[rooklz]

What

A recorded compile command can carry -save-temps=<value> with a value other than the stripped cwd and obj modes. The syntax check passes such a joined form to the compiler, assuming an unknown mode is rejected there (6229cd07eb made the test skip whenever a compiler accepts unknown modes). Newer Clang builds do accept them: any value is taken as an output prefix, so on Apple clang 21.0.0 the command

$ clang -fsyntax-only -save-temps=invalid t.cpp ; echo rc=$? ; ls
rc=0
t.cpp  t.ii

exits 0 and writes t.ii even under -fsyntax-only (with -c it also writes t.bc, t.s, t.o). GCC instead fails on the same option, so the outcome of the check is toolchain-dependent: a silent file write on one compiler, a failure on another.

Reject -save-temps= and --save-temps= values other than the four already-stripped forms before the compiler runs, like the forwarded-action and -Wp rejections: the rewritten test no longer probes the compiler and no longer skips, and asserts the mode is reported, the compiler is never invoked and no output appears. The new scope is recorded in docs/dev/TechnicalDebt.md.

The rejection is placed after the -Wp, handling with an explicit not in DROP guard, so it does not touch the option-stripping block that #11522 extends; both changes merge in either order.

Tested

macOS 26.2, Apple M5 (arm64), Apple clang 21.0.0 (clang-2100.1.1.101), CMake/Ninja, ANYPS5_RELINKER_ONLY=ON, BUILD_TESTING=ON (Python 3.13). Base: 801d5b9bc1 (upstream/main), tested head: fe80227fce.

• Measured on the host compiler: -fsyntax-only -save-temps=invalid and -save-temps=diag exit 0 and write t.ii (shown above); joined -o forms write nothing, so they stay untouched.
• Rewritten test test_unmodeled_saved_temporary_modes_are_rejected_before_the_compiler_runs: five modes (invalid, diag in both spellings, CWD, empty) each fail with the new usage error before the compiler runs; it needs no compiler and skips nowhere. The stripped cwd/obj coverage in test_saved_temporaries_do_not_rewrite_files_or_hide_compiler_errors is unchanged and passes.
• Direct: tools/tests/test_check_compile.py 46 tests OK (2 skipped: Windows/GCC-only), test_check_compile_modes.py 31 OK (1 skip).
• Complete relinker-only build passed. Full CTest: 361/361 tests passed (ctest -j 8).
• python3 tools/check_conventions.py --base upstream/main: all conventions checks passed. python3 tools/pr_overlap.py --branch fix/check-compile-save-temps-modes-v2: no conflicting pull requests; the unresolved shared-file notes are the ancient open PRs fix(agc): color compare emulation: review follow-up to #668 #907/feat(agc): retain and refresh sampled depth array storage #1368/feat(ngs2): add experimental I3DL2 reverb #1384/feat(agc): add experimental depth clear and copy recognition #1395/fix(avplayer): keep handed-out video frames until later frames are handed out #1601/feat(aot): discuss native ARM64 compiler architecture and integration #1835/fix(libkernel): keep the signal mask per thread #3986/feat(aot): consolidate native ARM64 compiler, runtime and consumers #5021/docs: fix Pages workflow link and separate library technical debt #6039, which touch other entries of docs/dev/TechnicalDebt.md.

Host-only change to the Python tool; no title, GPU or console involved. No performance claim.

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: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

[MarcZX10]

I reviewed exact head fe80227fce02056fefbcbacaaa9aaec4cf823517. The new guard rejects unsupported joined -save-temps= / --save-temps= values before invoking the compiler, while the exact supported cwd and obj forms remain handled by DROP. The regression checks both spellings and several invalid values with a compiler marker to prove no compiler invocation or output file occurs.

I ran python3 -m unittest tools.tests.test_check_compile tools.tests.test_check_compile_modes from this head: 77 tests passed, 14 skipped. The new regression ran and passed. The open #11522 change also edits syntax_arguments; a three-way merge preview against its current head is clean, and the edits apply at separate locations.

Hosted policy checks and ASan pass. Linux, Windows and both macOS relinker jobs are still pending; the full 361-case hosted suite has not completed yet. I found no actionable code issue in this review, but I am leaving this as a comment until the platform jobs finish.

[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 907	docs/dev/TechnicalDebt.md
PR 1368	docs/dev/TechnicalDebt.md
PR 1384	docs/dev/TechnicalDebt.md
PR 1395	docs/dev/TechnicalDebt.md
PR 1601	docs/dev/TechnicalDebt.md
PR 1835	docs/dev/TechnicalDebt.md
PR 3986	docs/dev/TechnicalDebt.md
PR 5021	docs/dev/TechnicalDebt.md

[MarcZX10]

The previously pending Ubuntu, Windows, both macOS relinker jobs, and ASan checks have now completed successfully. The 77 focused Python tests (14 skipped) passed in the reviewed head, including the new regression; the merge preview against #11522 was clean. I found no actionable issue in the exact reviewed diff. Product-level Linux/Windows jobs are skipped by repository policy, so this approval covers the tooling change and checks listed above.

[babyhuey]

Windows run of this head fe80227fce0, merged onto main 8ff4baf5348 (clean merge, local only), with the selection CI's skipped Windows product job uses: Windows 11, Ryzen 7 5800X, RTX 3080, WinLibs GCC 15.2, Release, BUILD_TESTING=ON. The tree was configured fresh after the merge, then select_tests.py select / ninja -f select.ninja selected / --target libs / select_tests.py ctest --parallel 4 --timeout 120.

• Selected 435 tests: 429 pass, 0 fail, 6 skip. No failures; the new check_compile_test_unmodeled_saved_temporary_modes_are_rejected_before_the_compiler_runs passes.

AI-assisted: yes (Claude Code)