matplotlib instances getting stuck?

[enochii]

Describe the issue

Hi, thanks for the project!

I noticed that instances from matplotlib__matplotlib-* (for instance, matplotlib__matplotlib-26291) often take over 1 hour and still do not finish, while instances from other projects (e.g., django, astropy) usually complete within a few minutes under the same setup.

From the logs, it seems to spend a long time during conda environment creation, especially at the solving stage:

Collecting package metadata (repodata.json): ...working...
2026-01-07 19:36:24,060 - INFO - done
Solving environment: ...working...

Is this expected for matplotlib?
Any suggestions on how to handle these instances?

Suggest an improvement to documentation

No response

[nielstron]

Hi @enochii ,

Thanks for letting me know about this issue! We received some independent reports that some instances, e.g. matplotlib__matplotlib-23314, matplotlib__matplotlib-24149, can not be built at all anymore. We are currently investigating what might be the reason for these issues.

[enochii]

Hi, @nielstron

Thanks for your kind reply!

Observing that matplot__matplot-26291 became stuck during conda environment creation, I extracted the corresponding environment.yml. This hanging behavior—when running conda env create --file environment.yml—can be reliably reproduced in my laptop using the YAML file shown below.

Interestingly, by adding - nodefaults under - conda-forge to restrict the solver’s search space (and by also considering the use of the --solver=libmamba option), the environment can be created within a few minutes. I wonder whether this approach might also help resolve the other cases you mentioned above, or whether it could introduce other issues.

# To set up a development environment using conda run:
#
#   conda env create -f environment.yml
#   conda activate mpl-dev
#   pip install -e .
#
name: testbed
channels:
  - conda-forge
dependencies:
  # runtime dependencies
  - cairocffi
  - contourpy>=1.0.1
  - cycler>=0.10.0
  - fonttools>=4.22.0
  - importlib-resources>=3.2.0
  - kiwisolver>=1.0.1
  - numpy>=1.21
  - pillow>=6.2
  - pybind11>=2.6.0
  - pygobject
  - pyparsing>=2.3.1
  - pyqt
  - python-dateutil>=2.1
  - setuptools
  - setuptools_scm
  - wxpython
  # building documentation
  - colorspacious
  - graphviz
  - ipython
  - ipywidgets
  - numpydoc>=0.8
  - packaging
  - pydata-sphinx-theme
  - pyyaml
  - sphinx>=1.8.1,!=2.0.0
  - sphinx-copybutton
  - sphinx-gallery>=0.12
  - sphinx-design
  - pip
  - pip:
      - mpl-sphinx-theme
      - sphinxcontrib-svg2pdfconverter
      - pikepdf
  # testing
  - coverage
  - flake8>=3.8
  - flake8-docstrings>=1.4.0
  - gtk4
  - ipykernel
  - nbconvert[execute]!=6.0.0,!=6.0.1,!=7.3.0,!=7.3.1
  - nbformat!=5.0.0,!=5.0.1
  - pandas!=0.25.0
  - psutil
  - pre-commit
  - pydocstyle>=5.1.0
  - pytest!=4.6.0,!=5.4.0
  - pytest-cov
  - pytest-rerunfailures
  - pytest-timeout
  - pytest-xdist
  - tornado
  - pytz
  - black

[enochii]

Hi @enochii ,

Thanks for letting me know about this issue! We received some independent reports that some instances, e.g. matplotlib__matplotlib-23314, matplotlib__matplotlib-24149, can not be built at all anymore. We are currently investigating what might be the reason for these issues.

Hi @nielstron

Seems that when applying the trick (mainly by adding - nodefaults) above to accelerate the conda creation process, some matplotlib instances, including the matplotlib__matplotlib-24149 you mentioned, can now be built and evaluated.

I also encountered EvaluationError on two instances, namely pylint-dev__pylint-7277 and astropy__astropy-8872.

[Martin7-1]

@enochii Hi! I encounter the same problem as yours. Could you please help me on how to add - nodefaults to environment.yml? Are you directly modifying environment.yml file in matplotlib repo?

Thanks in advance :)