Make ExtensionDtype.construct_array_type a regular method #36126
Labels
ExtensionArray
Extending pandas with custom dtypes or arrays.
Refactor
Internal refactoring of code
Strings
String extension data type and string data
Comments
TomAugspurger
added
API Design
Strings
|
+1 |
TomAugspurger
added a commit
to TomAugspurger/pandas
that referenced
this issue
Sep 5, 2020
This allows a single dtype to support multiple array classes.
For arrow-backed strings, we'll likely want a separate array class
for ease of implementation, clarity. But we'll have a parametrized
dtype.
```python
class StringDtype:
def __init__(self, storage="python"):
self.storage = storage
def construct_array_type(self): # regular method
if self.storage == "python":
return StringArray
else:
return ArrowStringArray
```
Closes pandas-dev#36126
4 tasks
|
removing milestone |
|
This has been done, not sure when. Closing. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Currently ExtensionDtype.consruct_array_type is a classmethod. This makes it hard to use a single dtype with multiple array classes where the array type depends on some parameter of the dtype.
I'd like to change construct_array_type to be a regular method rather than a classmethod. AFAICT, this is fine for pandas since everywhere we use construct_array_type we have an instance rather than a type. It should be fine for 3rd party arrays to continue to use classmethods since you can call a classmethod on an instance. The API breaking change would be for 3rd-party code relying on
ExtensionDtype.construct_array_type()working (i.e. calling it on the class).My motivation is for allowing a parametrized
StringDtypethat can be used for either the current StringArray and an Arrow-backed StringArray.The text was updated successfully, but these errors were encountered: