Deprecate position argument to Axes.apply_aspect() #23629
Conversation
35a21f9 to
d0584cd
Compare
|
I think if you search on stackoverflow there are quite a few answers that suggest calling apply_aspect to get the position of things after the proper aspect ratio is applied without forcing a full draw. |
|
Well, maybe only a couple: https://stackoverflow.com/search?q=%5Bmatplotlib%5D+apply_aspect |
From these:
Anyway, let's be super-safe and only deprecate the public position parameter for now. - No end user should manipulate the position through |
d0584cd to
9a68b94
Compare
9a68b94 to
bdcaa9c
Compare
|
Please update the commit message as well for that. |
7784ff2 to
38fbaad
Compare
There was a problem hiding this comment.
I've found a use of apply_aspect in the wild that this rename would likely break without the deprecation triggering
https://github.com/AudioSceneDescriptionFormat/splines/blob/master/doc/rotation/helper.py#L275
I'm not entirely sure what they are attempting to do there.
| @@ -0,0 +1,3 @@ | |||
| The parameter *position* in ``Axes.apply_aspect()`` | |||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |||
| ... is deprecated. It is considered internal API and now public use case is known. | |||
There was a problem hiding this comment.
| ... is deprecated. It is considered internal API and now public use case is known. | |
| ... is deprecated. It is considered internal API and no public use case is known. |
lib/matplotlib/axes/_base.py
Outdated
| position : None or .Bbox | ||
|
|
||
| If not ``None``, this defines the position of the |
There was a problem hiding this comment.
| position : None or .Bbox | |
| If not ``None``, this defines the position of the | |
| position : None or .Bbox | |
| If not ``None``, this defines the position of the |
|
One use of And one that passes a position: https://github.com/astropy/astropy/blob/main/astropy/visualization/wcsaxes/core.py#L444-L452 |
|
Deferred to 3.7. Let’s not rush this into 3.6. This needs a more thorough investigation. |
38fbaad to
57c4ddf
Compare
|
Digging through the history my understanding of the reason for the position parameter is the following: It is somewhat surprising that "_apply_aspect" takes an optional position as input, which seems more functionality than what the name suggests. The parameter was introduced in 66290aae. Generally, applying an aspect will modify the size and position of an Axes. The fact that position is updated anyway was used to funnel additional position information from a layoutinto the already existing positioning code. IMHO:
I see two ways forward:
I'm inclined to go with 1.
has meanwhile been changed to a draw. 👍 |
Superseeds matplotlib#23629. It is somewhat surprising that "_apply_aspect" takes an optional position as input, which seems more functionality than what the name suggests. The parameter was introduced in 66290aa. Generally, applying an aspect will modify the size and position of an Axes. The fact that position is updated anyway was used to funnel additional position information from a layout into the already existing positioning code. Deprecating and removing the parameter would be a medium compatibility hassle. Therefore, I chose only to document the status quo and its intention, so that users are less surprised. See matplotlib#23629 (comment)
Superseeds matplotlib#23629. It is somewhat surprising that "_apply_aspect" takes an optional position as input, which seems more functionality than what the name suggests. The parameter was introduced in 66290aa. Generally, applying an aspect will modify the size and position of an Axes. The fact that position is updated anyway was used to funnel additional position information from a layout into the already existing positioning code. Deprecating and removing the parameter would be a medium compatibility hassle. Therefore, I chose only to document the status quo and its intention, so that users are less surprised. See matplotlib#23629 (comment)
Superseeds matplotlib#23629. It is somewhat surprising that "_apply_aspect" takes an optional position as input, which seems more functionality than what the name suggests. The parameter was introduced in 66290aa. Generally, applying an aspect will modify the size and position of an Axes. The fact that position is updated anyway was used to funnel additional position information from a layout into the already existing positioning code. Deprecating and removing the parameter would be a medium compatibility hassle. Therefore, I chose only to document the status quo and its intention, so that users are less surprised. See matplotlib#23629 (comment)
Superseeds matplotlib#23629. It is somewhat surprising that "_apply_aspect" takes an optional position as input, which seems more functionality than what the name suggests. The parameter was introduced in 66290aa. Generally, applying an aspect will modify the size and position of an Axes. The fact that position is updated anyway was used to funnel additional position information from a layout into the already existing positioning code. Deprecating and removing the parameter would be a medium compatibility hassle. Therefore, I chose only to document the status quo and its intention, so that users are less surprised. See matplotlib#23629 (comment)
|
Not going forward with the deprecation. The parameter is now documented as internal and that's good enough. |
PR Summary
In itself
apply_aspect()feels too low level to be needed by an end-user. I'm not sure if it reasonably works at all on its own. It might be that the result is influenced by other artists and if something is left stale, then the application result is still not final. In that case, one would need a full draw (without rendering) anyway. OTOH ifapply_aspect()was really standalone, I don't know why we defer the application and don't apply it immediately when setting the aspect.Also, the position parameter of
apply_aspectdoes not seem a user-facing API. It is used internally by layout mechanisms. But shifting the position explicitly as part of aspect application seems overreach of the scope of the function and at least not something a user should do.I'm 99% percent sure
apply_aspect()does not need to be public API, but please speak up if you think otherwise. And even if we miss a justified use-case for a user. They should be able to give feedback during the deprecation period and we would reconsider then.