Unregister_scale() function to matplotlib.scale by dikshajangra12918-oss · Pull Request #31547 · matplotlib/matplotlib

dikshajangra12918-oss · 2026-04-22T13:36:43Z

PR Summary

Currently matplotlib has register_scale() but no way to remove a registered scale. This is a problem when a custom scale is created for just one plot — there is no way to clean up after that plot is drawn.

This PR adds unregister_scale(name) to fix that gap. Uses pop() for idiomatic dict cleanup.

Changes made:
~ Added unregister_scale() in lib/matplotlib/scale.py
~ Raises ValueError if scale name is not registered
~ Added tests in lib/matplotlib/tests/test_scale.py

Closes #5791

AI Disclosure

I used Claude (AI) for guidance on implementation approach. Code was written and verified by me.

PR Checklist

~ [x] Closes #5791 is in the PR description
~ [x] New code is tested

Added unregister_scale(name) function that allows users to remove previously registered custom scales from the registry. Raises ValueError if the scale name is not found.

Added two test cases in test_scale.py: 1. test_unregister_scale: registers a temporary custom scale and verifies it is removed correctly. 2. test_unregister_scale_invalid: verifies that ValueError is raised when trying to unregister a scale that doesn't exist.

story645 · 2026-04-22T18:16:20Z

        )


+def unregister_scale(name):


Suggested change

def unregister_scale(name):

def deregister_scale(name):

Sorry I am late to this. While I agree that “deregister” seems like the better term in isolation, I noticed that the color map and color sequence registries both use “unregister”. So should we be consistent with those?

Consistency is important. As non-native speaker I have no clear feeling what is more natural, but this is what AI says

Both are used in practice, but they carry slightly different connotations:

“unregister” is the more common choice in most APIs and codebases. It reads naturally as the inverse of register, and is widely used in standard libraries, frameworks, and event systems.

“deregister” is also correct English, but it’s less common in programming contexts and can feel a bit more formal or domain-specific (you’ll see it more in networking, telecom, or standards documents).

From a naming consistency and developer expectation standpoint, unregister is usually the better choice.

So it seems, we should go with unregister, which is also in line with ColormapRegistry.

yeah I'll third this even though I'm the one who requested the change to deregister. Though really I think it'd be nicer to have an API more like the colormap registery where it's just register and unregister - ETA: which is a follow up discussion.
the action here is to change it back to unregister_scale.

While you could have a ScaleRegistry, it's much less useful than a ColormapRegistry. The primary (public) use of the colormap registry is for users to be able to get a colormap by name. But nobody needs to get a scale class from a name. So the only public API at all is register_scale, unregister_scale, and that's primarily for some third party libs. I wouldn't bother moving away from functions here.

But nobody needs to get a scale class from a name

set_{x/y}scale but also I think norm_from_scale should support strings (am surprised it doesn't - and yes I'll open an issue if there isn't one already opened).

Co-authored-by: hannah <story645@gmail.com>

dikshajangra12918-oss · 2026-04-23T08:42:33Z

@story645,

The remaining failures are pre-existing issues with Qt backends and are unrelated to this PR.
All checks related to our changes are passing!

story645 · 2026-04-23T16:49:08Z

    assert 'temp_test_scale' not in mscale._scale_mapping


 def test_unregister_scale_invalid():


Thanks @story645!

I've updated the function and test names to use deregister_scale. Please taek a look.

still didn't catch everything,

story645 · 2026-04-24T04:34:43Z

        assert isinstance(ax.xaxis.get_transform(), CustomTransform)
    finally:
-        # cleanup - there's no public unregister_scale()
+        # cleanup - there's no public deregister_scale()


Why not just use the method you just wrote?

@story645,

Could you help me understand what exactly still needs to be fixed? I've updated the function name, tests, and stub file. The remaining failures seem to be pre-existing issues but I want to make sure I haven't missed anything on my end.

Any guidance would be really helpful!

@dikshajangra12918-oss the test that you added is failing
https://github.com/matplotlib/matplotlib/actions/runs/24874799863/job/72828932746?pr=31547#step:13:179

You should also consider the logic of having a comment that states "there's no public deregister_scale" when that is the very function that you are adding.

Hii,

All requested changes are done:

Fixed TempScale to use IdentityTransfor

Fixed whitespace issues

Removed outdated comment

Our test_deregister_scale is passing.
The remaining failures are pre-existing platform issues unrelated to this PR.
Please let me know if anything else needs fixing!

Our test_deregister_scale is passing.

That test is still not passing. Please read our guide on testing. In particular, it shows how to run a single test. I suggest you run your single test that way because you will then get an immediate result that is just about that test and not be confused whether other test failures are involved.
https://matplotlib.org/devdocs/devel/testing.html

Also, please read through your change using the "files changed" tab above. The comments about there not being a deregister_scale are still there.

I'm not able to understand why these 2 checks are failing- 'test_backened_getattr[backend_template]" and 'test_backend_getattr[backend_tkagg]'. Could you help me understand what's wrong here?

All checks related to our 'deregister_scale' changes are passing. Please let me know if anything else needs to be fixed!

@rcomer @story645 , I did research on the failing tests on Windows Python 3.13, which seems to be a pre-existing platform issue unrelated to this PR. Could you please confirm if that's the case?

@dikshajangra12918-oss yes that one is a known flaky test so not caused by this PR.

@rcomer Thankyou for confirming! Is this ready for merging, orr are there any other changes needed?

rcomer · 2026-04-28T11:43:45Z


+def deregister_scale(name):
+    """
+    Remove a custom scale from the registry.


What happens if someone tries to deregister a built in scale? What should happen?

tangently I'm wondering if we should add a deregister function to the colormap registry (or is .pop good enough since the registry itself is public?)

@story645 the colormap registry has
https://matplotlib.org/stable/api/cm_api.html#matplotlib.cm.ColormapRegistry.unregister

@rcomer I think it should raise a "ValueError" to prevent accidental removal of built-in scales. I'll add that check to 'deregister_scale()'. Does that sound right?

@story645 That's good point! Should I address the colormap registry in this PR as well, or would it be better as a separate issue?

@story645 the colormap registry has https://matplotlib.org/stable/api/cm_api.html#matplotlib.cm.ColormapRegistry.unregister

@rcomer Thanks for the reference!
Following the same pattern as 'ColormapRegistry.unregister', I'll update 'deregister_scale()" to raise a 'ValueError' if someone tries to deregister a built-in scale like 'log', 'linear', etc.

That's good point! Should I address the colormap registry in this PR as well, or would it be better as a separate issue?

As @rcomer pointed out, the function already exists for colormaps. I just missed it in the docs.

That's good point! Should I address the colormap registry in this PR as well, or would it be better as a separate issue?

As @rcomer pointed out, the function already exists for colormaps. I just missed it in the docs.

Got it, thanks for clarifying!

dikshajangra12918-oss · 2026-04-30T17:11:27Z

@story645 @rcomer AppyVeyor failure is unrelated to this PR - result_images directory not found error in visualize_tests.py on Windows CI

story645 · 2026-04-30T18:59:21Z

+    name : str
+        The name of the scale to remove.
+    """
+    _builtin_scales = {'linear', 'log', 'symlog', 'logit',


if this isn't defined anywhere else, I think this should be defined up top and then added to the scale registery. Use the pattern from

matplotlib/lib/matplotlib/cm.py

Lines 195 to 197 in f6e850d

if name in self._builtin_cmaps:

raise ValueError(f"cannot unregister {name!r} which is a builtin "

"colormap.")

dikshajangra12918-oss and others added 3 commits April 21, 2026 15:25

Add unregister_scale() function to matplotlib.scale

12c9d66

Added unregister_scale(name) function that allows users to remove previously registered custom scales from the registry. Raises ValueError if the scale name is not found.

Use pop() instead of del for removing scale

0a4417d

github-actions Bot added the topic: transforms and scales label Apr 22, 2026

dikshajangra12918-oss marked this pull request as draft April 22, 2026 14:17

apple added 3 commits April 22, 2026 22:49

Also remove from _scale_has_axis_parameter in unregister_scale

cc686b1

Fix formatting in test_unregister_scale

99690bc

Remove whitespace from blank line

d9bf8cf

dikshajangra12918-oss marked this pull request as ready for review April 22, 2026 18:05

story645 reviewed Apr 22, 2026

View reviewed changes

dikshajangra12918-oss and others added 2 commits April 23, 2026 05:10

Update lib/matplotlib/scale.py

2747f55

Co-authored-by: hannah <story645@gmail.com>

Add deregister_scale to stub file

e673532

Update tests to use deregister_scale

d56e2ef

story645 reviewed Apr 23, 2026

View reviewed changes

apple added 2 commits April 23, 2026 22:43

Rename test functions to use deregister_scale

0081c89

Replace all unregister_scale with deregister_scale

8400592

story645 reviewed Apr 24, 2026

View reviewed changes

apple added 4 commits April 24, 2026 11:30

Use deregister_scale() in test cleanup

b83f45e

Remove outdated comment about deregister_scale

aaf5b7a

Fix TempScale using IdentityTransform

799bf8e

Fix Whitespace in blank lines

dc559da

rcomer mentioned this pull request Apr 25, 2026

Add IndirectTransform for lazy coordinate resolution #31488

Draft

Remove outdated comments about no public deregister_scale()

41ded34

rcomer reviewed Apr 28, 2026

View reviewed changes

story645 mentioned this pull request Apr 28, 2026

DOC: added unregister to colormap guide #31580

Merged

5 tasks

apple added 4 commits April 30, 2026 20:00

Rename deregister_scale to unregister_scale and protect built-in scales

5398c62

Fix whitespace and missing newline in test_scale.py

390d67d

Fix trailing blank line in test_scale.py

e0a87e8

Add newline at end of test_scale.py

425d0cc

story645 requested changes Apr 30, 2026

View reviewed changes

		assert 'temp_test_scale' not in mscale._scale_mapping


		def test_unregister_scale_invalid():

	if name in self._builtin_cmaps:
	raise ValueError(f"cannot unregister {name!r} which is a builtin "
	"colormap.")

Uh oh!

Conversation

dikshajangra12918-oss commented Apr 22, 2026

PR Summary

AI Disclosure

PR Checklist

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

story645 Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dikshajangra12918-oss commented Apr 23, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dikshajangra12918-oss commented Apr 30, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

story645 Apr 28, 2026 •

edited

Loading