Hash pyi files after ruff post-processing in pyi_generator

[masenf]

pyi_hashes.json entries were computed as each stub was written, before
scan_all ran ruff format and ruff check --fix over the generated
files. Since ruff materially rewrites the stubs (quoting, wrapping,
import fixes), the registry recorded hashes of intermediate content
that never exists on disk afterwards: any generator change affecting
only the pre-format output flagged hash changes even when the final
.pyi files were byte-for-byte identical.

_scan_file now just records which .pyi files were written, and scan_all
computes the md5 hashes from the final on-disk content after ruff
post-processing. written_files becomes a per-instance list (it was a
mutable class attribute shared across generator instances) and the dead
modules/root/current_module class attributes are dropped.

pyi_hashes.json is regenerated with the new scheme (one-time value
churn for every entry; keys unchanged). Verified idempotent across
repeated --force runs and explicit-target merge runs.

[greptile-apps]

Greptile Summary

This PR fixes a bug where pyi_hashes.json entries were computed from the intermediate generator output before ruff format / ruff check --fix rewrote the stubs, causing spurious hash changes even when the final on-disk .pyi files were byte-for-byte identical. Hashes are now computed from the final on-disk content after ruff post-processing.

• _write_pyi_file is simplified to -> None; _scan_file now returns the .pyi path string instead of a (path, hash) tuple; scan_all computes hashes from disk after ruff runs and writes them only in the use_json path.
• PyiGenerator.written_files is moved from a mutable class attribute (shared across all instances) to a per-instance list initialized in a new __init__, fixing a latent shared-state bug.
• strict=False is tightened to strict=True in the non-prune zip branch; a new regression test (test_hashes.py) verifies the recorded hash matches the post-ruff file content end-to-end.

Confidence Score: 5/5

Safe to merge — the change is narrowly scoped to hash-timing and a class-attribute cleanup with no behavioral impact on stub generation itself.

The stub writing logic is unchanged; only the moment at which hashes are computed moves (from before ruff to after). The mutable class-attribute fix eliminates a potential cross-instance contamination. The new test exercises the full end-to-end path.

No files require special attention.

Important Files Changed

Filename	Overview
packages/reflex-base/src/reflex_base/utils/pyi_generator.py	Moves hash computation to after ruff post-processing; fixes shared mutable class attribute bug in PyiGenerator; tightens zip strict=True in non-prune path
tests/units/reflex_base/utils/pyi_generator/test_hashes.py	New regression test that verifies recorded hash matches on-disk content after ruff processes the generated stub
tests/units/reflex_base/utils/pyi_generator/test_regression.py	Updated written_files iteration to match new list[str] type (was list[tuple[str,str]])
pyi_hashes.json	All 124 hash values regenerated to reflect post-ruff file content; keys are unchanged
packages/reflex-base/news/+pyi_hashes_post_ruff.bugfix.md	Changelog entry describing the hash computation fix

_{Reviews (2): Last reviewed commit: "Re-trigger CI" | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

✅ 26 untouched benchmarks
⏩ 8 skipped benchmarks¹

---

_{Comparing claude/peaceful-curie-ew0aa9 (780ed9b) with main (fdb00db)}

Footnotes

1. 8 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩