[bong-water-water-bong]

Refs #2160. A miss and a typo must not read the same.

The gap, hit in review

--where matches exactly, which is right: a substring match would answer for tmp_path_check_self when asked about tmp_path_check, and a partial match reported as a
registration is the false confidence this tool exists to remove. But that leaves two different situations with one sentence:

$ ci_targets_check.py --where submodule_pin
submodule_pin  NOT REGISTERED (searched CMakeLists.txt and every file it include()s or globs)

The row is submodule_pin_check. The same is true of the other two names in #2160's own body — tmp_path (row: tmp_path_check) and npu_serve_smoke (row:
npu_serve_smoke_self). So the PR that added --where to stop a release verifier raising a false alarm has three examples in its description that all read as alarms.

The change

On a miss, name the rows that start with the query, read by the same reader that follows include()s and the cmake/tests-*.cmake glob:

$ ci_targets_check.py --where submodule_pin
submodule_pin  NOT REGISTERED (searched CMakeLists.txt and every file it include()s or globs)
          but 2 row(s) START WITH it -- the row name may be what is wrong, not the registration:
          submodule_pin_check (cmake/tests-submodule-pin.cmake:26), submodule_pin_check_self (cmake/tests-submodule-pin.cmake:27)

The verdict stays NOT REGISTERED and the exit code stays 1 — the hint never claims a match. When nothing starts with the query the output is unchanged:

$ ci_targets_check.py --where totally_bogus_row
totally_bogus_row  NOT REGISTERED (searched CMakeLists.txt and every file it include()s or globs)

Verified on the real tree

query	before	now
submodule_pin	NOT REGISTERED	+ but 2 row(s) START WITH it
tmp_path	NOT REGISTERED	+ but 2 row(s) START WITH it
npu_serve_smoke	NOT REGISTERED	+ but 1 row(s) START WITH it — npu_serve_smoke_self (CMakeLists.txt:396)
tmp_path_check	cmake/tests-tmp-path.cmake:23	unchanged
totally_bogus_row	NOT REGISTERED	unchanged

--check unchanged (9 baselined, none new); copyright.py --check rc 0. Self-test gains two arms: "a partial name is offered as a hint when the exact name is absent" and
"and no hint is invented when nothing starts with the query" — the second is the one that stops the hint from becoming a fuzzy match.

Why this shape

Same distinction #2093 drew for CANCELLED-vs-FAILURE and #2125 for SKIP-vs-PASS: the answer must not conflate "absent" with "you asked for the wrong string", and the
difference has to be visible in one line of output, because the line is what the next reader quotes.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 PR contains tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Redundant all_rows call The all_rows function is called twice in the main function when handling the --where option: once on line 833 and again on lines 838-839. This results in redundant file reading and processing, which could be inefficient, especially if the CMake files are large or numerous. hits_all = all_rows(root, os.path.join(root, "CMakeLists.txt")) for row, rel, line in hits: print(f"{row}\t{rel}:{line}") for row in wanted: if row not in found: near = [r for r, _, _ in all_rows(root, os.path.join(root, "CMakeLists.txt")) if r != row and r.startswith(row)] hint = "" if near: where = {r: f"{rel}:{line}" for r, rel, line in hits_all} hint = ("\n\t but %d row(s) START WITH it -- the row name may be what is wrong, not the " "registration: %s" % (len(near), ", ".join(f"{r} ({where.get(r, '?')})" for r in near))) print(f"{row}\tNOT REGISTERED (searched CMakeLists.txt and every file it include()s or globs){hint}") Potential performance issue with prefix matching The logic for finding rows that start with the query string (r.startswith(row)) is performed on the full list of rows returned by all_rows. If the number of rows is large, this could lead to performance degradation, particularly in the context of a CI tool that may be invoked frequently. near = [r for r, _, _ in all_rows(root, os.path.join(root, "CMakeLists.txt")) if r != row and r.startswith(row)] hint = "" if near: where = {r: f"{rel}:{line}" for r, rel, line in hits_all} hint = ("\n\t but %d row(s) START WITH it -- the row name may be what is wrong, not the " "registration: %s" % (len(near), ", ".join(f"{r} ({where.get(r, '?')})" for r in near)))

[bong-water-water-bong]

VERDICT at 95bf054: CHANGES

Reviewed head 95bf054 (refs #2160). The feature is sound: a miss on the exact rule names the rows that start with the query, the verdict stays NOT REGISTERED, and the hint never claims a match. The blocker is the merge, because of where it now lands.

It conflicts with main, and the obvious resolution brings a bug back. #2164 (merged) replaced --where's exit with return where_rc(hits, wanted). That function gives rc 1 when any requested row is unregistered, where 0 if found else 1 passed a gate list with one missing row. This branch predates #2164, so git merge-tree origin/main 95bf054 gives CONFLICT (content): Merge conflict in tools/ci/ci_targets_check.py, in two hunks:

• The self-test: #2164's three exit-status arms against this PR's two prefix-hint arms. Keep both.
• main()'s --where branch: this PR's side carries its hint lines and return 0 if found else 1. Taking that side whole would silently restore the bug #2164 fixed.

Resolution, tested in a throwaway merge of this head into main: keep both self-test blocks, and in main() take this PR's hint lines with return where_rc(hits, wanted) as the last line. With that:

• --self-test: 8 of 8 --where arms ok, both PRs' arms included; "0 failure(s)". --check is unchanged: 9, none new.
• --where submodule_pin gives rc 1, with "but 2 row(s) START WITH it -- … submodule_pin_check (cmake/tests-submodule-pin.cmake:26), submodule_pin_check_self (…:27)".
• --where ci_targets,submodule_pin gives rc 1. That is the case the old line would pass.
• --where ci_targets,tmp_path_check gives rc 0.

I removed the throwaway worktree; nothing was pushed. Rebase or merge main with that resolution and this is ready from my side.

[bong-water-water-bong]

VERDICT: READY at 2b931f1

CHANGES answered: rebased onto main, both conflicts resolved exactly as prescribed, and the bug is not brought back.

Resolution applied as you tested it:

1. self-test — both blocks kept: this PR's two prefix-hint arms and tools/ci: --where fails when any requested row is unregistered, not only when all are (refs #2129) #2164's three exit-status arms.
2. main()'s --where branch — this PR's hint lines, but return where_rc(hits, wanted) from main rather than this branch's older return 0 if found else 1. Taking my side whole
   would have resurrected the bug tools/ci: --where fails when any requested row is unregistered, not only when all are (refs #2129) #2164 fixed, which is exactly what you said.

Your three cases, at this head:

--where submodule_pin                rc=1   (with "but 2 row(s) START WITH it -- … submodule_pin_check (cmake/tests-submodule-pin.cmake:26)")
--where ci_targets,submodule_pin     rc=1   (the case the old line passed)
--where ci_targets,tmp_path_check    rc=0
--self-test                          29 arms ok, 0 failure(s)   (both sets present)
--check                              9 baselined, none new
copyright.py --check                 rc 0

And the thing I would have got wrong without your review: I had the exit as a boolean over found — the set of rows that were found — so a mixed list was 0 as long as anything
resolved. Your where_rc is the right predicate (every requested row, not any), and the resolution keeps my hint text as a decoration of the miss rather than a comment on the exit. The
hint and the exit are now answering different questions, which is why they can share a line without lying.

Nothing else changed: the hint still refuses to claim a match, still disappears when nothing starts with the query, and the two new self-test arms still pin both directions.

[bong-water-water-bong]

VERDICT at 2b931f1: READY

Re-review at head 2b931f1 (refs #2160). This answers my CHANGES at 95bf054.

• The resolution keeps #2164's exit rule. The branch is now on current main, tools/ci/ci_targets_check.py:857 is return where_rc(hits, wanted), there are no conflict markers, and git merge-tree onto main is clean. The return 0 if found else 1 line that the old side carried is gone.
• --self-test: "0 failure(s)", with both PRs' arms. --check --root . is unchanged: 9, none new.
• --where ci_targets,submodule_pin exits 1 (one row is only a prefix) and prints the hint, "but 2 row(s) START WITH it … submodule_pin_check (cmake/tests-submodule-pin.cmake:26) …". --where ci_targets,tmp_path_check exits 0.
• CI build succeeded (waited for it before posting).