tests: main's red bake and ub checks skip where their tool can't run; the ub fixture builds on Windows #1939

Merged

NotRexed merged 1 commit from fix/main-tests-bake-ub into main

2026-11-10 21:31:55 +00:00

NotRexed commented

2026-11-10 21:19:49 +00:00

(Migrated from github.com)

Main's game-free checks have been red since #1707 and #1910.

Check	Where	Failure	Change
`test_bake_mimo`, `test_bake_skeleton`	Linux	`SystemExit: /usr/bin/clang++: no C++23 mode` from `bake_mimo.cxx_std`. It's raised after the tests' "no compiler" skip, so the tests error instead of skipping.	The skip check calls `bake_mimo.cxx_std(bake_mimo.cxx())`, so a compiler without C++23 skips like a missing one.
`test_bake_ds`	Windows	`Could not find module 'libzstd.so.1'`. The enumerator loads the system libzstd through ctypes, and the runners have none.	Skips when `romenum.zstd()` can't load (`OSError`).
`test_ub_word_index`	Windows	`'getenv' is deprecated` (the Windows CRT), an error under `-Werror`.	`-D_CRT_SECURE_NO_WARNINGS` on the fixture's compile.

Locally (Windows):

test_bake_ds prints its SKIP line.
test_bake_skeleton: OK, with 3 skipped.
test_bake_mimo skips.
test_ub_word_index's fixture compiles with the flag. The getenv warning only shows with CI's MSVC-target clang, so CI checks that part.

test_log_batch's "off" timing, under 200 ms against the default 250 ms, still flakes now and then on a loaded runner. It isn't changed here: there's no safe margin to widen.

Bake tests from #1707 (blessedplummer), ub test from #1910 (i-love-computers-dude).

🤖 Generated with Claude Code

Main's game-free checks have been red since #1707 and #1910. | Check | Where | Failure | Change | |---|---|---|---| | `test_bake_mimo`, `test_bake_skeleton` | Linux | `SystemExit: /usr/bin/clang++: no C++23 mode` from `bake_mimo.cxx_std`. It's raised after the tests' "no compiler" skip, so the tests error instead of skipping. | The skip check calls `bake_mimo.cxx_std(bake_mimo.cxx())`, so a compiler without C++23 skips like a missing one. | | `test_bake_ds` | Windows | `Could not find module 'libzstd.so.1'`. The enumerator loads the system libzstd through ctypes, and the runners have none. | Skips when `romenum.zstd()` can't load (`OSError`). | | `test_ub_word_index` | Windows | `'getenv' is deprecated` (the Windows CRT), an error under `-Werror`. | `-D_CRT_SECURE_NO_WARNINGS` on the fixture's compile. | **Locally (Windows):** - `test_bake_ds` prints its SKIP line. - `test_bake_skeleton`: OK, with 3 skipped. - `test_bake_mimo` skips. - `test_ub_word_index`'s fixture compiles with the flag. The getenv warning only shows with CI's MSVC-target clang, so CI checks that part. `test_log_batch`'s "off" timing, under 200 ms against the default 250 ms, still flakes now and then on a loaded runner. It isn't changed here: there's no safe margin to widen. Bake tests from #1707 (blessedplummer), ub test from #1910 (i-love-computers-dude). 🤖 Generated with [Claude Code](https://claude.com/claude-code)

Arkex-ai commented

2026-11-10 21:32:34 +00:00

(Migrated from github.com)

I finished the underlying #1931 fixes locally: github.com/Arkex-ai/Hookshot@4e9ac9e941 . Full Windows: 321 passed / 3 failed / 21 skipped versus clean main 319 / 4 / 21; DS, Mimo and skeleton actually execute and pass. Linux Clang 18 integration passes too. The fixes use the existing pinned zstandard dependency and a constexpr probe matching the real assertion failure path.

Once that lands these bake gates can execute instead of skipping on the runners. I am aligning a concurrent reader fallback with #1872 before opening the focused PR. Exact-head fork CIs are running: https://github.com/Arkex-ai/Hookshot/actions/runs/42905082651 .

I finished the underlying #1931 fixes locally: https://github.com/Arkex-ai/Hookshot/commit/4e9ac9e941650d0add34ce694647ae4f3e8fe8c8 . Full Windows: 321 passed / 3 failed / 21 skipped versus clean main 319 / 4 / 21; DS, Mimo and skeleton actually execute and pass. Linux Clang 18 integration passes too. The fixes use the existing pinned zstandard dependency and a constexpr probe matching the real assertion failure path. Once that lands these bake gates can execute instead of skipping on the runners. I am aligning a concurrent reader fallback with #1872 before opening the focused PR. Exact-head fork CIs are running: https://github.com/Arkex-ai/Hookshot/actions/runs/42905082651 .

Arkex-ai commented

2026-11-10 21:37:36 +00:00

(Migrated from github.com)

Draft #1954 now has the complete focused patch and measured validation: #1954 . The overlapping reader is called out in the description so we can settle on one implementation; discovery/cache work remains with #1872.

Draft #1954 now has the complete focused patch and measured validation: https://git.magicbean.dev/MagicBeanDev/Hookshot/pulls/1954 . The overlapping reader is called out in the description so we can settle on one implementation; discovery/cache work remains with #1872.

Commenting is not possible because the repository is archived.

No reviewers