Repository navigation
P1 Repo-wide vendor-neutral terminology and regression guard - #319
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 938af90622
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| contents = path.read_text(encoding="utf-8", errors="replace") | ||
| for number, line in enumerate(contents.splitlines(), start=1): | ||
| match = PATTERN.search(line) | ||
| if match: |
There was a problem hiding this comment.
Clean up remaining matches before expanding the naming scan
At the reviewed commit, the expanded scan rejects two tracked documents that this patch leaves unchanged: docs/INTEGRATION_SIDE_BRANCH_TRIAGE_2026-10-22.md:58 contains SIPROTEC, and docs/integration/M3H_MAINLINE_RELEASE_SOURCE_CLOSURE_2026-10-24.md:68 contains Omicron. Checking the committed blobs against the new pattern confirms both violations. Because .github/workflows/product-naming-guard.yml runs this script on every pull request and main push, the change makes Product Naming Guard fail on its own repository contents. Update those remaining references before enabling the broader scan.
Useful? React with 👍 / 👎.
| if PATTERN.fullmatch(path.name.casefold()): | ||
| violations.append(f"filename: {relative}") |
There was a problem hiding this comment.
Preserve restricted-name detection across complete paths
Using fullmatch() against path.name only detects a filename consisting entirely of a restricted token. Names such as iedscout.cpp and legacy_iedscout.hpp now pass because their extensions prevent a full match, while iedscout/README.md passes because the directory component is discarded. All three were rejected by the previous substring check against the complete relative path, so this weakens the existing path guard despite expanding content coverage. Match restricted tokens within the complete relative path instead.
Useful? React with 👍 / 👎.
Public-facing hygiene
No runtime IEC 61850/SMV/PTP algorithms or released binaries are modified.
Merge gate: Product Naming Guard must pass, along with C++ and affected workflow CI. Inspect exact changed paths and historical PR attribution before merging.