI6N0GLq7bmrIHXO4AA0+GbMqO5Kcp5PPA+xl9UxbovwV61tzeWvs4fQDS1FXyBjg4BEYLlL3u8e5gZ2//nmn19wNkDXKhWfC+CAAAAAZiS0dEAAAAAAAA+UO7fwAAAAlwSFlzAAALEwAACxMBAJqcGAAAAAd0SU1FB+QIEg0fJQXnbmsAAAKVSURBVDjLhZJPSFRRFMa/c++b55tGTZpSMZRStCyJFlEoLkSyWtQiyI1FUWRtIooWFS2yKHcG0aICN1IWCNWmQhfixqQokDAHpY3lFJiTZo7ju/e9e0+LwP6o9W3O6vvxfeccwjK6dPEirrS2IkmUE2loeCGkTBFwjIAxw4yinh4AAC0HMIlbSL0zmHs72SV7extldjaElDOS6CoDNwCgsLsbYjmA+q6Rk//xaN6p5kbRfIJDIjZK5YbWtjHQWRCNYqS+fukEmQebIYQTD3R6eJ7z883W83C8LZRpucRIJkl6HtZWVNBIIgH5t3n2fhUIBmxNu1K6WmdSUIl2aJLIab4MGEFhcvz41OfPgyGwuIIkA0Cc01o1KaXBzIC7Clnjd2j2yWFS1WsSBR2POiURNvX1/arw6W4ZYlEHjqD1YaAH5+f9XCEIvq8QiTgAiIIgNGZ4stDZ1ZIqaWwBfk9QFJdwBcOEpsv31UoiwFoGEUFKB8YYWLb7Ubk6FSZvLyQWAPD+1WPM2HKExlxXyt9mrWE34pIxhqJRD9ZastZ2Z2a/Pg2NRenZiQUAAUDHbmBvEzayj0FfF3qx2ArWWpMQPwMqpWbSGbXGy3KCdWdSf+xMAMDBZxorD5kGt67b8/KqGDwHImIpBRsTGiLsiXpuMOcvPrlYGMzlXulOxPbdI17biCwxTsYwMXOn6zovBQGbL6SWBjAzAGwgMNjNY7fuJnj7QxhZ8EFk5RxRyqL49JclP1YCgNYa/f3910pKSvLi8Tjp+TR9Q36XjhYf4NmxtFQTaHueXhJAZWVlcF0X1loeHR0NBgYG3sRisZORSGTo29QUampr8S8Jay2mp6dzieh1ZWXljpqamtogCIbCMPyvGQB+AKK0L000MH1KAAAAAElFTkSuQmCC',  # noqa E501
        'save': b'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',  # noqa E501
        'first': b'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',  # noqa E501
        'previous': b'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',  # noqa E501
        'next': b'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',  # noqa E501
        'last': b'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',  # noqa E501
        'insert': b'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',  # noqa E501
        'delete': b'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',  # noqa E501
        'duplicate': b'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',  # noqa E501
        'icon' : b'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',  # noqa E501
        "search": "Search",
        # fmt: on
        # Markers
        # ----------------------------------------
        "marker_virtual": "\u2731",
        "marker_required": "\u2731",
        "marker_required_color": "red2",
        # Sorting icons
        # ----------------------------------------
        "marker_sort_asc": "\u25BC",
        "marker_sort_desc": "\u25B2",
        # Info Popup defaults
        # ----------------------------------------
        "popup_info_auto_close_seconds": 1,
        "popup_info_alpha_channel": 0.85,
        # Default sizes for elements
        # ---------------------------
        # Label Size
        # Sets the default label (text) size when `field()` is used.
        # A label is static text that is displayed near the element to describe it.
        "default_label_size": (20, 1),  # (width, height)
        # Element Size
        # Sets the default element size when `field()` is used.
        # The size= parameter of `field()` will override this.
        "default_element_size": (30, 1),  # (width, height)
        # Mline size
        # Sets the default multi-line text size when `field()` is used.
        # The size= parameter of `field()` will override this.
        "default_mline_size": (30, 7),  # (width, height)
    }
    """
    Default Themepack.
    """

    def __init__(self, tp_dict: Dict[str, str] = {}) -> None:
        self.tp_dict = ThemePack.default

    def __getattr__(self, key):
        # Try to get the key from the internal tp_dict first.
        # If it fails, then check the default dict.
        try:
            return self.tp_dict[key]
        except KeyError:
            try:
                return ThemePack.default[key]
            except KeyError:
                raise AttributeError(f"ThemePack object has no attribute '{key}'")

    def __call__(self, tp_dict: Dict[str, str] = {}) -> None:
        """
        Update the ThemePack object from tp_dict.

        Example minimal ThemePack: NOTE: You can add additional keys if desired
            tp_example = {
                'ttk_theme : the name of a ttk theme
                'edit_protect' : either base64 image (eg b''), or string eg '', f''
                'quick_edit' : either base64 image (eg b''), or string eg '', f''
                'save' : either base64 image (eg b''), or string eg '', f''
                'first' : either base64 image (eg b''), or string eg '', f''
                'previous' : either base64 image (eg b''), or string eg '', f''
                'next' : either base64 image (eg b''), or string eg '', f''
                'last' : either base64 image (eg b'') or a string eg '', f''
                'insert' : either base64 image (eg b''), or string eg '', f''
                'delete' : either base64 image (eg b''), or string eg '', f''
                'duplicate' : either base64 image (eg b''), or string eg '', f''
                'search' : either base64 image (eg b''), or string eg '', f''
                'marker_virtual' : string eg '', f'',  unicode
                'marker_required' : string eg '', f'',  unicode
                'marker_required_color': string eg 'red', Tuple eg (255,0,0)
                'marker_sort_asc': string eg '', f'',  unicode
                'marker_sort_desc': string eg '', f'',  unicode
            }
        For Base64, you can convert a whole folder using https://github.com/PySimpleGUI/PySimpleGUI-Base64-Encoder # fmt: skip
        Remember to us b'' around the string.

        :param tp_dict: (optional) A dict formatted as above to create the ThemePack
            from. If one is not supplied, a default ThemePack will be generated.  Any
            keys not present in the supplied tp_dict will be generated from the default
            values.  Additionally, tp_dict may contain additional keys not specified in
            the minimal default ThemePack.
        :returns: None
        """  # noqa: E501
        # For default use cases, load the default directly to avoid the overhead
        # of __getattr__() going through 2 key reads
        if tp_dict == {}:
            tp_dict = ThemePack.default

        self.tp_dict = tp_dict


# set a default themepack
themepack = ThemePack()


# ======================================================================================
# LANGUAGEPACKS
# ======================================================================================
# Change the language text used throughout the program.
class LanguagePack:

    """
    LanguagePacks are user-definable collections of strings that allow for localization
    of strings and messages presented to the end user.

    Creating your own is easy as well! In fact, a LanguagePack can be as simple as one
    line if you just want to change one aspect of the default LanguagePack. Example:
        # I want the save popup to display this text in English in all caps
        lp_en = {'save_success': 'SAVED!'}
    """

    default = {
        # ------------------------------------------------------------------------------
        # Buttons
        # ------------------------------------------------------------------------------
        "button_cancel": " Cancel ",
        "button_ok": "  OK  ",
        "button_yes": " Yes ",
        "button_no": "  No  ",
        # ------------------------------------------------------------------------------
        # Prepopulate record values/prepends
        # ------------------------------------------------------------------------------
        # Text, Varchar, Char, Null Default, used exclusively for description_column
        "description_column_str_null_default": "New Record",
        # Prepended to parent description_column
        "duplicate_prepend": "Copy of ",
        # ------------------------------------------------------------------------------
        # Startup progress bar
        # ------------------------------------------------------------------------------
        "startup_form": "Creating Form",
        "startup_init": "Initializing",
        "startup_datasets": "Adding datasets",
        "startup_relationships": "Adding relationships",
        "startup_binding": "Binding window to Form",
        # ------------------------------------------------------------------------------
        # Progress bar displayed during sqldriver operations
        # ------------------------------------------------------------------------------
        "sqldriver_init": "{name} connection",
        "sqldriver_connecting": "Connecting to database",
        "sqldriver_execute": "Executing SQL commands",
        # ------------------------------------------------------------------------------
        # Default ProgressAnimate Phrases
        # ------------------------------------------------------------------------------
        "animate_phrases": [
            "Please wait...",
            "Still working...",
        ],
        # ------------------------------------------------------------------------------
        # Info Popups - no buttons
        # ------------------------------------------------------------------------------
        # Info Popup Title - universal
        "info_popup_title": "Info",
        # Form save_records
        "form_save_success": "Updates saved successfully.",
        "form_save_none": "There were no updates to save.",
        # DataSet save_record
        "dataset_save_empty": "There were no updates to save.",
        "dataset_save_none": "There were no changes to save!",
        "dataset_save_success": "Updates saved successfully.",
        # ------------------------------------------------------------------------------
        # Yes/No Popups
        # ------------------------------------------------------------------------------
        # Form prompt_save
        "form_prompt_save_title": "Unsaved Changes",
        "form_prompt_save": "You have unsaved changes!\nWould you like to save them first?",  # fmt: skip # noqa: E501
        # DataSet prompt_save
        "dataset_prompt_save_title": "Unsaved Changes",
        "dataset_prompt_save": "You have unsaved changes!\nWould you like to save them first?",  # fmt: skip # noqa: E501
        # Form save_records
        "form_save_problem_title": "Problem Saving",
        "form_save_partial": "Some updates were saved successfully;",
        "form_save_problem": "There was a problem saving updates to the following tables:\n{tables}.",  # fmt: skip # noqa: E501
        # DataSet save_record
        "dataset_save_callback_false_title": "Callback Prevented Save",
        "dataset_save_callback_false": "Updates not saved.",
        "dataset_save_keyed_fail_title": "Problem Saving",
        "dataset_save_keyed_fail": "Query failed: {exception}.",
        "dataset_save_fail_title": "Problem Saving",
        "dataset_save_fail": "Query failed: {exception}.",
        # ------------------------------------------------------------------------------
        # Delete
        # ------------------------------------------------------------------------------
        # DataSet delete_record
        "delete_title": "Confirm Deletion",
        "delete_cascade": "Are you sure you want to delete this record?\nKeep in mind that child records:\n({children})\nwill also be deleted!",  # fmt: skip # noqa: E501
        "delete_single": "Are you sure you want to delete this record?",
        # Failed Ok Popup
        "delete_failed_title": "Problem Deleting",
        "delete_failed": "Query failed: {exception}.",
        "delete_recursion_limit_error": "Delete Cascade reached max recursion limit.\nDELETE_CASCADE_RECURSION_LIMIT",  # fmt: skip # noqa: E501
        # ------------------------------------------------------------------------------
        # Duplicate
        # ------------------------------------------------------------------------------
        # Popup when record has children
        "duplicate_child_title": "Confirm Duplication",
        "duplicate_child": "This record has child records:\n(in {children})\nWhich records would you like to duplicate?",  # fmt: skip # noqa: E501
        "duplicate_child_button_dupparent": "Only duplicate this record.",
        "duplicate_child_button_dupboth": "Duplicate this record and its children.",
        # Popup when record is single
        "duplicate_single_title": "Confirm Duplication",
        "duplicate_single": "Are you sure you want to duplicate this record?",
        # Failed Ok Popup
        "duplicate_failed_title": "Problem Duplicating",
        "duplicate_failed": "Query failed: {exception}.",
        # ------------------------------------------------------------------------------
        # Quick Editor
        # ------------------------------------------------------------------------------
        "quick_edit_title": "Quick Edit - {data_key}",
        # ------------------------------------------------------------------------------
        # Error when importing module for driver
        # ------------------------------------------------------------------------------
        "import_module_failed_title": "Problem importing module",
        "import_module_failed": "Unable to import module neccessary for {name}\nException: {exception}\n\nTry `pip install {requires}`",  # fmt: skip # noqa: E501
    }
    """
    Default LanguagePack.
    """

    def __init__(self, lp_dict={}):
        self.lp_dict = type(self).default

    def __getattr__(self, key):
        # Try to get the key from the internal lp_dict first.
        # If it fails, then check the default dict.
        try:
            return self.lp_dict[key]
        except KeyError:
            try:
                return type(self).default[key]
            except KeyError:
                raise AttributeError(f"LanguagePack object has no attribute '{key}'")

    def __call__(self, lp_dict={}):
        """Update the LanguagePack instance."""
        # For default use cases, load the default directly to avoid the overhead
        # of __getattr__() going through 2 key reads
        if lp_dict == {}:
            lp_dict = type(self).default

        self.lp_dict = lp_dict


# set a default languagepack
lang = LanguagePack()


# ======================================================================================
# ABSTRACTION LAYERS
# ======================================================================================
# Database abstraction layers for a uniform API
# --------------------------------------------------------------------------------------


# This is a dummy class for documenting convenience functions
class Abstractions:

    """
    Supporting multiple databases in your application can quickly become very
    complicated and unmanageable. pysimplesql abstracts all of this complexity and
    presents a unified API via abstracting the main concepts of database programming.
    See the following documentation for a better understanding of how this is
    accomplished. `Column`, `ColumnInfo`, `SQLDriver`, `Sqlite`, `Mysql`, `Postgres`.

    Note: This is a dummy class that exists purely to enhance documentation and has no
    use to the end user.
    """

    pass


# ======================================================================================
# COLUMN ABSTRACTION
# ======================================================================================
# The column abstraction hides the complexity of dealing with SQL columns, getting their
# names, default values, data types, primary key status and notnull status
# --------------------------------------------------------------------------------------
class Column:

    """
    The `Column` class is a generic column class.  It holds a dict containing the column
    name, type  whether the column is notnull, whether the column is a primary key and
    the default value, if any. `Column`s are typically stored in a `ColumnInfo`
    collection. There are multiple ways to get information from a `Column`, including
    subscript notation, and via properties. The available column info via these methods
    are name, domain, notnull, default and pk See example:

    .. literalinclude:: ../doc_examples/Column.1.py
        :language: python
        :caption: Example code
    """

    def __init__(
        self,
        name: str,
        domain: str,
        notnull: bool,
        default: None,
        pk: bool,
        virtual: bool = False,
    ):
        self._column = {
            "name": name,
            "domain": domain,
            "notnull": notnull,
            "default": default,
            "pk": pk,
            "virtual": virtual,
        }

    def __str__(self):
        return f"Column: {self._column}"

    def __repr__(self):
        return f"Column: {self._column}"

    def __getitem__(self, item):
        return self._column[item]

    def __setitem__(self, key, value):
        self._column[key] = value

    def __lt__(self, other, key):
        return self._column[key] < other._column[key]

    def __contains__(self, item):
        return item in self._column

    def __getattr__(self, key):
        return self._column[key]

    def __setattr__(self, key, value):
        if key == "_column":
            super().__setattr__(key, value)
        else:
            self._column[key] = value

    def cast(self, value: any) -> any:
        """
        Cast a value to the appropriate data type as defined by the column info for the
        column. This can be useful for comparing values between the database and the
        GUI.

        :param value: The value you would like to cast
        :returns: The value, cast to a type as defined by the domain
        """
        # convert the data into the correct data type using the domain in ColumnInfo
        domain = self.domain

        # String type casting
        if domain in ["TEXT", "VARCHAR", "CHAR"]:
            # convert to str
            value = str(value)

        # Integer type casting
        elif domain in ["INT", "INTEGER", "BOOLEAN"]:
            try:
                value = int(value)
            except (ValueError, TypeError):
                value = str(value)

        # float type casting
        elif domain in ["REAL", "DOUBLE", "DECIMAL", "FLOAT"]:
            try:
                value = float(value)
            except ValueError:
                value = str(value)

        # Date casting
        elif domain == "DATE":
            try:
                value = datetime.strptime(value, "%Y-%m-%d").date()
            except TypeError:
                logger.debug(
                    f"Unable to cast {value} to a datetime.date object. "
                    f"Casting to string instead."
                )
                value = str(value)

        elif domain == "TIMESTAMP":
            timestamp_formats = ["%Y-%m-%d %H:%M:%S", "%Y-%m-%d %H:%M:%S.%f"]

            parsed = False
            for timestamp_format in timestamp_formats:
                try:
                    value = datetime.strptime(value, timestamp_format)
                    # value = dt.datetime()
                    parsed = True
                    break
                except ValueError:
                    pass

            if not parsed:
                logger.debug(
                    "Unable to cast datetime/time/timestamp. Casting to string instead."
                )
                value = str(value)

        # other date/time casting
        elif domain in [
            "TIME",
            "DATETIME",
        ]:  # TODO: i'm sure there is a lot of work to do here
            try:
                value = datetime.date(value)
            except TypeError:
                print(
                    "Unable to case datetime/time/timestamp. Casting to string instead."
                )
                value = str(value)
        return value


class ColumnInfo(List):

    """
    Column Information Class.

    The `ColumnInfo` class is a custom container that behaves like a List containing a
    collection of `Columns`. This class is responsible for maintaining information about
    all the columns (`Column`) in a table. While the individual `Column` elements of
    this collection contain information such as default values, primary key status, SQL
    data type, column name, and the notnull status - this class ties them all together
    into a collection and adds functionality to set default values for null columns and
    retrieve a dict representing a table row with all defaults already assigned.
    See example below:
    .. literalinclude:: ../doc_examples/ColumnInfo.1.py
        :language: python
        :caption: Example code
    """

    def __init__(self, driver: SQLDriver, table: str):
        self.driver = driver
        self.table = table

        # List of required SQL types to check against when user sets custom values
        self._domains = [
            "TEXT",
            "VARCHAR",
            "CHAR",
            "INTEGER",
            "REAL",
            "DOUBLE",
            "FLOAT",
            "DECIMAL",
            "BOOLEAN",
            "TIME",
            "DATE",
            "DATETIME",
            "TIMESTAMP",
        ]

        # Defaults to use for Null values returned from the database. These can be
        # overwritten by the user and support function calls as well by using
        # ColumnInfo.set_null_default() and ColumnInfo.set_null_defaults()
        self.null_defaults = {
            "TEXT": lang.description_column_str_null_default,
            "VARCHAR": lang.description_column_str_null_default,
            "CHAR": lang.description_column_str_null_default,
            "INT": 0,
            "INTEGER": 0,
            "REAL": 0.0,
            "DOUBLE": 0.0,
            "FLOAT": 0.0,
            "DECIMAL": 0.0,
            "BOOLEAN": 0,
            "TIME": lambda x: datetime.now().strftime("%H:%M:%S"),
            "DATE": lambda x: date.today().strftime("%Y-%m-%d"),
            "TIMESTAMP": lambda x: datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
            "DATETIME": lambda x: datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
        }
        super().__init__()

    def __contains__(self, item):
        if isinstance(item, str):
            return self._contains_key_value_pair("name", item)
        return super().__contains__(item)

    def __getitem__(self, item):
        if isinstance(item, str):
            return next((i for i in self if i.name == item), None)
        return super().__getitem__(item)

    def pk_column(self) -> Union[str, None]:
        """
        Get the pk_column for this colection of column_info.

        :returns: A string containing the column name of the PK column, or None if one
            was not found
        """
        for c in self:
            if c.pk:
                return c.name
        return None

    def names(self) -> List[str]:
        """
        Return a List of column names from the `Column`s in this collection.

        :returns: List of column names
        """
        return self._get_list("name")

    def col_name(self, idx: int) -> str:
        """
        Get the column name located at the specified index in this collection of
        `Column`s.

        :param idx: The index of the column to get the name from
        :returns: The name of the column at the specified index
        """
        return self[idx].name

    def default_row_dict(self, dataset: DataSet) -> dict:
        """
        Return a dictionary of a table row with all defaults assigned.

        This is useful for inserting new records to prefill the GUI elements.

        :param dataset: a pysimplesql DataSet object
        :returns: dict
        """
        d = {}
        for c in self:
            default = c.default
            domain = c.domain

            # First, check to see if the default might be a database function
            if self._looks_like_function(default):
                table = self.driver.quote_table(self.table)
                # TODO: may need AS column to support all databases?
                q = f"SELECT {default} AS val FROM {table};"

                rows = self.driver.execute(q)
                if rows.attrs["exception"] is None:
                    try:
                        default = rows.iloc[0]["val"]
                    except KeyError:
                        try:
                            default = rows.iloc[0]["VAL"]
                        except KeyError:
                            default = ""
                    d[c.name] = default
                    continue
                logger.warning(
                    f"There was an exception getting the default: {rows.exception}"
                )

            # The stored default is a literal value, lets try to use it:
            if default in [None, "None"]:
                try:
                    null_default = self.null_defaults[domain]
                except KeyError:
                    # Perhaps our default dict does not yet support this datatype
                    null_default = None

                # skip primary keys
                if not c.pk:
                    # put default in description_column
                    if c.name == dataset.description_column:
                        default = null_default

                    # put defaults in other fields
                    elif domain not in [
                        "TEXT",
                        "VARCHAR",
                        "CHAR",
                    ]:  # (don't put 'New Record' in other txt fields)
                        # If our default is callable, call it.
                        if callable(null_default):
                            default = null_default()
                        # Otherwise, assign it
                        else:
                            default = null_default
                    # string-like, not description_column
                    default = ""
            else:
                # Load the default that was fetched from the database
                # during ColumnInfo creation
                if domain in ["TEXT", "VARCHAR", "CHAR"]:
                    # strip quotes from default strings as they seem to get passed with
                    # some database-stored defaults
                    # strip leading and trailing quotes
                    default = c.default.strip("\"'")

            d[c.name] = default
            logger.debug(
                f"Default fetched from database function. Default value is: {default}"
            )
        if dataset.transform is not None:
            dataset.transform(dataset, d, TFORM_DECODE)
        return d

    def set_null_default(self, domain: str, value: object) -> None:
        """
        Set a Null default for a single SQL type.

        :param domain: The SQL type to set the default for
            ('INTEGER', 'TEXT', 'BOOLEAN', etc.)
        :param value: The new value to set the SQL type to. This can be a literal or
            even a callable
        :returns: None
        """
        if domain not in self._domains:
            RuntimeError(
                f"Unsupported SQL Type: {domain}. Supported types are: {self._domains}"
            )

        self.null_defaults[domain] = value

    def set_null_defaults(self, null_defaults: dict) -> None:
        """
        Set Null defaults for all SQL types.

        supported types:  'TEXT','VARCHAR', 'CHAR', 'INTEGER', 'REAL', 'DOUBLE',
            'FLOAT', 'DECIMAL', 'BOOLEAN', 'TIME', 'DATE', 'DATETIME', 'TIMESTAMP'
        :param null_defaults: A dict of SQL types and default values. This can be a
            literal or even a callable
        :returns: None
        """
        # Check if the null_defaults dict has all the required keys:
        if not all(key in null_defaults for key in self._domains):
            RuntimeError(
                f"The supplied null_defaults dictionary does not havle all required SQL"
                f" types. Required: {self._domains}"
            )

        self.null_defaults = null_defaults

    def get_virtual_names(self) -> List[str]:
        """
        Get a list of virtual column names.

        :returns: A List of column names that are virtual, or [] if none are present in
            this collections
        """
        return [c for c in self if not c.virtual]

    def _contains_key_value_pair(self, key, value):  # used by __contains__
        return any(key in d and d[key] == value for d in self)

    @staticmethod
    def _looks_like_function(
        s: str,
    ):  # TODO: check if something looks like a statement for complex defaults?  Regex?
        # check if the string is empty
        if not s:
            return False

        # If the entire string is in all caps, it looks like a function
        # (like in MySQL CURRENT_TIMESTAMP)
        if s.isupper():
            return True

        # find the index of the first opening parenthesis
        open_paren_index = s.find("(")

        # if there is no opening parenthesis, the string is not a function
        if open_paren_index == -1:
            return False

        # check if there is a name before the opening parenthesis
        name = s[:open_paren_index].strip()
        if not name.isidentifier():
            return False

        # find the index of the last closing parenthesis
        close_paren_index = s.rfind(")")

        # if there is no closing parenthesis, the string is not a function
        if close_paren_index == -1 or close_paren_index <= open_paren_index:
            return False

        # if all checks pass, the string looks like a function
        return True

    def _get_list(self, key: str) -> List:
        # returns a list of any key in the underlying Column instances. For example,
        # column names, types, defaults, etc.
        return [d[key] for d in self]


# ======================================================================================
# DATABASE ABSTRACTION
# ======================================================================================
# The database abstraction hides the complexity of dealing with multiple databases. The
# concept relies on individual "drivers" that derive from the SQLDriver class, and
# return a pandas DataFrame populated with the data, along with attrs set for the
# lastrowid and exceptions passed from the driver.
# --------------------------------------------------------------------------------------
class Result:
    """
    This is a "dummy" Result object that is a convenience for constructing a DataFrame
    that has the expected attrs set.
    """

    @classmethod
    def set(
        cls,
        row_data: dict = None,
        lastrowid: int = None,
        exception: Exception = None,
        column_info: ColumnInfo = None,
    ):
        """
        Create a pandas DataFrame with the row data and expected attrs set.

        :param row_data: A list of dicts of row data
        :param lastrowid: The inserted row ID from the last INSERT statement
        :param exception: Exceptions passed back from the SQLDriver
        :param column_info: An optional ColumnInfo object
        """
        df = pd.DataFrame(row_data)
        df.attrs["lastrowid"] = lastrowid
        df.attrs["exception"] = exception
        df.attrs["column_info"] = column_info

        # Store virtual flags for each row
        df.attrs["virtual"] = pd.Series(
            [False] * len(df.index), index=df.index, dtype="int64"
        )
        return df


class ReservedKeywordError(Exception):
    pass


class SQLDriver:

    """
    Abstract SQLDriver class.  Derive from this class to create drivers that conform to
    PySimpleSQL.  This ensures that the same code will work the same way regardless of
    which database is used.  There are a few important things to note: The commented
    code below is broken into methods that **MUST** be implemented in the derived class,
    methods that.

    **SHOULD** be implemented in the derived class, and methods that **MAY** need to be
    implemented in the derived class for it to work as expected. Most derived drivers
    will at least partially work by implementing the **MUST** have methods.

    See the source code for `Sqlite`, `Mysql` and `Postgres` for examples of how to
    construct your own driver.

    NOTE: SQLDriver.execute() should return a pandas DataFrame.  Additionally, by
    pysimplesql convention, the attrs["lastrowid"] should always be None unless and
    INSERT query is executed with SQLDriver.execute() or a record is inserted with
    SQLDriver.insert_record()
    """

    # ---------------------------------------------------------------------
    # MUST implement
    # in order to function
    # ---------------------------------------------------------------------
    def __init__(
        self,
        name: str,
        requires: List[str],
        placeholder="%s",
        table_quote="",
        column_quote="",
        value_quote="'",
    ):
        """
        Create a new SQLDriver instance This must be overridden in the derived class,
        which must call super().__init__(), and when finished call self.win_pb.close()
        to close the database.
        """
        # Be sure to call super().__init__() in derived class!
        self.con = None
        self.name = name
        self.requires = requires
        self._check_reserved_keywords = True
        self.win_pb = ProgressBar(
            lang.sqldriver_init.format_map(LangFormat(name=name)), 100
        )
        self.win_pb.update(lang.sqldriver_connecting, 0)

        # Each database type expects their SQL prepared in a certain way.  Below are
        # defaults for how various elements in the SQL string should be quoted and
        # represented as placeholders. Override these in the derived class as needed to
        # satisfy SQL requirements

        # The placeholder for values in the query string.  This is typically '?' or'%s'
        self.placeholder = placeholder  # override this in derived __init__()

        # These are the quote characters for tables, columns and values.
        # It varies between different databases

        # override this in derived __init__() (defaults to no quotes)
        self.quote_table_char = table_quote
        # override this in derived __init__() (defaults to no quotes)
        self.quote_column_char = column_quote
        # override this in derived __init__() (defaults to single quotes)
        self.quote_value_char = value_quote

    def import_failed(self, exception) -> None:
        popup = Popup()
        requires = ", ".join(self.requires)
        popup.ok(
            lang.import_module_failed_title,
            lang.import_module_failed.format_map(
                LangFormat(name=self.name, requires=requires, exception=exception)
            ),
        )
        exit(0)

    def check_reserved_keywords(self, value: bool) -> None:
        """
        SQLDrivers can check to make sure that field names respect their own reserved
        keywords.  By default, all SQLDrivers will check for their respective keywords.
        You can choose to disable this feature with this method.

        :param value: True to check for reserved keywords in field names, false to skip
            this check
        :return: None
        """
        self._check_reserved_keywords = value

    def connect(self, *args, **kwargs):
        """
        Connect to a database.

        Connect to a database in the connect() method, assigning the connection to
        self.con.

        Implementation varies by database, you may need only one parameter, or
        several depending on how a connection is established with the target database.
        """
        raise NotImplementedError

    def execute(
        self,
        query,
        values=None,
        column_info: ColumnInfo = None,
        auto_commit_rollback: bool = False,
    ):
        """
        Implements the native SQL implementation's execute() command.

        :param query: The query string to execute
        :param values: Values to pass into the query to replace the placeholders
        :param column_info: An optional ColumnInfo object
        :param auto_commit_rollback: Automatically commit or rollback depending on
            whether an exception was handled. Set to False by default.  Set to True to
            have exceptions and commit/rollbacks happen automatically
        :return:
        """
        raise NotImplementedError

    def execute_script(self, script: str, silent: bool = False):
        raise NotImplementedError

    def get_tables(self):
        raise NotImplementedError

    def column_info(self, table):
        raise NotImplementedError

    def pk_column(self, table):
        raise NotImplementedError

    def relationships(self):
        raise NotImplementedError

    # ---------------------------------------------------------------------
    # SHOULD implement
    # based on specifics of the database
    # ---------------------------------------------------------------------
    # This is a generic way to estimate the next primary key to be generated.
    # Note that this is not always a reliable way, as manual inserts which assign a
    # primary key value don't always update the sequencer for the given database.  This
    # is just a default way to "get things working", but the best bet is to override
    # this in the derived class and get the value right from the sequencer.
    def next_pk(self, table: str, pk_column: str) -> int:
        max_pk = self.max_pk(table, pk_column)
        if max_pk is not None:
            return max_pk + 1
        return 1

    def check_keyword(self, keyword: str, key: str = None) -> None:
        """
        Check keyword to see if it is a reserved word.  If it is raise a
        ReservedKeywordError. Checks to see if the database name is in keys and uses the
        database name for the key if it exists, otherwise defaults to 'common' in the
        RESERVED set. Override this with the specific key for the database if needed for
        best results.

        :param keyword: the value to check against reserved words
        :param key: The key in the RESERVED set to check in
        :returns: None
        """
        if not self.check_reserved_keywords:
            return

        if key is None:
            # First try using the name of the driver
            key = self.name.lower() if self.name.lower() in RESERVED else "common"

        if keyword.upper() in RESERVED[key] or keyword.upper in RESERVED["common"]:
            raise ReservedKeywordError(
                f"`{keyword}` is a reserved keyword and cannot be used for table or "
                f"column names."
            )

    # ---------------------------------------------------------------------
    # MAY need to be implemented
    # These default implementations will likely work for most SQL databases.
    # Override any of the following methods as needed.
    # ---------------------------------------------------------------------
    @staticmethod
    def quote(val: str, chars: str):
        if not len(chars):
            return val
        left_quote = chars[0]
        right_quote = chars[1] if len(chars) == 2 else left_quote
        return f"{left_quote}{val}{right_quote}"

    def quote_table(self, table: str):
        return self.quote(table, self.quote_table_char)

    def quote_column(self, column: str):
        return self.quote(column, self.quote_column_char)

    def quote_value(self, value: str):
        return self.quote(value, self.quote_value_char)

    def commit(self):
        self.con.commit()

    def rollback(self):
        self.con.rollback()

    def close(self):
        self.con.close()

    def default_query(self, table):
        table = self.quote_table(table)
        return f"SELECT {table}.* FROM {table}"

    def default_order(self, description_column):
        description_column = self.quote_column(description_column)
        return f" ORDER BY {description_column} ASC"

    def relationship_to_join_clause(self, r_obj: Relationship):
        parent = self.quote_table(r_obj.parent_table)
        child = self.quote_table(r_obj.child_table)
        fk = self.quote_column(r_obj.fk_column)
        pk = self.quote_column(r_obj.pk_column)

        return f"{r_obj.join_type} {parent} ON {child}.{fk}={parent}.{pk}"

    def min_pk(self, table: str, pk_column: str) -> int:
        rows = self.execute(f"SELECT MIN({pk_column}) as min_pk FROM {table}")
        return rows.iloc[0]["min_pk"]

    def max_pk(self, table: str, pk_column: str) -> int:
        rows = self.execute(f"SELECT MAX({pk_column}) as max_pk FROM {table}")
        return rows.iloc[0]["max_pk"]

    def generate_join_clause(self, dataset: DataSet) -> str:
        """
        Automatically generates a join clause from the Relationships that have been set.

        This typically isn't used by end users.

        :returns: A join string to be used in a sqlite3 query
        :rtype: str
        """
        join = ""
        for r in dataset.frm.relationships:
            if dataset.table == r.child_table:
                join += f" {self.relationship_to_join_clause(r)}"
        return join if not dataset.join_clause else dataset.join_clause

    @staticmethod
    def generate_where_clause(dataset: DataSet) -> str:
        """
        Generates a where clause from the Relationships that have been set, as well as
        the DataSet's where clause.

        This is not typically used by end users.

        :returns: A where clause string to be used in a sqlite3 query
        :rtype: str
        """
        where = ""
        for r in dataset.frm.relationships:
            if dataset.table == r.child_table and r.on_update_cascade:
                table = dataset.table
                parent_pk = dataset.frm[r.parent_table].get_current(r.pk_column)

                # Children without cascade-filtering parent aren't displayed
                if not parent_pk:
                    parent_pk = "NULL"

                clause = f" WHERE {table}.{r.fk_column}={str(parent_pk)}"
                if where:
                    clause = clause.replace("WHERE", "AND")
                where += clause

        if not where:
            # There was no where clause from Relationships..
            where = dataset.where_clause
        else:
            # There was an auto-generated portion of the where clause.
            # We will add the table's where clause to it
            where = where + " " + dataset.where_clause.replace("WHERE", "AND")

        return where

    @staticmethod
    def generate_query(
        dataset: DataSet,
        join_clause: bool = True,
        where_clause: bool = True,
        order_clause: bool = True,
    ) -> str:
        """
        Generate a query string using the relationships that have been set.

        :param dataset: A `DataSet` object
        :param join_clause: True to auto-generate `join` clause, False to not
        :param where_clause: True to auto-generate `where` clause, False to not
        :param order_clause: True to auto-generate `order by` clause, False to not
        :returns: a query string for use with sqlite3
        """
        return (
            f"{dataset.query}"
            f' {dataset.join_clause if join_clause else ""}'
            f' {dataset.where_clause if where_clause else ""}'
            f' {dataset.order_clause if order_clause else ""}'
        )

    def delete_record(
        self, dataset: DataSet, cascade=True
    ):  # TODO: get ON DELETE CASCADE from db
        # Get data for query
        table = self.quote_table(dataset.table)
        pk_column = self.quote_column(dataset.pk_column)
        pk = dataset.get_current(dataset.pk_column)

        # Create clauses
        delete_clause = f"DELETE FROM {table} "  # leave a space at end for joining
        where_clause = f"WHERE {table}.{pk_column} = {pk}"

        # Delete child records first!
        if cascade:
            recursion = 0
            result = self.delete_record_recursive(
                dataset, "", where_clause, table, pk_column, recursion
            )

        # Then delete self
        if result == DELETE_RECURSION_LIMIT_ERROR:
            return DELETE_RECURSION_LIMIT_ERROR
        q = delete_clause + where_clause + ";"
        return self.execute(q)

    def delete_record_recursive(
        self, dataset: DataSet, inner_join, where_clause, parent, pk_column, recursion
    ):
        for child in Relationship.get_delete_cascade_tables(dataset.table):
            # Check to make sure we arn't at recursion limit
            recursion += 1  # Increment, since this is a child
            if recursion >= DELETE_CASCADE_RECURSION_LIMIT:
                return DELETE_RECURSION_LIMIT_ERROR

            # Get data for query
            fk_column = self.quote_column(
                Relationship.get_delete_cascade_fk_column(child)
            )
            pk_column = self.quote_column(dataset.frm[child].pk_column)
            child_table = self.quote_table(child)
            select_clause = f"SELECT {child_table}.{pk_column} FROM {child} "
            delete_clause = f"DELETE FROM {child} WHERE {pk_column} IN "

            # Create new inner join and add it to beginning of passed in inner_join
            inner_join_clause = (
                f"INNER JOIN {parent} ON {parent}.{pk_column} = "
                f"{child}.{fk_column} {inner_join}"
            )

            # Call function again to create recursion
            result = self.delete_record_recursive(
                dataset.frm[child],
                inner_join_clause,
                where_clause,
                child,
                self.quote_column(dataset.frm[child].pk_column),
                recursion,
            )

            # Break out of recursive call if at recursion limit
            if result == DELETE_RECURSION_LIMIT_ERROR:
                return DELETE_RECURSION_LIMIT_ERROR

            # Create query and execute
            q = (
                delete_clause
                + "("
                + select_clause
                + inner_join_clause
                + where_clause
                + ");"
            )
            self.execute(q)
            logger.debug(f"Delete query executed: {q}")

            # Reset limit for next Child stack
            recursion = 0
        return None

    def duplicate_record(self, dataset: DataSet, children: bool) -> pd.DataFrame:
        # https://stackoverflow.com/questions/1716320/how-to-insert-duplicate-rows-in-sqlite-with-a-unique-id # fmt: skip # noqa E501
        # This can be done using * syntax without knowing the schema of the table
        # (other than primary key column). The trick is to create a temporary table
        # using the "CREATE TABLE AS" syntax.
        description = self.quote_value(
            f"{lang.duplicate_prepend}"
            f"{dataset.get_description_for_pk(dataset.get_current_pk())}"
        )
        table = self.quote_table(dataset.table)
        tmp_table = self.quote_table(f"temp_{dataset.table}")
        pk_column = self.quote_column(dataset.pk_column)
        description_column = self.quote_column(dataset.description_column)

        # Create tmp table, update pk column in temp and insert into table
        query = [
            f"DROP TABLE IF EXISTS {tmp_table};",
            (
                f"CREATE TEMPORARY TABLE {tmp_table} AS SELECT * FROM {table} WHERE "
                f"{pk_column}={dataset.get_current(dataset.pk_column)};"
            ),
            (
                f"UPDATE {tmp_table} SET {pk_column} = "
                f"{self.next_pk(dataset.table, dataset.pk_column)};"
            ),
            f"UPDATE {tmp_table} SET {description_column} = {description}",
            f"INSERT INTO {table} SELECT * FROM {tmp_table};",
            f"DROP TABLE IF EXISTS {tmp_table};",
        ]
        for q in query:
            res = self.execute(q)
            if res.attrs["exception"]:
                return res

        # Now we save the new pk
        pk = res.attrs["lastrowid"]

        # create list of which children we have duplicated
        child_duplicated = []
        # Next, duplicate the child records!
        if children:
            for _ in dataset.frm.datasets:
                for r in dataset.frm.relationships:
                    if (
                        r.parent_table == dataset.table
                        and r.on_update_cascade
                        and (r.child_table not in child_duplicated)
                    ):
                        child = self.quote_table(r.child_table)
                        tmp_child = self.quote_table(f"temp_{r.child_table}")
                        pk_column = self.quote_column(
                            dataset.frm[r.child_table].pk_column
                        )
                        fk_column = self.quote_column(r.fk_column)

                        # Update children's pk_columns to NULL and set correct parent
                        # PK value.
                        queries = [
                            f"DROP TABLE IF EXISTS {tmp_child};",
                            (
                                f"CREATE TEMPORARY TABLE {tmp_child} AS "
                                f"SELECT * FROM {child} WHERE {fk_column}="
                                f"{dataset.get_current(dataset.pk_column)};"
                            ),
                            # don't next_pk(), because child can be plural.
                            f"UPDATE {tmp_child} SET {pk_column} = NULL;",
                            f"UPDATE {tmp_child} SET {fk_column} = {pk}",
                            f"INSERT INTO {child} SELECT * FROM {tmp_child};",
                            f"DROP TABLE IF EXISTS {tmp_child};",
                        ]
                        for q in queries:
                            res = self.execute(q)
                            if res.exception:
                                return res

                        child_duplicated.append(r.child_table)
        # If we made it here, we can return the pk.  Since the pk was stored earlier,
        # we will just send and empty dataframe. TODO: will this work as expeted still?
        return Result.set(lastrowid=pk)

    def save_record(
        self, dataset: DataSet, changed_row: dict, where_clause: str = None
    ) -> pd.DataFrame:
        pk = dataset.get_current_pk()
        pk_column = dataset.pk_column

        # Remove the pk column and any virtual columns
        changed_row = {
            self.quote_column(k): v
            for k, v in changed_row.items()
            if k != pk_column and k not in dataset.column_info.get_virtual_names()
        }

        # Set empty fields to None
        for k, v in changed_row.items():
            if v == "":  # noqa: PLC1901
                changed_row[k] = None

        # quote appropriately
        table = self.quote_table(dataset.table)
        pk_column = self.quote_column(pk_column)

        # Create the WHERE clause
        if where_clause is None:
            where_clause = f"WHERE {pk_column} = {pk}"

        # Generate an UPDATE query
        query = f"UPDATE {table} SET {', '.join(f'{k}={self.placeholder}' for k in changed_row)} {where_clause};"  # fmt: skip # noqa: E501
        values = list(changed_row.values())

        result = self.execute(query, tuple(values))
        # manually clear the rowid since it is not needed for updated records
        # (we already know the key)
        result.lastrowid = None
        return result

    def insert_record(self, table: str, pk: int, pk_column: str, row: dict):
        # Remove the pk column
        row = {self.quote_column(k): v for k, v in row.items() if k != pk_column}

        # Set empty fields to None
        for k, v in row.items():
            if v == "":  # noqa: PLC1901
                row[k] = None

        # quote appropriately
        table = self.quote_table(table)

        # Remove the primary key column to ensure autoincrement is used!
        query = (
            f"INSERT INTO {table} ({', '.join(key for key in row)}) VALUES "
            f"({','.join(self.placeholder for _ in range(len(row)))}); "
        )
        values = [value for key, value in row.items()]
        return self.execute(query, tuple(values))


# --------------------------------------------------------------------------------------
# SQLITE3 DRIVER
# --------------------------------------------------------------------------------------
class Sqlite(SQLDriver):
    """
    The SQLite driver supports SQLite3 databases.
    """

    def __init__(
        self,
        db_path=None,
        sql_script=None,
        sql_script_encoding: str = "utf-8",
        sqlite3_database=None,
        sql_commands=None,
    ):
        super().__init__(
            name="SQLite",
            requires=["sqlite3"],
            placeholder="?",
            table_quote='"',
            column_quote='"',
        )

        self.import_required_modules()

        new_database = False
        if db_path is not None:
            logger.info(f"Opening database: {db_path}")
            new_database = not os.path.isfile(db_path)
            self.connect(db_path)  # Open our database

        self.imported_database = False
        if sqlite3_database is not None:
            self.con = sqlite3_database
            new_database = False
            self.imported_database = True

        self.win_pb.update(lang.sqldriver_execute, 50)
        self.con.row_factory = sqlite3.Row
        if sql_commands is not None and new_database:
            # run SQL script if the database does not yet exist
            logger.info("Executing sql commands passed in")
            logger.debug(sql_commands)
            self.con.executescript(sql_commands)
            self.con.commit()
        if sql_script is not None and new_database:
            # run SQL script from the file if the database does not yet exist
            logger.info("Executing sql script from file passed in")
            self.execute_script(sql_script, sql_script_encoding)

        self.db_path = db_path
        self.win_pb.close()

    def import_required_modules(self):
        global sqlite3  # noqa PLW0603
        try:
            import sqlite3
        except ModuleNotFoundError as e:
            self.import_failed(e)

    def connect(self, database):
        self.con = sqlite3.connect(database)

    def execute(
        self,
        query,
        values=None,
        silent=False,
        column_info=None,
        auto_commit_rollback: bool = False,
    ) -> pd.DataFrame:
        if not silent:
            logger.info(f"Executing query: {query} {values}")

        cursor = self.con.cursor()
        exception = None
        try:
            cur = cursor.execute(query, values) if values else cursor.execute(query)
        except sqlite3.Error as e:
            exception = e
            logger.warning(
                f"Execute exception: {type(e).__name__}: {e}, using query: {query}"
            )
            if auto_commit_rollback:
                self.rollback()
        else:
            if auto_commit_rollback:
                self.commit()

        try:
            rows = cur.fetchall()
        except:  # noqa E722
            rows = []

        lastrowid = cursor.lastrowid if cursor.lastrowid is not None else None
        return Result.set(
            [dict(row) for row in rows], lastrowid, exception, column_info
        )

    def close(self):
        # Only do cleanup if this is not an imported database
        if not self.imported_database:
            # optimize the database for long-term benefits
            if self.db_path != ":memory:":
                q = "PRAGMA optimize;"
                self.con.execute(q)
            # Close the connection
            self.con.close()

    def get_tables(self):
        q = (
            "SELECT name FROM sqlite_master "
            'WHERE type="table" AND name NOT like "sqlite%";'
        )
        cur = self.execute(q, silent=True)
        return list(cur["name"])

    def column_info(self, table):
        # Return a list of column names
        q = f"PRAGMA table_info({self.quote_table(table)})"
        rows = self.execute(q, silent=True)
        names = []
        col_info = ColumnInfo(self, table)

        for index, row in rows.iterrows():
            name = row["name"]
            names.append(name)
            domain = row["type"]
            notnull = row["notnull"]
            default = row["dflt_value"]
            pk = row["pk"]
            col_info.append(
                Column(
                    name=name, domain=domain, notnull=notnull, default=default, pk=pk
                )
            )

        return col_info

    def pk_column(self, table):
        q = f"PRAGMA table_info({self.quote_table(table)})"
        result = self.execute(q, silent=True)
        return result.loc[result["pk"] == 1, "name"].iloc[0]

    def relationships(self):
        # Return a list of dicts {from_table,to_table,from_column,to_column,requery}
        relationships = []
        tables = self.get_tables()
        for from_table in tables:
            rows = self.execute(
                f"PRAGMA foreign_key_list({self.quote_table(from_table)})", silent=True
            )

            for index, row in rows.iterrows():
                dic = {}
                # Add the relationship if it's in the requery list
                if row["on_update"] == "CASCADE":
                    dic["update_cascade"] = True
                else:
                    dic["update_cascade"] = False
                if row["on_delete"] == "CASCADE":
                    dic["delete_cascade"] = True
                else:
                    dic["delete_cascade"] = False
                dic["from_table"] = from_table
                dic["to_table"] = row["table"]
                dic["from_column"] = row["from"]
                dic["to_column"] = row["to"]
                relationships.append(dic)
        return relationships

    def execute_script(self, script, encoding):
        with open(script, "r", encoding=encoding) as file:
            logger.info(f"Loading script {script} into database.")
            self.con.executescript(file.read())


# --------------------------------------------------------------------------------------
# FLATFILE DRIVER
# --------------------------------------------------------------------------------------
# The CSV driver uses SQlite3 in the background
# to use pysimplesql directly with CSV files
class Flatfile(Sqlite):

    """
    The Flatfile driver adds support for flatfile databases such as CSV files to
    pysimplesql.

    The flatfile data is loaded into an internal SQlite database, where it can be used
    and manipulated like any other database file.  Each timem records are saved, the
    contents of the internal SQlite database are written back out to the file. This
    makes working with flatfile data as easy and consistent as any other database.
    """

    def __init__(
        self,
        file_path: str,
        delimiter: str = ",",
        quotechar: str = '"',
        header_row_num: int = 0,
        table: str = None,
        pk_col: str = None,
    ) -> None:
        """
        Create a new Flatfile driver instance.

        :param file_path: The path to the flatfile
        :param delimiter: The delimiter for the flatfile. Defaults to ','.  Tabs ('\t')
            are another popular option
        :param quotechar: The quoting character specified by the flatfile.
            Defaults to '"'
        :param header_row_num: The row containing the header column names.
            Defaults to 0
        :param table: The name to give this table in pysimplesql. Default is 'Flatfile'
        :param pk_col: The column name that acts as a primary key for the dataset. See
            below how to use this parameter:
           - If no pk_col parameter is supplied, then a generic primary key column named
             'pk' will be generated with AUTO INCREMENT and PRIMARY KEY set. This is a
             virtual column and will not be written back out to the flatfile.
           - If the pk_col parameter is supplied, and it exists in the header row, then
             it will be used as the primary key for the dataset.  If this column does
             not exist in the header row, then a virtual primary key column with this
             name will be created with AUTO INCREMENT and PRIMARY KEY set. As above, the
             virtual primary key column that was created will not be written to the
             flatfile.
        """
        # First up the SQLite driver that we derived from
        super().__init__(":memory:")  # use an in-memory database

        # Change Sqlite Sqldriver init set values to Flatfile-specific
        self.name = "Flatfile"
        self.requires = ["csv,sqlite3"]
        self.placeholder = "?"  # update

        self.import_required_modules()

        self.connect(":memory:")
        self.file_path = file_path
        self.delimiter = delimiter
        self.quotechar = quotechar
        self.header_row_num = header_row_num
        self.pk_col = pk_col if pk_col is not None else "pk"
        self.pk_col_is_virtual = False
        self.table = table if table is not None else "Flatfile"
        self.con.row_factory = sqlite3.Row

        # Store any text up to the header line, so they can be restored
        self.pre_header = []

        # Open the CSV file and read the header row to get column names
        with open(file_path, "r") as f:
            reader = csv.reader(f, delimiter=self.delimiter, quotechar=self.quotechar)
            # skip lines as determined by header_row_num
            for _i in range(self.header_row_num):
                self.pre_header.append(next(reader))

            # Grab the header row information
            self.columns = next(reader)

        if self.pk_col not in self.columns:
            # The pk column was not found, we will make it virutal
            self.columns.insert(0, self.pk_col)
            self.pk_col_is_virtual = True

        # Construct the SQL commands to create the table to represent the flatfile data
        q_cols = ""
        for col in self.columns:
            if col == self.pk_col:
                q_cols += f'{col} {"INTEGER PRIMARY KEY AUTOINCREMENT" if self.pk_col_is_virtual else "PRIMARY KEY"}'  # fmt: skip # noqa: E501
            else:
                q_cols += f"{col} TEXT"

            if col != self.columns[-1]:
                q_cols += ", "

        query = f"CREATE TABLE {self.table} ({q_cols})"
        self.execute(query)

        # Load the CSV data into the table
        with open(self.file_path, "r") as f:
            reader = csv.reader(f, delimiter=self.delimiter, quotechar=self.quotechar)
            # advance to past the header column
            for _i in range(self.header_row_num + 1):
                next(reader)

            # We only want to insert the pk_column if it is not virtual. We will remove
            # it now, as it has already served its purpose to create the table
            if self.pk_col_is_virtual:
                self.columns.remove(self.pk_col)

            query = (
                f'INSERT INTO {self.table} ({", ".join(self.columns)}) VALUES '
                f'({", ".join(["?" for _ in self.columns])})'
            )
            for row in reader:
                self.execute(query, row)

        self.commit()  # commit them all at the end
        self.win_pb.close()

    def import_required_modules(self):
        global csv  # noqa PLW0603
        global sqlite3  # noqa PLW0603
        try:
            import csv
            import sqlite3
        except ModuleNotFoundError as e:
            self.import_failed(e)

    def save_record(
        self, dataset: DataSet, changed_row: dict, where_clause: str = None
    ) -> pd.DataFrame:
        # Have SQlite save this record
        result = super().save_record(dataset, changed_row, where_clause)

        if result.exception is None:
            # No it is safe to write our data back out to the CSV file

            # Update the DataSet object's DataFra,e with the changes, so then
            # the entire DataFrame can be written back to file sequentially
            dataset.rows[dataset.current_index] = changed_row

            # open the CSV file for writing
            with open(self.file_path, "w", newline="\n") as csvfile:
                # create a csv writer object
                writer = csv.writer(
                    csvfile, delimiter=self.delimiter, quotechar=self.quotechar
                )

                # Skip the number of lines defined by header_row_num.
                # Write out the stored pre_header lines
                for line in self.pre_header:
                    writer.writerow(line)

                # write the header row
                writer.writerow(list(self.columns))

                # write the DataFrame out.
                # Use our columns to exclude the possible virtual pk
                rows = []
                for r in dataset.rows:
                    rows.append([r[c] for c in self.columns])

                logger.debug(f"Writing the following data to {self.file_path}")
                logger.debug(rows)
                writer.writerows(rows)

        return result


# --------------------------------------------------------------------------------------
# MYSQL DRIVER
# --------------------------------------------------------------------------------------
class Mysql(SQLDriver):
    """
    The Mysql driver supports MySQL databases.
    """

    def __init__(
        self, host, user, password, database, sql_script=None, sql_commands=None
    ):
        super().__init__(name="MySQL", requires=["mysql-connector-python"])

        self.import_required_modules()

        self.name = "MySQL"  # is this redundant?
        self.host = host
        self.user = user
        self.password = password
        self.database = database
        self.con = self.connect()

        self.win_pb.update(lang.sqldriver_execute, 50)
        if sql_commands is not None:
            # run SQL script if the database does not yet exist
            logger.info("Executing sql commands passed in")
            logger.debug(sql_commands)
            self.con.executescript(sql_commands)
            self.con.commit()
        if sql_script is not None:
            # run SQL script from the file if the database does not yet exist
            logger.info("Executing sql script from file passed in")
            self.execute_script(sql_script)

        self.win_pb.close()

    def import_required_modules(self):
        global mysql  # noqa PLW0603
        try:
            import mysql.connector
        except ModuleNotFoundError as e:
            self.import_failed(e)

    def connect(self, retries=3):
        attempt = 0
        while attempt < retries:
            try:
                con = mysql.connector.connect(
                    host=self.host,
                    user=self.user,
                    password=self.password,
                    database=self.database,
                    # connect_timeout=3,
                )
                return con
            except mysql.connector.Error as e:
                print(f"Failed to connect to database ({attempt + 1}/{retries})")
                print(e)
                attempt += 1
                sleep(1)
        raise Exception("Failed to connect to database")

    def execute(
        self,
        query,
        values=None,
        silent=False,
        column_info=None,
        auto_commit_rollback: bool = False,
    ):
        if not silent:
            logger.info(f"Executing query: {query} {values}")
        cursor = self.con.cursor(dictionary=True)
        exception = None
        try:
            cursor.execute(query, values) if values else cursor.execute(query)
        except mysql.connector.Error as e:
            exception = e.msg
            logger.warning(
                f"Execute exception: {type(e).__name__}: {e}, using query: {query}"
            )
            if auto_commit_rollback:
                self.rollback()
        else:
            if auto_commit_rollback:
                self.commit()

        try:
            rows = cursor.fetchall()
        except:  # noqa E722
            rows = []

        lastrowid = cursor.lastrowid if cursor.lastrowid else None

        return pd.DataFrame(
            [dict(row) for row in rows], lastrowid, exception, column_info
        )

    def get_tables(self):
        query = (
            "SELECT TABLE_NAME FROM information_schema.tables WHERE table_schema = %s"
        )
        rows = self.execute(query, [self.database], silent=True)
        return [row["TABLE_NAME"] for row in rows]

    def column_info(self, table):
        # Return a list of column names
        query = "DESCRIBE {}".format(table)
        rows = self.execute(query, silent=True)
        col_info = ColumnInfo(self, table)

        for row in rows:
            name = row["Field"]
            # Check if the value is a bytes-like object, and decode if necessary
            type_value = (
                row["Type"].decode("utf-8")
                if isinstance(row["Type"], bytes)
                else row["Type"]
            )
            # Capitalize and get rid of the extra information of the row type
            # I.e. varchar(255) becomes VARCHAR
            domain = type_value.split("(")[0].upper()
            notnull = row["Null"] == "NO"
            default = row["Default"]
            pk = row["Key"] == "PRI"
            col_info.append(
                Column(
                    name=name, domain=domain, notnull=notnull, default=default, pk=pk
                )
            )

        return col_info

    def pk_column(self, table):
        query = "SHOW KEYS FROM {} WHERE Key_name = 'PRIMARY'".format(table)
        cur = self.execute(query, silent=True)
        row = cur.fetchone()
        return row["Column_name"] if row else None

    def relationships(self):
        # Return a list of dicts {from_table,to_table,from_column,to_column,requery}
        tables = self.get_tables()
        relationships = []
        for from_table in tables:
            query = (
                "SELECT * FROM information_schema.key_column_usage WHERE "
                "referenced_table_name IS NOT NULL AND table_name = %s"
            )
            rows = self.execute(query, (from_table,), silent=True)

            for row in rows:
                dic = {}
                # Get the constraint information
                on_update, on_delete = self.constraint(row["CONSTRAINT_NAME"])
                if on_update == "CASCADE":
                    dic["update_cascade"] = True
                else:
                    dic["update_cascade"] = False
                if on_delete == "CASCADE":
                    dic["delete_cascade"] = True
                else:
                    dic["delete_cascade"] = False
                dic["from_table"] = row["TABLE_NAME"]
                dic["to_table"] = row["REFERENCED_TABLE_NAME"]
                dic["from_column"] = row["COLUMN_NAME"]
                dic["to_column"] = row["REFERENCED_COLUMN_NAME"]
                relationships.append(dic)
        return relationships

    def execute_script(self, script):
        with open(script, "r"):
            logger.info(f"Loading script {script} into database.")
            # TODO

    # Not required for SQLDriver
    def constraint(self, constraint_name):
        query = (
            "SELECT UPDATE_RULE, DELETE_RULE FROM "
            "INFORMATION_SCHEMA.REFERENTIAL_CONSTRAINTS WHERE CONSTRAINT_NAME = "
            f"'{constraint_name}'"
        )
        rows = self.execute(query, silent=True)
        return rows[0]["UPDATE_RULE"], rows[0]["DELETE_RULE"]


# --------------------------------------------------------------------------------------
# MARIADB DRIVER
# --------------------------------------------------------------------------------------
# MariaDB is a fork of MySQL and backward compatible.  It technically does not need its
# own driver, but that could change in the future, plus having its own named class makes
# it more clear for the end user.
class Mariadb(Mysql):
    """
    The Mariadb driver supports MariaDB databases.
    """

    def __init__(
        self, host, user, password, database, sql_script=None, sql_commands=None
    ):
        super().__init__(host, user, password, database, sql_script, sql_commands)
        self.name = "MariaDB"


# --------------------------------------------------------------------------------------
# POSTGRES DRIVER
# --------------------------------------------------------------------------------------
class Postgres(SQLDriver):
    """
    The Postgres driver supports PostgreSQL databases.
    """

    def __init__(
        self,
        host,
        user,
        password,
        database,
        sql_script=None,
        sql_commands=None,
        sync_sequences=False,
    ):
        super().__init__(
            name="Postgres", requires=["psycopg2", "psycopg2.extras"], table_quote='"'
        )

        self.import_required_modules()

        self.host = host
        self.user = user
        self.password = password
        self.database = database
        self.con = self.connect()

        # experiment to see if I can make a nocase collation
        # query ="CREATE COLLATION NOCASE (provider = icu, locale = 'und-u-ks-level2');"
        # self.execute(query)

        if sync_sequences:
            # synchronize the sequences with the max pk for each table. This is useful
            # if manual records were inserted without calling nextval() to update the
            # sequencer
            q = "SELECT sequence_name FROM information_schema.sequences;"
            sequences = self.execute(q, silent=True)
            for s in sequences:
                seq = s["sequence_name"]

                # get the max pk for this table
                q = (
                    "SELECT column_name, table_name FROM information_schema.columns "
                    f"WHERE column_default LIKE 'nextval(%{seq}%)'"
                )
                rows = self.execute(q, silent=True, auto_commit_rollback=True)
                row = rows.fetchone()
                table = row["table_name"]
                pk_column = row["column_name"]
                max_pk = self.max_pk(table, pk_column)

                # update the sequence
                # TODO: This needs fixed.  pysimplesql_user does have permissions on the
                # sequence, but this still bombs out
                seq = self.quote_table(seq)
                if max_pk > 0:
                    q = f"SELECT setval('{seq}', {max_pk});"
                else:
                    q = f"SELECT setval('{seq}', 1, false);"
                self.execute(q, silent=True, auto_commit_rollback=True)

        self.win_pb.update("executing SQL commands", 50)
        if sql_commands is not None:
            # run SQL script if the database does not yet exist
            logger.info("Executing sql commands passed in")
            logger.debug(sql_commands)
            self.con.executescript(sql_commands)
            self.con.commit()
        if sql_script is not None:
            # run SQL script from the file if the database does not yet exist
            logger.info("Executing sql script from file passed in")
            self.execute_script(sql_script)
        self.win_pb.close()

    def import_required_modules(self):
        global psycopg2  # noqa PLW0603
        try:
            import psycopg2
            import psycopg2.extras
        except ModuleNotFoundError as e:
            self.import_failed(e)

    def connect(self, retries=3):
        attempt = 0
        while attempt < retries:
            try:
                con = psycopg2.connect(
                    host=self.host,
                    user=self.user,
                    password=self.password,
                    database=self.database,
                    # connect_timeout=3,
                )
                return con
            except psycopg2.Error as e:
                print(f"Failed to connect to database ({attempt + 1}/{retries})")
                print(e)
                attempt += 1
                sleep(1)
        raise Exception("Failed to connect to database")

    def execute(
        self,
        query: str,
        values=None,
        silent=False,
        column_info=None,
        auto_commit_rollback: bool = False,
    ):
        if not silent:
            logger.info(f"Executing query: {query} {values}")
        cursor = self.con.cursor(cursor_factory=psycopg2.extras.RealDictCursor)
        exception = None
        try:
            cursor.execute(query, values) if values else cursor.execute(query)
        except psycopg2.Error as e:
            exception = e
            logger.warning(
                f"Execute exception: {type(e).__name__}: {e}, using query: {query}"
            )
            if auto_commit_rollback:
                self.rollback()
        else:
            if auto_commit_rollback:
                self.commit()

        try:
            rows = cursor.fetchall()
        except psycopg2.ProgrammingError:
            rows = []

        # In Postgres, the cursor does not return a lastrowid.  We will not set it here,
        # we will instead set it in save_records() due to the RETURNING stement of the
        # query
        return pd.DataFrame(
            [dict(row) for row in rows], exception=exception, column_info=column_info
        )

    def get_tables(self):
        query = (
            "SELECT table_name FROM information_schema.tables WHERE "
            "table_schema='public' AND table_type='BASE TABLE';"
        )
        # query = "SELECT tablename FROM pg_tables WHERE table_schema='public'"
        rows = self.execute(query, silent=True)
        return [row["table_name"] for row in rows]

    def column_info(self, table: str) -> ColumnInfo:
        # Return a list of column names
        query = f"SELECT * FROM information_schema.columns WHERE table_name = '{table}'"
        rows = self.execute(query, silent=True)

        col_info = ColumnInfo(self, table)
        pk_column = self.pk_column(table)
        for row in rows:
            name = row["column_name"]
            domain = row["data_type"].upper()
            notnull = row["is_nullable"] != "YES"
            default = row["column_default"]
            # Fix the default value by removing the datatype that is appended to the end
            if default is not None and "::" in default:
                default = default[: default.index("::")]

            pk = name == pk_column
            col_info.append(
                Column(
                    name=name, domain=domain, notnull=notnull, default=default, pk=pk
                )
            )

        return col_info

    def pk_column(self, table):
        query = (
            "SELECT column_name FROM information_schema.table_constraints tc JOIN "
            "information_schema.key_column_usage kcu ON tc.constraint_name = "
            "kcu.constraint_name WHERE tc.constraint_type = 'PRIMARY KEY' AND "
            f"tc.table_name = '{table}' "
        )
        cur = self.execute(query, silent=True)
        row = cur.fetchone()
        return row["column_name"] if row else None

    def relationships(self):
        # Return a list of dicts {from_table,to_table,from_column,to_column,requery}
        tables = self.get_tables()
        relationships = []
        for from_table in tables:
            query = (
                "SELECT conname, conrelid::regclass, confrelid::regclass, "
                "confupdtype, confdeltype, a1.attname AS column_name, a2.attname "
                "AS referenced_column_name "
                "FROM pg_constraint "
                "JOIN pg_attribute AS a1 ON conrelid = a1.attrelid AND "
                "a1.attnum = ANY(conkey) "
                "JOIN pg_attribute AS a2 ON confrelid = a2.attrelid AND "
                "a2.attnum = ANY(confkey) "
                f"WHERE confrelid = '\"{from_table}\"'::regclass AND contype = 'f'"
            )

            rows = self.execute(query, (from_table,), silent=True)

            for row in rows:
                dic = {}
                # Get the constraint information
                # constraint = self.constraint(row['conname'])
                if row["confupdtype"] == "c":
                    dic["update_cascade"] = True
                else:
                    dic["update_cascade"] = False
                if row["confdeltype"] == "c":
                    dic["delete_cascade"] = True
                else:
                    dic["delete_cascade"] = False
                dic["from_table"] = row["conrelid"].strip('"')
                dic["to_table"] = row["confrelid"].strip('"')
                dic["from_column"] = row["column_name"]
                dic["to_column"] = row["referenced_column_name"]
                relationships.append(dic)
        return relationships

    def min_pk(self, table: str, pk_column: str) -> int:
        table = self.quote_table(table)
        pk_column = self.quote_column(pk_column)
        rows = self.execute(
            f"SELECT COALESCE(MIN({pk_column}), 0) AS min_pk FROM {table};", silent=True
        )
        return rows.fetchone()["min_pk"]

    def max_pk(self, table: str, pk_column: str) -> int:
        table = self.quote_table(table)
        pk_column = self.quote_column(pk_column)
        rows = self.execute(
            f"SELECT COALESCE(MAX({pk_column}), 0) AS max_pk FROM {table};", silent=True
        )
        return rows.fetchone()["max_pk"]

    def next_pk(self, table: str, pk_column: str) -> int:
        # Working with case-sensitive tables is painful in Postgres.  First, the
        # sequence must be quoted in a manner similar to tables, then the quoted
        # sequence name has to be also surrounded in single quotes to be treated
        # literally and prevent folding of the casing.
        seq = f"{table}_{pk_column}_seq"  # build the default sequence name
        seq = self.quote_table(seq)  # quote it like a table

        # wrap the quoted string in singe quotes.  Phew!
        q = f"SELECT nextval('{seq}') LIMIT 1;"
        rows = self.execute(q, silent=True)
        return rows.fetchone()["nextval"]

    def insert_record(self, table: str, pk: int, pk_column: str, row: dict):
        # insert_record() for Postgres is a little different from the rest. Instead of
        # relying on an autoincrement, we first already "reserved" a primary key
        # earlier, so we will use it directly quote appropriately
        table = self.quote_table(table)

        # Remove the primary key column to ensure autoincrement is used!
        query = (
            f"INSERT INTO {table} ({', '.join(key for key in row)}) VALUES "
            f"({','.join('%s' for _ in range(len(row)))}); "
        )
        values = [value for key, value in row.items()]
        result = self.execute(query, tuple(values))

        result.lastid = pk
        return result

    def execute_script(self, script):
        pass


# --------------------------------------------------------------------------------------
# MS SQLSERVER DRIVER
# --------------------------------------------------------------------------------------
class Sqlserver(SQLDriver):
    """
    The Sqlserver driver supports Microsoft SQL Server databases.
    """

    def __init__(
        self, host, user, password, database, sql_script=None, sql_commands=None
    ):
        super().__init__(
            name="Sqlserver", requires=["pyodbc"], table_quote="[]", placeholder="?"
        )

        self.import_required_modules()

        self.name = "Sqlserver"  # is this redundant?
        self.host = host
        self.user = user
        self.password = password
        self.database = database
        self.con = self.connect()

        if sql_commands is not None:
            # run SQL script if the database does not yet exist
            logger.info("Executing sql commands passed in")
            logger.debug(sql_commands)
            self.con.executescript(sql_commands)
            self.con.commit()
        if sql_script is not None:
            # run SQL script from the file if the database does not yet exist
            logger.info("Executing sql script from file passed in")
            self.execute_script(sql_script)

        self.win_pb.close()

    def import_required_modules(self):
        global pyodbc  # noqa PLW0603
        try:
            import pyodbc
        except ModuleNotFoundError as e:
            self.import_failed(e)

    def connect(self, retries=3, timeout=3):
        attempt = 0
        while attempt < retries:
            try:
                con = pyodbc.connect(
                    f"DRIVER={{ODBC Driver 17 for SQL Server}};"
                    f"SERVER={self.host};"
                    f"DATABASE={self.database};"
                    f"UID={self.user};"
                    f"PWD={self.password}",
                    timeout=timeout,
                )
                return con
            except pyodbc.Error as e:
                print(f"Failed to connect to database ({attempt + 1}/{retries})")
                print(e)
                attempt += 1
                sleep(1)
        raise Exception("Failed to connect to database")

    def execute(
        self,
        query,
        values=None,
        silent=False,
        column_info=None,
        auto_commit_rollback: bool = False,
    ):
        if not silent:
            logger.info(f"Executing query: {query} {values}")
        cursor = self.con.cursor()
        exception = None
        try:
            cursor.execute(query, values) if values else cursor.execute(query)
        except pyodbc.Error as e:
            exception = e
            logger.warning(
                f"Execute exception: {type(e).__name__}: {e}, using query: {query}"
            )
            if auto_commit_rollback:
                self.rollback()
        else:
            if auto_commit_rollback:
                self.commit()

        try:
            rows = cursor.fetchall()
        except:  # noqa E722
            rows = []

        lastrowid = cursor.rowcount if cursor.rowcount else None

        return pd.DataFrame(
            [
                dict(zip([column[0] for column in cursor.description], row))
                for row in rows
            ],
            lastrowid,
            exception,
            column_info,
        )

    def get_tables(self):
        query = (
            "SELECT table_name FROM information_schema.tables WHERE table_catalog = ?"
        )
        rows = self.execute(query, [self.database], silent=True)
        return [row["table_name"] for row in rows]

    def column_info(self, table):
        # Return a list of column names
        query = "SELECT * FROM INFORMATION_SCHEMA.COLUMNS WHERE TABLE_NAME = ?"
        rows = self.execute(query, [table], silent=True)
        col_info = ColumnInfo(self, table)

        # Get the primary key column(s)
        pk_columns = []
        pk_query = """
            SELECT COLUMN_NAME
            FROM INFORMATION_SCHEMA.KEY_COLUMN_USAGE
            WHERE TABLE_NAME = ?
        """
        pk_rows = self.execute(pk_query, [table], silent=True)
        for pk_row in pk_rows:
            pk_columns.append(pk_row["COLUMN_NAME"])

        for row in rows:
            name = row["COLUMN_NAME"]
            domain = row["DATA_TYPE"].upper()
            notnull = row["IS_NULLABLE"] == "NO"
            default = row["COLUMN_DEFAULT"]
            pk = name in pk_columns
            col_info.append(
                Column(
                    name=name, domain=domain, notnull=notnull, default=default, pk=pk
                )
            )

        return col_info

    def relationships(self):
        # Return a list of dicts {from_table,to_table,from_column,to_column,requery}
        tables = self.get_tables()
        relationships = []
        for from_table in tables:
            query = (
                "SELECT "
                "   OBJECT_NAME(f.parent_object_id) AS from_table, "
                "   OBJECT_NAME(f.referenced_object_id) AS to_table, "
                "   COL_NAME(fc.parent_object_id, fc.parent_column_id) AS from_column, "
                "   COL_NAME(fc.referenced_object_id, fc.referenced_column_id) AS to_column, "  # noqa: E501
                "   f.update_referential_action_desc AS update_cascade, "
                "   f.delete_referential_action_desc AS delete_cascade "
                "FROM "
                "   sys.foreign_keys AS f "
                "   INNER JOIN sys.foreign_key_columns AS fc "
                "       ON f.object_id = fc.constraint_object_id "
                "WHERE "
                f"   OBJECT_NAME(f.parent_object_id) = '{from_table}'"
            )

            rows = self.execute(query, silent=True)

            for row in rows:
                dic = {}
                dic["from_table"] = row["from_table"]
                dic["to_table"] = row["to_table"]
                dic["from_column"] = row["from_column"]
                dic["to_column"] = row["to_column"]
                dic["update_cascade"] = row["update_cascade"] == "CASCADE"
                dic["delete_cascade"] = row["delete_cascade"] == "CASCADE"
                relationships.append(dic)
        return relationships

    def pk_column(self, table):
        query = (
            "SELECT "
            "   COLUMN_NAME "
            "FROM "
            "   INFORMATION_SCHEMA.KEY_COLUMN_USAGE "
            "WHERE "
            f"   TABLE_NAME = '{table}' "
            "   AND CONSTRAINT_NAME LIKE 'PK%'"
        )

        rows = self.execute(query, silent=True)

        if rows:
            return rows[0]["COLUMN_NAME"]
        return None


# --------------------------------------------------------------------------------------
# MS ACCESS DRIVER
# --------------------------------------------------------------------------------------
class MSAccess(SQLDriver):
    """
    The MSAccess driver supports Microsoft Access databases.
    Note that only database interactions are supported, including stored Queries, but
    not operations dealing with Forms, Reports, etc.
    """

    def __init__(self, database_file):
        super().__init__(
            name="MSAccess", requires=["Jype1"], table_quote="[]", placeholder="?"
        )

        self.import_required_modules()

        self.database_file = database_file
        self.con = self.connect()

    import os
    import sys

    def import_required_modules(self):
        global jpype  # noqa PLW0603
        try:
            import jpype  # pip install JPype1
        except ModuleNotFoundError as e:
            self.import_failed(e)

    def connect(self):
        # Get the path to the 'lib' folder
        current_path = os.path.dirname(os.path.abspath(__file__))
        lib_path = os.path.join(current_path, "lib", "UCanAccess-5.0.1.bin")

        jars = [
            "ucanaccess-5.0.1.jar",
            os.path.join("lib", "commons-lang3-3.8.1.jar"),
            os.path.join("lib", "commons-logging-1.2.jar"),
            os.path.join("lib", "hsqldb-2.5.0.jar"),
            os.path.join("lib", "jackcess-3.0.1.jar"),
            os.path.join("loader", "ucanload.jar"),
        ]
        classpath = os.pathsep.join([os.path.join(lib_path, jar) for jar in jars])

        if not jpype.isJVMStarted():
            jpype.startJVM(
                jpype.getDefaultJVMPath(), "-ea", f"-Djava.class.path={classpath}"
            )

        driver_manager = jpype.JPackage("java").sql.DriverManager
        con_str = f"jdbc:ucanaccess://{self.database_file}"
        return driver_manager.getConnection(con_str)

    def execute(
        self,
        query,
        values=None,
        silent=False,
        column_info=None,
        auto_commit_rollback: bool = False,
    ):
        if not silent:
            logger.info(f"Executing query: {query} {values}")

        exception = None
        has_result_set = False
        try:
            if values:
                stmt = self.con.prepareStatement(query)
                for index, value in enumerate(values, start=1):
                    stmt.setObject(index, value)
                has_result_set = stmt.execute()
            else:
                stmt = self.con.createStatement()
                has_result_set = stmt.execute(query)
        except Exception as e:  # noqa: BLE001
            exception = e
            if not silent:
                logger.warning(
                    f"Execute exception: {type(e).__name__}: {e}, using query: {query}"
                )
            if auto_commit_rollback:
                self.rollback()

        if has_result_set:
            rs = stmt.getResultSet()
            metadata = rs.getMetaData()
            column_count = metadata.getColumnCount()
            rows = []

            while rs.next():
                row = {}
                for i in range(1, column_count + 1):
                    column_name = str(metadata.getColumnName(i))
                    value = rs.getObject(i)

                    if isinstance(value, jpype.JPackage("java").lang.String):
                        value = str(value)
                    elif isinstance(value, jpype.JPackage("java").lang.Integer):
                        value = int(value)
                    elif isinstance(value, jpype.JPackage("java").math.BigDecimal):
                        value = float(value.doubleValue())
                    elif isinstance(value, jpype.JPackage("java").lang.Double):
                        value = float(value)
                    if isinstance(value, jpype.JPackage("java").sql.Timestamp):
                        timestamp_str = value.toInstant().toString()[:-1]
                        if "." in timestamp_str:
                            timestamp_format = "%Y-%m-%dT%H:%M:%S.%f"
                        else:
                            timestamp_format = "%Y-%m-%dT%H:%M:%S"
                        dt_value = datetime.strptime(timestamp_str, timestamp_format)
                        value = dt_value.strftime("%Y-%m-%d")
                    elif isinstance(value, jpype.JPackage("java").sql.Date):
                        date_str = value.toString()
                        date_format = "%Y-%m-%d"
                        value = datetime.strptime(date_str, date_format).date()
                    elif isinstance(value, jpype.JPackage("java").sql.Time):
                        time_str = value.toString()
                        time_format = "%H:%M:%S"
                        value = datetime.strptime(time_str, time_format).time()
                    elif value is not None:
                        value = value
                    # TODO: More conversions?

                    row[column_name] = value
                rows.append(row)

            return pd.DataFrame(rows, None, exception, column_info)

        affected_rows = stmt.getUpdateCount()
        return pd.DataFrame([], affected_rows, exception, column_info)

    def column_info(self, table):
        meta_data = self.con.getMetaData()
        rs = meta_data.getColumns(None, None, table, None)

        col_info = ColumnInfo(self, table)
        pk_columns = [self.pk_column(table)]

        while rs.next():
            name = str(rs.getString("COLUMN_NAME"))
            domain = str(rs.getString("TYPE_NAME")).upper()
            notnull = str(rs.getString("IS_NULLABLE")) == "NO"
            default = str(rs.getString("COLUMN_DEF"))
            pk = name in pk_columns

            col_info.append(
                Column(
                    name=name, domain=domain, notnull=notnull, default=default, pk=pk
                )
            )

        return col_info

    def pk_column(self, table):
        meta_data = self.con.getMetaData()
        rs = meta_data.getPrimaryKeys(None, None, table)
        if rs.next():
            return str(rs.getString("COLUMN_NAME"))
        return None

    def get_tables(self):
        metadata = self.con.getMetaData()
        rs = metadata.getTables(None, None, "%", ["TABLE"])
        tables = []

        while rs.next():
            tables.append(str(rs.getString("TABLE_NAME")))

        return tables

    def relationships(self):
        # Get the mapping of uppercase table and column names to their original case
        table_mapping = {table.upper(): table for table in self.get_tables()}
        column_mappings = {
            table: {col.name.upper(): col.name for col in self.column_info(table)}
            for table in self.get_tables()
        }

        query = (
            "SELECT"
            "  fk.TABLE_NAME AS from_table,"
            "  pk.TABLE_NAME AS to_table,"
            "  fk.COLUMN_NAME AS from_column,"
            "  pk.COLUMN_NAME AS to_column,"
            "  rc.UPDATE_RULE AS on_update,"
            "  rc.DELETE_RULE AS on_delete"
            " FROM"
            "  INFORMATION_SCHEMA.REFERENTIAL_CONSTRAINTS rc"
            " INNER JOIN INFORMATION_SCHEMA.KEY_COLUMN_USAGE fk"
            " ON rc.CONSTRAINT_NAME = fk.CONSTRAINT_NAME"
            " INNER JOIN INFORMATION_SCHEMA.KEY_COLUMN_USAGE pk"
            " ON rc.UNIQUE_CONSTRAINT_NAME = pk.CONSTRAINT_NAME"
            " WHERE"
            "  fk.TABLE_SCHEMA = 'PUBLIC'"
            " AND pk.TABLE_SCHEMA = 'PUBLIC'"
        )

        stmt = self.con.createStatement()
        rs = stmt.executeQuery(query)
        relationships = []

        while rs.next():
            from_table_upper = str(rs.getString("from_table"))
            to_table_upper = str(rs.getString("to_table"))
            from_column_upper = str(rs.getString("from_column"))
            to_column_upper = str(rs.getString("to_column"))

            dic = {}
            dic["from_table"] = table_mapping[from_table_upper]
            dic["to_table"] = table_mapping[to_table_upper]
            dic["from_column"] = column_mappings[dic["from_table"]][from_column_upper]
            dic["to_column"] = column_mappings[dic["to_table"]][to_column_upper]
            dic["update_cascade"] = rs.getString("on_update") == "CASCADE"
            dic["delete_cascade"] = rs.getString("on_delete") == "CASCADE"
            relationships.append(dic)

        return relationships

    def max_pk(self, table: str, pk_column: str) -> int:
        rows = self.execute(f"SELECT MAX({pk_column}) as max_pk FROM {table}")
        return rows.fetchone()["MAX_PK"]  # returned as upper case

    def _get_column_definitions(self, table_name):
        # Creates a comma separated list of column names and types to be used in a
        # CREATE TABLE statement
        columns = self.column_info(table_name)
        cols = ""
        for c in columns:
            cols += f"{c['name']} {c['domain']}, "
        cols = cols[:-2]
        return cols

    def duplicate_record(self, dataset: DataSet, children: bool) -> pd.DataFrame:
        # https://stackoverflow.com/questions/1716320/how-to-insert-duplicate-rows-in-sqlite-with-a-unique-id # fmt: skip # noqa: E501
        # This can be done using * syntax without knowing the schema of the table
        # (other than primary key column). The trick is to create a temporary table
        # using the "CREATE TABLE AS" syntax.

        description = self.quote_value(
            f"{lang.duplicate_prepend}"
            f"{dataset.get_description_for_pk(dataset.get_current_pk())}"
        )
        table = self.quote_table(dataset.table)
        tmp_table = self.quote_table(f"temp_{dataset.table}")
        pk_column = self.quote_column(dataset.pk_column)
        description_column = self.quote_column(dataset.description_column)

        # Create tmp table, update pk column in temp and insert into table
        f"WHERE {pk_column}={dataset.get_current(dataset.pk_column)}"
        query = [
            f"DROP TABLE IF EXISTS [{tmp_table}];",
            f"CREATE TABLE [{tmp_table}] ({self._get_column_definitions(table)});",
            (
                f"INSERT INTO [{tmp_table}] (SELECT * FROM [{table}] "
                f"WHERE {pk_column}={dataset.get_current(dataset.pk_column)});"
            ),
            (
                f"UPDATE [{tmp_table}] SET {pk_column} = "
                f"{self.next_pk(dataset.table, dataset.pk_column)};"
            ),
            f"UPDATE [{tmp_table}] SET {description_column} = {description}",
            f"INSERT INTO [{table}] SELECT * FROM [{tmp_table}];",
            f"DROP TABLE IF EXISTS [{tmp_table}]",
        ]
        for q in query:
            res = self.execute(q)
            if res.exception:
                return res

        # Now we save the new pk
        pk = res.lastrowid

        # create list of which children we have duplicated
        child_duplicated = []
        # Next, duplicate the child records!
        if children:
            for _ in dataset.frm.datasets:
                for r in dataset.frm.relationships:
                    if (
                        r.parent_table == dataset.table
                        and r.on_update_cascade
                        and (r.child_table not in child_duplicated)
                    ):
                        child = self.quote_table(r.child_table)
                        tmp_child = self.quote_table(f"temp_{r.child_table}")
                        pk_column = self.quote_column(
                            dataset.frm[r.child_table].pk_column
                        )
                        fk_column = self.quote_column(r.fk_column)

                        # Update children's pk_columns to NULL and set correct parent
                        # PK value.
                        queries = [
                            f"DROP TABLE IF EXISTS [{tmp_child}]",
                            (
                                f"CREATE TABLE [{tmp_table}] "
                                f"({self._get_column_definitions(table)});"
                            ),
                            (
                                f"INSERT INTO [{tmp_table}] (SELECT * FROM [{table}] "
                                f"WHERE {pk_column}="
                                f"{dataset.get_current(dataset.pk_column)});"
                            ),
                            # don't next_pk(), because child can be plural.
                            f"UPDATE [{tmp_child}] SET {pk_column} = NULL;",
                            f"UPDATE [{tmp_child}] SET {fk_column} = {pk}",
                            f"INSERT INTO [{child}] SELECT * FROM [{tmp_child}];",
                            f"DROP TABLE IF EXISTS [{tmp_child}]",
                        ]
                        for q in queries:
                            res = self.execute(q)
                            if res.exception:
                                return res

                        child_duplicated.append(r.child_table)
        # If we made it here, we can return the pk.  Since the pk was stored earlier,
        # we will just send and empty DataFrame
        return Result.set(lastrowid=pk)


# --------------------------
# TYPEDDICTS AND TYPEALIASES
# --------------------------
class Driver:
    """
    The `Driver` class allows for easy driver creation. It is a simple wrapper around
    the various `SQLDriver` classes.
    """

    sqlite: callable = Sqlite
    flatfile: callable = Flatfile
    mysql: callable = Mysql
    mariadb: callable = Mariadb
    postgres: callable = Postgres
    sqlserver: callable = Sqlserver
    msaccess: callable = MSAccess


SaveResultsDict = Dict[str, int]
CallbacksDict = Dict[str, Callable[[Form, sg.Window], Union[None, bool]]]
PromptSaveValue = (
    int  # Union[PROMPT_SAVE_PROCEED, PROMPT_SAVE_DISCARDED, PROMPT_SAVE_NONE]
)


class SimpleTransform(TypedDict):
    decode: Dict[str, Callable[[str, str], None]]
    encode: Dict[str, Callable[[str, str], None]]


SimpleTransformsDict = Dict[str, SimpleTransform]


# ======================================================================================
# ALIASES
# ======================================================================================
languagepack = lang
Database = Form
Table = DataSet
record = field  # for reverse capability