60 of 470 tracked C/C++ sources are outside clang-tidy-diff's reach, and 11 of them only because one CMake flag is missing

[Yaraslaut]

60 of 470 tracked C/C++ sources are outside clang-tidy-diff's reach, and 11 of them only because one CMake flag is missing

What this is

#481 widened the clang-tidy job's configure so examples/ would appear in
compile_commands.json at all, and deleted the -regex that used to exclude
it. PR #649 then added a filter to that job: a changed source with no entry
in the compile database is dropped from the diff before clang-tidy-diff.py
sees it, because clang tooling interpolates a neighbouring entry's compile
command and the result is a clang-diagnostic-error about the configure
rather than a finding about the changed line. Each drop is printed as a
::warning::.

That makes the hole visible instead of fatal. It does not close it. This issue
records its size and separates the part that is cheap to fix from the part
that is not.

Measurement

On 3c6dd24f (PR #649's head), with the clang-tidy job's own configure
flags, clang 22.1.8, Qt 6 present:

$ # every tracked C/C++ source vs. build/clang-debug/compile_commands.json
tracked C/C++ sources: 470
not in compile_commands.json: 60
    30  tests/lint
     9  examples/bank/gui
     9  examples/bank/gui_wasm
     6  tests/compile_checks
     2  examples/bank/tests
     1  examples/bookmarks/gui_wasm
     1  examples/common/wasm_spike
     1  examples/pastebin/gui_wasm
     1  examples/polls/gui_wasm

The database itself: 703 entries naming 695 distinct in-workspace sources, 270
of them under examples/.

The three groups are not the same problem

1. examples/bank/gui (9) and examples/bank/tests (2) — a missing flag,
nothing more. The clang-tidy job passes -DMORPH_BUILD_BANK_EXAMPLE=ON
but not -DMORPH_BUILD_BANK_GUI=ON. examples/bank/CMakeLists.txt:159 gates
the GUI sources on the latter, and examples/bank/CMakeLists.txt:198 says the
bank controllers "only exist when MORPH_BUILD_BANK_GUI=ON", which is why the
two examples/bank/tests sources go with them. The linux-all-features job
already sets it (.github/workflows/ci.yml:1855), so these sources do build
on a native Linux Qt configure — they are simply not offered to the one job
that reads code. Adding the flag is a one-line change; it costs a wider
configure (more Qt Quick), which is why it is a decision and not a drive-by.

2. tests/lint (30) and tests/compile_checks (6) — not translation units
by design. tests/lint/** are text fixtures scanned by shell checkers;
their own headers say "Nothing here is compiled" — that is #650, and PR #649's
filter changes that issue's option space (see the comment there).
tests/compile_checks/** are built by configure-time try_compile() /
try_run() with specific COMPILE_DEFINITIONS (-DMORPH_CLIENT_ONLY and
friends); analysing them under the ordinary flag set is meaningless, as
client_only_facade_no_model_header.cpp demonstrated by producing 17
clang-diagnostic-errors attributed upward into bridge.hpp, registry.hpp
and model.hpp.

3. The 13 WASM sources — need a toolchain this job does not have. Four
*/gui_wasm/main_wasm.cpp, examples/common/wasm_spike/main_wasm.cpp, and
nine under examples/bank/gui_wasm. wasm-ladder.yml and wasm-demo.yml
build them with Emscripten; neither runs clang-tidy (grep -l clang-tidy .github/workflows/*.yml matches only ci.yml and drift-guard.yml). So no
lint check of any kind has ever read a line of these files. Making the
clang-tidy job configure an Emscripten toolchain is a large change; running
clang-tidy inside the existing WASM legs may be cheaper, since they already
have the toolchain — but neither has been costed.

Verification status

Reproduced for the measurement above: the counts come from an actual local
configure with the job's flags, not from reading CMake. Read, not
reproduced: the claim that adding -DMORPH_BUILD_BANK_GUI=ON brings the 11
bank sources into the database is inferred from
examples/bank/CMakeLists.txt:25,159,198 and from linux-all-features
setting the flag — I did not re-run the configure with it, and did not measure
what it costs in configure time. Not attempted: any clang-tidy run over
the WASM sources.

Also measured, and clean: bugprone-exception-escape over the three
compiled production sources (src/qt/qt_websocket_backend.cpp,
src/qt/qt_websocket_server.cpp, src/main.cpp) with
-checks='-*,bugprone-exception-escape' reports nothing. That probe was run
because PR #649 fixed a real instance of that check in a test double whose
destructor settles a Completion, and the same shape would be a
std::terminate in production; it does not appear there.

What would change the verdict

Close this when either (a) the bank GUI sources are in the clang-tidy job's
database, or the decision not to put them there is written down with its cost;
and (b) the WASM sources are lint-analysed somewhere, or the decision not to
is written down. Re-open — or rather, re-measure — if the "not in
compile_commands.json" count rises above 60 for a reason other than a new
tests/lint/ fixture directory.

Related

• clang-tidy-diff's configure omits the application ladder, so no rung is in the compile database it lints against #481 — why the job configures the ladder, and why it carries no -regex.
• clang-tidy-diff fails with a clang-diagnostic-error on any edit to the two sources that self-include their .moc #624 — the same structural problem with a build-time generated header.
• tests/lint/ fixtures are inside clang-tidy-diff's scope, and nothing tells the author that until CI is red #650 — the same structural problem with tests/lint/ fixtures.
• core/ladder: retire IBackend's four async twins and move the examples' prose onto the structural surface (fixes #570, fixes #571) #649 — the filter that makes these drops visible rather than fatal.

[Yaraslaut]

Triage: valid — and the 11-source part is confirmed with a one-flag cause

Measured on 4563aff3.

The tree size is right: git ls-files '*.cpp' '*.cc' '*.cxx' '*.c' → 473 (the issue says 470; it was filed two merges ago, and nothing since added sources — close enough that the split, not the total, is the load-bearing part).

The 11 bank-GUI sources are out of reach for exactly the reason stated, and the fix is one flag. Comparing the two jobs' configures:

clang-tidy:          BANK_EXAMPLE FORMS_QML FUZZERS HMAC_EXAMPLES LADDER
                     LOAD_TESTS NET OFFLINE_SQLITE QT
linux-all-features:  BANK_EXAMPLE BANK_GUI FORMS_QML HMAC_EXAMPLES LADDER
                     LOAD_TESTS NET OFFLINE_SQLITE QT TESTS

MORPH_BUILD_BANK_GUI is declared OFF by default (CMakeLists.txt:28), linux-all-features turns it on, and the clang-tidy job does not. So those 11 sources are unlinted not by design but by omission — and another job already proves the option configures and builds on the same runner image, which removes the usual objection.

The "no lint job anywhere" claim for the WASM sources holds structurally:

$ grep -l "clang-tidy" .github/workflows/*.yml
.github/workflows/ci.yml
.github/workflows/drift-guard.yml

Neither is a WASM job, so nothing lints the 13.

triage: valid, enhancement, area: ci.

The split is what makes this actionable, and it should not be collapsed

Three causes with three different answers, and treating them as one number ("60 unlinted") would hide that:

1. The 11 — fix by passing the flag. Cheap and clearly right.
2. The 36 that are not translation units by design — nothing to fix; they should be recorded as out of scope so the count stops looking like debt.
3. The 13 WASM — a real gap needing a job that does not exist, i.e. the expensive one, and the only one worth arguing about.

Expect (1) to surface findings the way #646 did. Adding a configure flag makes 11 previously-unanalysed sources visible to clang-tidy-diff the moment anyone edits them. #646 is the precedent: 84 findings appeared in two files when #647 made them parse, and clearing them was its own ticket. Whoever takes this should measure the 11 files' findings before landing the flag and decide deliberately whether to clear them in the same PR or file them — not discover it from a red leg on someone else's unrelated change.

Also worth knowing: #649's filter now drops changed sources absent from the compile database, with floors of 600 in-workspace / 200 under examples/. Adding BANK_GUI only grows the database, so the floors are unaffected — but the filter is why these 60 are currently silent rather than red, which is a change since this issue was filed.

Not verified by me: the 60/36/13 breakdown itself, and the filer labels the 11-source claim as read from CMake rather than from a re-configure. The flag difference above is measured; the per-file counts are not.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VptDWG2fKr2vBnLSJcgzgW