Modernize IO using only the API

[bashtage]

This is an attempt to modernize IO using only the API functions. Just a start

• read_pickle(filepath_or_buffer[, ...])
• DataFrame.to_pickle(path[, compression, ...])
• read_table(filepath_or_buffer[, sep, ...])
• read_csv(filepath_or_buffer[, sep, ...])
• DataFrame.to_csv([path_or_buf, sep, na_rep, ...])
• read_fwf(filepath_or_buffer[, colspecs, ...])
• read_clipboard([sep])
• DataFrame.to_clipboard([excel, sep])
• read_excel(io[, sheet_name, header, names, ...])
• DataFrame.to_excel(excel_writer[, ...])
• ExcelFile.parse([sheet_name, header, names, ...])
• Styler.to_excel(excel_writer[, sheet_name, ...])
• ExcelWriter(path[, engine, date_format, ...])
• read_json([path_or_buf, orient, typ, dtype, ...])
• json_normalize(data[, record_path, meta, ...])
• DataFrame.to_json([path_or_buf, orient, ...])
• build_table_schema(data[, index, ...])
• read_html(io[, match, flavor, header, ...])
• DataFrame.to_html([buf, columns, col_space, ...])
• Styler.to_html([buf, table_uuid, ...])
• read_xml(path_or_buffer[, xpath, ...])
• DataFrame.to_xml([path_or_buffer, index, ...])
• DataFrame.to_latex([buf, columns, ...])
• Styler.to_latex([buf, column_format, ...])
• HDFStore: PyTables (HDF5)
• read_hdf(path_or_buf[, key, mode, errors, ...])
• HDFStore.put(key, value[, format, index, ...])
• HDFStore.append(key, value[, format, axes, ...])
• HDFStore.get(key)
• HDFStore.select(key[, where, start, stop, ...])
• HDFStore.info()
• HDFStore.keys([include])
• HDFStore.groups()
• HDFStore.walk([where])
• read_feather(path[, columns, use_threads, ...])
• DataFrame.to_feather(path, **kwargs)
• read_parquet(path[, engine, columns, ...])
• DataFrame.to_parquet([path, engine, ...])
• read_orc(path[, columns])
• read_sas(filepath_or_buffer[, format, ...])
• read_spss(path[, usecols, convert_categoricals])
• read_sql_table(table_name, con[, schema, ...])
• read_sql_query(sql, con[, index_col, ...])
• read_sql(sql, con[, index_col, ...])
• DataFrame.to_sql(name, con[, schema, ...])
• read_gbq(query[, project_id, index_col, ...])
• read_stata(filepath_or_buffer[, ...])
• DataFrame.to_stata(path[, convert_dates, ...])
• StataReader.data_label
• StataReader.value_labels()
• StataReader.variable_labels()
• StataWriter.write_file()

[twoertwein]

I wouldn't mind removing those. Might be good to be in sync with @phofl's PR pandas-dev/pandas#46308

[twoertwein]

I think object might be better here than Any. I think the rule of thumb is use object if it can literally be any object and use Any when it is too complex to type.

[bashtage]

Looking at typeshed they use Any both ways for dump and load.

https://github.com/python/typeshed/blob/d4287a7f08305d95d24d3e0487941ec9bb35f4ae/stdlib/pickle.pyi#L121-L128

[bashtage]

Seems object can cause problems with type checking

https://stackoverflow.com/a/39817126/2551705

although this is probably more to do with loading than saving.

[Dr-Irv]

Can you confirm that read_csv() and read_table() are the same as what was previously in parsers.pyi ?

[twoertwein]

@bashtage might be easier to break it into multiple PRs: easier to review and probably also easier for you (less rebasing since the individual PRs will be closed more quickly)