René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"[MNT]: Lazy imports in python 3.15+","articleBody":"A ways off, but creating this tracking issue to collect ideas as they arise.\n\nPython 3.15 (October 2026) will introduce [PEP 810 - Explicit lazy imports](https://peps.python.org/pep-0810). This has the potential to significantly improve startup time performance - as one example Meta reports 70% reduction in initialization time by converting its libraries to use lazy imports ([link](https://developers.facebook.com/blog/post/2022/06/15/python-lazy-imports-with-cinder/)). I expect the performance increase here to compound as our dependencies roll in lazy import support as well. With how widely used matplotlib is, I expect people to start asking for this before too long.\n\nFor backwards compatibility, we will need to use the `__lazy_modules__` approach rather than the `lazy` import syntax:\n\n```python\n__lazy_modules__ = [\"json\", \"collections.abc\"]\n\nimport requests  # eager, since not in list\nimport json  # lazy\nfrom json import dumps, loads  # lazy, each name is a lazy proxy\nfrom collections.abc import Mapping  # lazy\nfrom numpy import *  # eager, but we should move away from star imports\n```\n\nSome things may currently rely on eager import-time execution (global state, ImportError checks, backend/plugin registration (?), etc). Anything relying on eager execution can be left out of the `__lazy_modules__` list and fixed at a later time.\n\nThere are a few existing issues talking about deferred imports, which I believe will be subsumed by this. We can talk about whether it's worth putting effort into those anyways for older python versions: https://github.com/matplotlib/matplotlib/issues/22247, (maybe) https://github.com/matplotlib/matplotlib/issues/30578.\n\nRough plan:\n\n- [ ] Benchmark existing startup time (should measure both `import matplotlib` and `import matplotlib.pyplot as plt`)\n- [ ] Audit code for anything relying on eager import-time execution\n- [ ] Audit code for deferred imports which can be brought up to top-level\n- [ ] Audit code for star imports\n- [ ] Any code relying on a particular import _order_ should be fixed\n- [ ] Move imports over to `__lazy_modules__` (excluding things relying on eager imports)\n- [ ] Benchmark new startup time\n- [ ] Add CI test coverage for all-eager and all-lazy import cases to catch breakage: `python -X lazy_imports=none -m pytest` `python -X lazy_imports=all -m pytest`\n\nAlternatively this could be an incremental roll-in, focusing on the heavier weight modules.","author":{"url":"https://github.com/scottshambaugh","@type":"Person","name":"scottshambaugh"},"datePublished":"2026-01-04T17:30:30.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":4},"url":"https://github.com/30933/matplotlib/issues/30933"}