Repository navigation
Add an author preflight for Pyright changes - #12037
Bill Schnurr (bschnurr) merged 2 commits into
Conversation
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
🔒 Automated review in progress — Rich Chiodo (@rchiodo) is auto-reviewing this PR. |
|
Reviewed commit: Result: 🔴 Verification detailsVerification: Automated verification is best effort. The PR author remains responsible for testing. Summary: [unavailable] Container verification could not start and local execution was not authorized for this PR HEAD: C:\Users\rchiodo\AppData\Local\Programs\Podman\podman.EXE --connection pyrx-automation failed: stderr: Error: unable to start container "35b98b7c3c1af0c646c3287227daca1a51cfb3be5271c6bc109c7cf8da1eda10": crun: open `memory.max` for writing: No such file or directory: OCI runtime attempted to invoke a command that was not found Test runs: none recorded. |
Rich Chiodo (rchiodo)
left a comment
There was a problem hiding this comment.
Approved via Review Center.
|
According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅ |
Why
A PR can fix its motivating example and still regress a nearby case. In #11951, a call spelled
issubclasscould refer to a shadowing function rather than the builtin, and a bad return in a reachable branch went undiagnosed. The goal is to help authors catch assumptions like this while the change is still local or in draft, without waiting for a full review.What changes
pyright-pr-review. It reviews selected local changes or a draft PR directly, without launching the five specialists. For typing changes, it compares the motivating case, a nearby valid case, and a nearby invalid case against the base. It checks exact inferred types and diagnostic locations rather than just error counts. It reports what ran, how to replay findings, and what remains untested. It does not edit tracked files, post to GitHub, or issue a merge verdict. Full review remains the default.issubclassexample, shadow the name and check whether the diagnostic inside the reachable branch is preserved. Preflight probes the riskiest contracts first and marks the rest untested.What the review feedback added
A qualitative pass over public feedback on the original 60-PR queue surfaced gaps beyond shortcut assumptions:
end. Check a later consumer for shared-state changes and the full lifecycle for protocol changes.How to try it
Select
pyright-pr-reviewand ask: “Preflight my local analyzer changes against upstream main. Run the motivating case and nearby valid and invalid controls. Show me how to replay any failures. Do not edit or publish anything.”Validation
Prettier,
git diff --check, and links between the agent, skill, and guide passed. The new mode has not been exercised end to end. This PR changes instructions and docs only; it does not change the analyzer or add a CI gate.