[Arkex-ai]

Current main integration: head 81b8ff584fc0a18b2a42b592fa0483feacc89d81, base d224dc0680ad418b1ab111555c7d19cc483dcfdc. The runner migration is integrated and focused package/declaration checks pass. The current upstream texture diagnostic fields are retained; the fixture models their dimensions. Complete local suite: 278 passed, 3 failed, 20 skipped in 131 s (windows, 2 at once). Exact main d224dc0: 277 passed, 3 failed, 20 skipped in 131 s (windows, 2 at once). Same failures: nx2/test_nup_out_of_space.py, nx2/test_nup_paths.py, tests/test_precompile_compute_guard.py. LLVM-Mingw and UTF-8 explicitly pinned; initial collector failures remain preserved separately.

Fresh exact-head CI dispatches (do not infer success from this list):

• source-regressions.yml: https://github.com/Arkex-ai/Hookshot/actions/runs/42869994655
• windows-builder.yml: https://github.com/Arkex-ai/Hookshot/actions/runs/42869999908
• linux-appimage.yml: https://github.com/Arkex-ai/Hookshot/actions/runs/42870004781

Historical native coverage below stays tied to its recorded build; no new API13 native run claimed.

A transient image-view allocation failure entered the texture's cache as a null handle, so later requests never tried again even after memory became available. The actual factory/cache regression reproduces that permanent failure while its healthy reuse control passes.

This keeps host/device allocation failures retryable, limits uncached view attempts after a failure to the next texture frame, and preserves existing valid cache entries and permanent negative caching. It covers general sample/storage/attachment views, ASTC all/layer storage views, and GPU-unswizzle storage views. A successful BC single-layer compatibility fallback still wins; a failed fallback does not erase an earlier OOM result. A newly created view is rolled back if its cache entry cannot be committed.

Current revision 8e19c6af3948573eaf80fcc2cdff49ca36c84059 changes only the authored fixture: diagnostic environment reads are deterministic inside its namespace, avoiding ambient-runner dependence and native Windows CRT getenv deprecation. No compiler warnings or assertions are disabled. All48 cases and8 behavioral mutations pass LLVM/GCC/ASan/UBSan again; three exact-head fork workflows are refreshed. The complete production GPU tree matches b9be8119b9f198dd221101a90d23271a0550d57d byte for byte; build/native evidence below retains that recorded hash and is not relabeled.

Recorded validation at b9be8119b9f198dd221101a90d23271a0550d57d, based on main d8f0513f4555d3f6fa2a6d2df6fa76015ee17dbc:

• 48 production-code cases pass Windows LLVM, Linux GCC and GCC ASan/UBSan. They execute the actual factory, complete general view function, ASTC cache functions and GPU-unswizzle cache/early-return region. Vulkan capabilities/results and cache allocation failures are authored.
• Eight behavioral mutations are rejected: permanent caching of transient failures, missing throttling, throttling valid cached handles, missing either OOM class, leaking an uncommitted view, losing an earlier fallback OOM and trusting an undefined error-output handle.
• Full Windows suite: 241 passed / 2 failed / 20 skipped, versus current main 240 / 2 / 20. Both retain the same local NUP symlink-privilege and missing-Windows-SDK precompile-guard failures.
• Official API12 standalone host/backend build succeeds. Its clean official guest generation remains tied to 1a429f6; the exact main delta was audited: NUP extraction progress is outside this lift.py --exefs input, SDL polling adds no HLE declarations, and the renderer ABI is unchanged. Current host/backend are rebuilt and copied generation bytes are hash-checked. No generated game source is edited.
• Exact-head isolated default serial and parallel/batch checkpoint runs with FG2x both exit0. Each produces three CRC/zlib/RGB-validated3440x1440 captures and completes an observed1444x720 / maximized1440x829 / restored1444x720 roundtrip. Original save hashes are unchanged. Capture size is independent of current desktop/window size; no physical ultrawide-monitor coverage is inferred.

Formats, swizzles, subresources, arena/staging behavior, queue waits and the renderer ABI are unchanged by this contribution. The higher-level failed-upload generation policy remains: some uploads retry only after a storage write. This repairs the next actual view request; it does not promise immediate whole-texture or global GPU-OOM recovery. Formatted buffer views and device-loss recovery are separate contracts. No physical allocation fault, driver-validation, campaign or performance result is claimed.

Actual open-PR diffs were reviewed: #1332 BC view diagnostics/compatibility, #1505 BC scratch fallback/upload diagnostics, #1130 incremental ASTC and #1541 stage diagnostics do not provide this retry/cache-commit policy. Their separate behaviors are preserved. Main's accepted staging publication is integrated. Only source, authored tests and documentation are included; no game-derived artifacts.

Main BC retry/staging fallback changes introduced a shared source-fixture compilation regression. At 23a845afa97733be027edf07cfb8a7591d850371, the two staging-publication fixture files are integrated from fe1c22ceb699830965d75168bfe13efc9db690ef: existing assertions remain, mapped fallback/unmapped cleanup/reuse and multi-atom ASTC visibility are exercised. GPU/host/lifter/third-party source trees are identical to the previous head; earlier native results retain their actual hashes. This CI repair is tracked separately and adds no production behavior to this contribution. Exact-head fork workflows were refreshed.

Current-head validation

Current head 81b8ff5, based on main d224dc0. Earlier local/runtime results describe their recorded source hashes.

• Linux AppImage: in_progress.
• Windows builder: success.
• Game-free source regressions: failure.

Fork runs are separate from upstream Actions approval; no maintainer approval gate is bypassed.

[Arkex-ai]

Rebased onto main d224dc06 and resolved both conflicts. The texture failure log retains main's added width/height/usage details inside the existing retry callback; the authored fixture now models the dimensions used by that log. Test arguments moved into the test's RUN_ALL_ARGS per #1822. Focused retry and declaration checks pass, with all existing assertions retained.

Head 81b8ff584fc0a18b2a42b592fa0483feacc89d81. New checks are pending:

• source-regressions.yml: https://github.com/Arkex-ai/Hookshot/actions/runs/42869994655
• windows-builder.yml: https://github.com/Arkex-ai/Hookshot/actions/runs/42869999908
• linux-appimage.yml: https://github.com/Arkex-ai/Hookshot/actions/runs/42870004781

Full current-base validation remains pending; no fresh API13 game run claimed.