[babyhuey]

What

tools/pr_overlap.py now compares exports by NID, so an explicit APS5_EXPORT("<nid>", <function>) in one pull request is reported against a named export with the same NID in another (#9921).

exports() collects APS5_VABI definitions by name and skips _nid_no_patch functions, so APS5_EXPORT("Pz262plj20w", placeholder_nid_no_patch) added nothing. A pull request implementing FT_Outline_GetInsideBorder_nid_postfix, whose NID is Pz262plj20w, then showed no shared export with it, and merging both fails the NID patch step with duplicate patched export name: Pz262plj20w.

• New aliases(diff) collects the NIDs of added APS5_EXPORT lines, using nid_names.source_aliases, the parser check_conventions.py already uses. scan() stores them in pr["aliases"].
• overlaps() now calls same_exports(), which compares export_ids(): compute_nid of each named export plus the aliases. A match is listed by its function name when either side has the name, and by NID when both sides are aliases. NIDs that a shared dependency adds are still left out.
• exports() is unchanged. An _nid_no_patch helper without an APS5_EXPORT line still adds nothing.

Like the existing name comparison, the NID comparison is not scoped to a module. I tried scoping it by source directory, and that misses the case in this issue: #9922 implements FT_Outline_GetInsideBorder in core/libs/prx/libSceFreeType/src/OutlineBorders.cpp, which its CMake builds into the provider that has #9462's placeholder:

add_library(libSceFreeTypeOt SHARED EXCLUDE_FROM_ALL Export.cpp ../libSceFreeType/src/OutlineBorders.cpp)
add_library(libSceFreeTypeOl SHARED EXCLUDE_FROM_ALL Export.cpp ../libSceFreeType/src/OutlineBorders.cpp)

Scoping by the built PRX would need the CMake source lists, and many use variables, included .cmake files and globs. So a match in different providers is reported too. In the open-PR scan below, those are 43 pairs: the four bulk declaration PRs (#9451, #9453, #9462, #9498), or #9922 which carries #9462's declarations, against implementations in sibling providers (for example #9453's libSceFreeTypeOl placeholder for FT_Get_Kerning against #8558's libSceFreeType implementation). They don't fail the NID patch step on their own, but each is the same function declared as not implemented in one PR and implemented in the other.

#8970 also edits tools/pr_overlap.py (comparison against exports already on main, as noted on #9921). The two touch neighbouring lines, so whichever merges second needs a small rebase.

Tested

• Ubuntu 24.04 (the anyps5-build CI image, Python 3.12.3): python3 -m unittest discover -s tools/tests -p "test_pr_overlap*.py" runs 70 tests, OK (3 skipped, the same skips as on main). The two new tests in test_pr_overlap_exports.py fail on main's pr_overlap.py and pass with this change.
• Windows 11 Pro 25H2 (10.0.26200), Python 3.13.15, this head: the same unittest discover runs 70 tests, OK (5 skipped: the three bash and jq workflow-step tests, the POSIX control-character path test and SubdirectoryTests, whose fake gh is a shell script, the same as on main), and both new tests pass. python pr_overlap.py --help from tools/ starts without an import error.
• The pair from bug(pr-overlap): explicit APS5_EXPORT NIDs are omitted from overlap checks #9921: scan() and overlaps() on feat(libSceFreeTypeOt): declare 150 PS5 3.20 exports #9462 (718e1755ac) and feat(freetype): implement system outline border queries #9922 (beb7b21f3d) treated as unrelated pull requests. main reports no shared exports. This branch reports FT_Outline_GetInsideBorder and FT_Outline_GetOutsideBorder (feat(freetype): implement system outline border queries #9922's implementations against feat(libSceFreeTypeOt): declare 150 PS5 3.20 exports #9462's Pz262plj20w and sKneJjUXWkQ), plus the 148 NIDs feat(freetype): implement system outline border queries #9922 carries from feat(libSceFreeTypeOt): declare 150 PS5 3.20 exports #9462. In the real scan, feat(freetype): implement system outline border queries #9922 lists Depends on: #9462, so that pair stays suppressed.
• Full open-PR scan (python3 tools/pr_overlap.py --base main --out ... on main 895f47f4ee, 1,285 open PRs present in both runs) with main's script and with this head: 43 pairs gain shared exports, all of them the cross-provider pairs above (feat(libSceFreeType): query Type 1 private dictionaries #9933 is a new one against feat(libSceFreeTypeOl): declare 150 PS5 3.20 exports #9453). No pair loses any. (One row, feat(libc): export _FDsign/_Dsign like MSVC CRT (PPSA34015) #6085 against duplicate-feat(libc): add _FDsign #9951, appears only in the main run because duplicate-feat(libc): add _FDsign #9951 was opened between the two scans.)
• python3 -I tools/check_conventions.py with this title and body: passed. git diff --check: clean.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: Fixes bug(pr-overlap): explicit APS5_EXPORT NIDs are omitted from overlap checks #9921
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

[MarcZX10]

I reviewed the alias parser and NID comparison path. It normalizes named _nid_postfix exports through compute_nid, collects only added APS5_EXPORT aliases through the shared C++ token parser, suppresses identities inherited from common dependencies, and preserves the no-alias behavior for _nid_no_patch helpers. The unit cases cover alias collection/removal, alias-to-name matching, alias-to-alias output and shared-dependency suppression. I also ran the exact-head test_pr_overlap*.py suite from commit 8f2261b14bf6d8bc6a888e04832ead9b318ea1a6: 69 tests passed. The first local attempt lacked the workflow fixture in my /dev/shm snapshot; after adding that exact-head fixture, the suite passed. Hosted eligibility, conventions, intake and the completed relinker/ASan checks pass; Windows relinker is still running and product jobs are skipped.

The live base has advanced: this head is 1 commit ahead/11 behind current main caa089fd735d0a5d615279e0f567cb638ad65690, and the open queue is now 1,287 PRs. Please sync to current main and refresh the full overlap scan against the live queue before merge. The published 44-pair scan is useful and transparent, but it ran on main 5366b8b91 with 1,285 PRs. I found no implementation defect in the reviewed diff; this follow-up is for current-base/current-queue validation.

[MarcZX10]

Follow-up with exact-head evidence: I reran the test_pr_overlap*.py suite on 8f2261b14bf6d8bc6a888e04832ead9b318ea1a6 (69/69 pass), then ran the PR branch scan against the live main and open-PR queue. It reported no confirmed overlapping PR pair; two comparisons remain manual because the shared files conflict with the base: #988 (tools/pr_overlap.py) and #8970 (tools/pr_overlap.py, tools/tests/test_pr_overlap_exports.py). I inspected those diffs: #988 adds scan/job-summary logging, while #8970 handles export collisions against base main; they cover distinct behavior from this PR’s PR-to-PR explicit-alias comparison. Please preserve both changes when integrating. Current hosted eligibility/conventions/intake/labels and cross-platform relinker/ASAN checks pass; product builds are skipped.

[github-actions]

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 988	tools/pr_overlap.py
PR 8970	tools/pr_overlap.py tools/tests/test_pr_overlap_exports.py

[babyhuey]

@MarcZX10 Synced and rescanned. The head is now 957b65fa2f, rebased onto main 895f47f4ee, the same patch (it picks up #9666's change to the same file without a conflict).

• Linux (CI image): test_pr_overlap*.py 70 tests, OK (3 skipped). Windows 11, Python 3.13: 70 tests, OK (5 skipped, all platform skips from main, including fix(tools): read pr_overlap paths from the top of the repository #9666's new SubdirectoryTests), and python pr_overlap.py --help from tools/ imports nid_names fine.
• Full scan on the live queue, main's script and this head on 895f47f4ee, 1,285 open PRs in both runs: 43 pairs gain shared exports, the same cross-provider kind as before (feat(libSceFreeType): query Type 1 private dictionaries #9933 is new, against feat(libSceFreeTypeOl): declare 150 PS5 3.20 exports #9453), and none lose any. The only row that differs the other way, feat(libc): export _FDsign/_Dsign like MSVC CRT (PPSA34015) #6085 against duplicate-feat(libc): add _FDsign #9951, is because duplicate-feat(libc): add _FDsign #9951 was opened between the two runs. I've updated the description.

#988 and #8970 keep their behaviour; whichever of #8970 and this lands second needs the small rebase noted in the description.

[adriadam10]

Checked aliases()/same_exports() against nid_names.compute_nid and source_aliases (only added APS5_EXPORT lines are collected, shared-dependency NIDs are still excluded, exports() is unchanged). Both new tests need the new functions, so they fail on main. Merged.