Repository navigation
Scope the browser floor guard to compiled packages - #402
Merged
Merged
Conversation
I claimed threadpoolctl>=3.6.0 was a live browser-install bug against a Pyodide shipping 3.5.0. It is not. threadpoolctl publishes a py3-none-any wheel, so micropip pulls any version straight from PyPI and never touches the bundled build. The floor costs a download, not an install. The constraint is real only where no pure-Python wheel exists, because then Pyodide's build is the only one a browser can have. Checked against PyPI: numpy, scipy, h5py and matplotlib publish none, so their floors genuinely are a ceiling; threadpoolctl does, so its floor is not constrained this way. So the test now enforces the four compiled packages and records threadpoolctl in PYODIDE_PURE_PYTHON, deliberately unenforced, with the reason. Still verified non-vacuous: raising the non-darwin scipy floor to 1.15.0, above the 1.14.1 Pyodide ships, fails three tests by name. The four floors that shipped in #399 were correct and remain unchanged. What was wrong was the stated reason for one of them. Correction also filed on #400, whose original example was the same mistake.
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
If your organization's extra usage balance is empty, an organization admin can add extra usage credits at claude.ai/admin-settings/usage. If its monthly spend limit was reached, an admin can raise it on the same page. If neither applies, contact Anthropic support.
Once extra usage is available, reopen this pull request to trigger a review.
|
Claude encountered an error after 0s —— View job I'll analyze this and get back to you. |
This was referenced Sep 21, 2026
neuromechanist
added a commit
to nemarOrg/nemar-cli
that referenced
this pull request
Sep 21, 2026
Both the memory entry and the eegprep-lean contract said every dependency floor is a ceiling in the browser. That is only true of compiled packages. The test is whether the package publishes a pure-Python wheel, not whether Pyodide bundles it. A py3-none-any wheel means micropip fetches any version straight from PyPI and the bundled build goes unused. numpy, scipy, h5py and matplotlib publish none and are genuinely capped; threadpoolctl is bundled but pure Python and is not. The memory cited threadpoolctl>=3.6.0 on the epic branch as a live broken install, which was wrong. Kept as the instructive part rather than quietly deleted: being bundled made it look identical to the scipy case, and presence in pyodide-lock.json says nothing about whether another version is reachable. Over-applying the rule has its own cost, which is false alarms and a guard nobody trusts. Superseded claims removed rather than left beside the correction, per the directory's own rules. Test fix is sccn/eegprep#402.
This was referenced Sep 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What was wrong
I justified part of #399 by saying
threadpoolctl>=3.6.0was a live browser-install bug against aPyodide that ships 3.5.0. That is wrong. threadpoolctl publishes
threadpoolctl-3.6.0-py3-none-any.whl, so micropip pulls it straight from PyPI and never touchesthe bundled build. The floor costs a download, not an install.
What is actually true
Pyodide's bundled build is the only one a browser can have when the package publishes no
pure-Python wheel. Checked against PyPI:
py3-none-anyThe change
tests/test_browser_dependency_floors.pynow enforcesPYODIDE_COMPILED, the four packages wherethe constraint binds, and records
PYODIDE_PURE_PYTHONseparately with the reason it isdeliberately not enforced. The
pyproject.tomlcomment carried the same wrong claim and iscorrected.
The four floors that shipped in #399 are correct and unchanged. What was wrong was the stated
reason for one of them, and a test that enforced a constraint that does not exist.
Verification
Still non-vacuous, which is the point of the guard: raising the non-darwin scipy floor to 1.15.0,
above the 1.14.1 Pyodide ships, fails three tests by name. 10 passed unmodified.
ruff checkandruff format --checkclean.Correction also filed as a comment on #400, whose original example was this same mistake. That
issue stays open, because the gate does have a real blind spot for compiled packages, but the naive
fix (comparing lockfile-resolved versions) reports false failures since
uv.lockresolves ahead ofPyodide by design.