BLD: pin matplotlib to < 3.11.0

[adamwitmer]

As mentioned in issue #36, the newest release of matplotlib (3.11.0) includes changes to text and font handling that break tests that rely on pixel level image comparisons (https://matplotlib.org/stable/release/prev_whats_new/whats_new_3.11.0#fonts-and-text). This PR pins matplotlib to versions below 3.11.0 to avoid implementing suggested work-arounds from the matplotlib release notes including:

1. Regenerating all test images and removing support for versions lower than 3.11.0. I don't think we want to exclude all older versions of matplotlib in our workflow.
2. Removing text from plots or using placeholders. We specifically check font style with image comparison.
3. Loosening image comparison tolerances. We are concerned with pixel level changes that may be missed by loosening tolerances, and we have already loosened tolerances to account for sub-pixel level inconsistencies between testing platforms.
4. Changing the image comparison algorithm used for testing. "Perpetual hashing" was mentioned as one such option, but would require substantial refactoring of the current testing architecture.

AI Disclosure: No AI tools were used in generating original commit to this PR. Claude was used to check what was common practice for implementing dependency version checks in open source packages, and suggestions were cross-referenced with widely used open source packages such as pandas

[tylerjereddy]

If you want to do this temporarily that's fine, but:

• I'd still expect you to support >= 3.11.0 by the original deadline Mihee asked for; you may need to just check how much more disk space is needed to support reference images for both versions in tests.
• You should probably error out at import time if the user is using >= 3.11.0, so that you fail "fast and early" until 3.11.x series support is verified/tested.

[tylerjereddy]

In addition to that change, please always include an AI disclosure in your original comment, as noted previously.

[adamwitmer]

I added a check to neat_ml/__init__.py to raise an ImportError when importing unsupported versions of matplotlib. I asked an LLM what the best practice was for erroring out at runtime when an unsupported version of a dependency is being used and it suggested using the packaging.version.Version package. pandas does something similar, but with a vendored version of Version (https://github.com/pandas-dev/pandas/blob/main/pandas/util/version/__init__.py) that they use to check some version compatibility, e.g. https://github.com/pandas-dev/pandas/blob/main/pandas/compat/numpy/__init__.py

Locally, this error does raise when importing matplotlib==3.11.0, i.e.:

______________________________________________________________________________________________________________________________ ERROR collecting neat_ml/tests ______________________________________________________________________________________________________________________________
/Users/awitmer/anaconda3/envs/neat_ml_3_12/lib/python3.12/importlib/__init__.py:90: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
<frozen importlib._bootstrap>:1381: in _gcd_import
    ???
<frozen importlib._bootstrap>:1354: in _find_and_load
    ???
<frozen importlib._bootstrap>:1304: in _find_and_load_unlocked
    ???
<frozen importlib._bootstrap>:488: in _call_with_frames_removed
    ???
<frozen importlib._bootstrap>:1381: in _gcd_import
    ???
<frozen importlib._bootstrap>:1354: in _find_and_load
    ???
<frozen importlib._bootstrap>:1304: in _find_and_load_unlocked
    ???
<frozen importlib._bootstrap>:488: in _call_with_frames_removed
    ???
<frozen importlib._bootstrap>:1381: in _gcd_import
    ???
<frozen importlib._bootstrap>:1354: in _find_and_load
    ???
<frozen importlib._bootstrap>:1325: in _find_and_load_unlocked
    ???
<frozen importlib._bootstrap>:929: in _load_unlocked
    ???
<frozen importlib._bootstrap_external>:994: in exec_module
    ???
<frozen importlib._bootstrap>:488: in _call_with_frames_removed
    ???
neat_ml/__init__.py:6: in <module>
    raise ImportError("Workflow currently only supports `matplotlib<3.11.0`")
E   ImportError: Workflow currently only supports `matplotlib<3.11.0`

However, currently this check is not covered by our tests, is it common practice to add an explicit test for these checks to the repos test suite?

[tylerjereddy]

is it common practice to add an explicit test for these checks to the repos test suite?

No

pandas does something similar, but with a vendored version of Version

Right, because with millions of users adding a new dependency for something so simple as a version check is undesirable, but doesn't matter for us.

[tylerjereddy]

ok, this looks good to go to me, thanks