fix(ci): remove stale test_pic_preallocation sanitizer deselect

.github/workflows/sanitizers.yml

@@ -127,12 +127,13 @@ jobs:
           # Exclude test_integer_motion_v2_coverage: triggers a SIGABRT under
           # ASan due to an intentional huge-allocation pattern in the motion
           # coverage path that the ASan allocator cannot satisfy.
-          # Exclude test_pic_preallocation: triggers a SIGABRT under ASan in
-          # addition to its pre-existing Linux exclusion; the test relies on
-          # allocation semantics that ASan's allocator intercepts fatally.
+          # test_pic_preallocation was removed from this list after PRs #765
+          # and #769 fixed the PREV_REF refcount leak and the dict leak in
+          # flush_context_threaded; all 8 sub-tests now pass under ASan+LSan,
+          # UBSan, and TSan (verified 2026-06-06).
           TESTS=$(meson introspect build --tests 2>&1 \
                   | python3 -c "import json,sys; [print('libvmaf:'+t['name']) for t in json.load(sys.stdin)]" \
-                  | grep -vE 'test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$|test_pic_preallocation$' || true)
+                  | grep -vE 'test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$' || true)
           echo "test count: $(echo "$TESTS" | wc -l)"
           if [ -z "$TESTS" ]; then
             echo "::error::Test enumeration produced empty list" >&2
@@ -203,11 +204,13 @@ jobs:
           # T-GPU-POOL-UAF-OOM-TSAN-ABORT.
           # Exclude test_integer_motion_v2_coverage: intentional huge-allocation
           # pattern triggers SIGABRT under the TSan allocator.
-          # Exclude test_pic_preallocation: SIGABRT under TSan for the same
-          # intentional-huge-alloc reason as the ASan exclusion.
+          # test_pic_preallocation was removed from this list after PRs #765
+          # and #769 fixed the PREV_REF refcount leak and the dict leak in
+          # flush_context_threaded; all 8 sub-tests now pass under ASan+LSan,
+          # UBSan, and TSan (verified 2026-06-06).
           TESTS=$(meson introspect build --tests 2>&1 \
                   | python3 -c "import json,sys; [print('libvmaf:'+t['name']) for t in json.load(sys.stdin)]" \
-                  | grep -vE 'test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$|test_pic_preallocation$' || true)
+                  | grep -vE 'test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$' || true)
           echo "test count: $(echo "$TESTS" | wc -l)"
           if [ -z "$TESTS" ]; then
             echo "::error::Test enumeration produced empty list" >&2

.github/workflows/tests-and-quality-gates.yml

@@ -305,10 +305,11 @@ jobs:
               # allocator aborts with SIGABRT instead of returning NULL.
               # Exclude test_integer_motion_v2_coverage: intentional
               # huge-allocation pattern triggers SIGABRT under ASan.
-              # Exclude test_pic_preallocation: SIGABRT under ASan from
-              # intentional huge-alloc; OOM path covered by unsanitized suite.
-              # (PR #767 added these to sanitizers.yml but missed this job.)
-              EXCLUDE='test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$|test_pic_preallocation$'
+              # test_pic_preallocation was removed from this list after PRs #765
+              # and #769 fixed the PREV_REF refcount leak and the dict leak in
+              # flush_context_threaded; all 8 sub-tests now pass under ASan+LSan
+              # (verified 2026-06-06).
+              EXCLUDE='test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$'
               ;;
             undefined)
               # UBSan deselects: test_model (NULL-to-nonnull memcpy
@@ -321,10 +322,13 @@ jobs:
               # eliminating the SIGILL on non-AVX2 CI runners; confirmed
               # clean under UBSan on current master.
               # Exclude test_y4m_alloc_failure: RLIMIT_AS incompatible with UBSan runtime allocation.
-              # Exclude test_gpu_picture_pool_uaf, test_integer_motion_v2_coverage,
-              # test_pic_preallocation: same intentional huge-alloc / SIGABRT
-              # reason as ASan exclusions above. (PR #767 gap — see sanitizers.yml.)
-              EXCLUDE='test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$|test_pic_preallocation$'
+              # Exclude test_gpu_picture_pool_uaf, test_integer_motion_v2_coverage:
+              # same intentional huge-alloc / SIGABRT reason as ASan exclusions above.
+              # test_pic_preallocation was removed from this list after PRs #765 and
+              # #769 fixed the PREV_REF refcount leak and the dict leak in
+              # flush_context_threaded; all 8 sub-tests now pass under UBSan
+              # (verified 2026-06-06).
+              EXCLUDE='test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$'
               ;;
             thread)
               # TSan deselects: test_model (same defect as ASan/UBSan
@@ -334,10 +338,13 @@ jobs:
               # mutex-domain mismatch; nightly TSan job was green on
               # 2026-05-09 and 2026-05-10 (state.md entry).
               # Exclude test_y4m_alloc_failure: RLIMIT_AS incompatible with TSan internal allocator.
-              # Exclude test_gpu_picture_pool_uaf, test_integer_motion_v2_coverage,
-              # test_pic_preallocation: intentional huge-alloc / SIGABRT under
-              # TSan allocator. (PR #767 gap — see sanitizers.yml and TSan job there.)
-              EXCLUDE='test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$|test_pic_preallocation$'
+              # Exclude test_gpu_picture_pool_uaf, test_integer_motion_v2_coverage:
+              # intentional huge-alloc / SIGABRT under TSan allocator.
+              # test_pic_preallocation was removed from this list after PRs #765 and
+              # #769 fixed the PREV_REF refcount leak and the dict leak in
+              # flush_context_threaded; all 8 sub-tests now pass under TSan
+              # (verified 2026-06-06).
+              EXCLUDE='test_model$|test_y4m_alloc_failure$|test_gpu_picture_pool_uaf$|test_integer_motion_v2_coverage$'
               ;;
             *)
               echo "::error::ADR-0347 deselect list is not defined for sanitizer=${{ matrix.sanitizer }}" >&2

docs/state.md

@@ -548,7 +548,7 @@ _Bugs closed in the last ~90 days. Older entries roll off into
 | **T-SANITIZER-CLI-PARSE-CLEAN — `test_cli_parse` no longer needs sanitizer deselect** — the test was bundled into the broad T-SANITIZER-DEFECTS-REVEALED-758 exclusion after PR #758 made the sanitizer matrix enumerate the real C test set. Current master no longer reproduces the recorded `test_backend_cpu` non-zero sanitizer exit: the full `test_cli_parse` binary passes cleanly under ASan+LSan, UBSan, and TSan. | this PR (`fix/sanitizer-cli-parse-deselect-2026-05-15`) | — (stale sanitizer deselect cleanup; no ADR per CLAUDE §12 r8) | `ASAN_OPTIONS=detect_leaks=1:halt_on_error=1:abort_on_error=1:print_summary=1 ./build-asan-cli/test/test_cli_parse`, `UBSAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1:print_stacktrace=1 ./build-ubsan-cli/test/test_cli_parse`, and `TSAN_OPTIONS=halt_on_error=1 ./build-tsan-cli/test/test_cli_parse` all pass. The workflow removes `test_cli_parse` from the ASan / UBSan / TSan `EXCLUDE` regexes; `test_predict` is retired by the companion cleanup. |
 | **T-SANITIZER-PREDICT-CLEAN — `test_predict` no longer needs sanitizer deselect** — the test was bundled into the broad T-SANITIZER-DEFECTS-REVEALED-758 exclusion after PR #758 made the sanitizer matrix enumerate the real C test set. Current master no longer reproduces the recorded UBSan / TSan findings: the full `test_predict` binary passes cleanly under ASan+LSan, UBSan, and TSan. | this PR (`fix/sanitizer-predict-deselect-2026-05-15`) | — (stale sanitizer deselect cleanup; no ADR per CLAUDE §12 r8) | `ASAN_OPTIONS=detect_leaks=1:halt_on_error=1:abort_on_error=1:print_summary=1 ./build-asan-predict/test/test_predict`, `UBSAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1:print_stacktrace=1 ./build-ubsan-predict/test/test_predict`, and `TSAN_OPTIONS=halt_on_error=1 ./build-tsan-predict/test/test_predict` all pass. The workflow removes `test_predict` from the ASan / UBSan / TSan `EXCLUDE` regexes; `test_cli_parse` was retired by the companion cleanup. |
 | **T-VK-VIF-1.4-RESIDUAL-NVIDIA-DEFERRED — Vulkan VIF API-1.4 residual on NVIDIA RTX 4090 + driver 595.71.05** — Phase 3b left `integer_vif_scale2` failing 45/48 frames at max abs `1.527e-02` and 5-run non-deterministic int64 accumulator magnitudes after stronger fences failed. Phase 3c replaces the seven `subgroupAdd(int64_t)` accumulator reductions in `vif.comp` with an explicit `subgroupShuffleXor` butterfly helper, avoiding the NVIDIA int64 subgroup-add lowering path. | PR #787 (`fix/vulkan-vif-int64-subgroup-reduction`) | [ADR-0269](adr/0269-vif-ciede-precise-step-a.md) Phase-3c status update; [research-0108](research/0108-vulkan-vif-int64-subgroup-reduction-2026-05-14.md) | `glslc --target-env=vulkan1.3 -O core/src/feature/vulkan/shaders/vif.comp -o /tmp/vif.spv`; `ninja -C build-vulkan-int64 tools/vmaf`; cross-backend VIF gate at `places=4`: NVIDIA device 0 0/48 on all scales (scale2 max `2.000000e-06`, repeated 5 times), Arc device 1 0/48, RADV device 2 0/48. |
-| **T-SANITIZER-PIC-PREALLOCATION-CLEAN — `test_pic_preallocation` no longer needs sanitizer deselect** — the test was bundled into the broad T-SANITIZER-DEFECTS-REVEALED-758 exclusion after PR #758 made the sanitizer matrix enumerate the real C test set. Current master no longer reproduces the original LSan / TSan signatures: the test passes cleanly under ASan+LSan, UBSan, and TSan. | this PR (`fix/sanitizer-pic-preallocation-2026-05-14`) | — (stale sanitizer deselect cleanup; no ADR per CLAUDE §12 r8) | `ASAN_OPTIONS=detect_leaks=1:halt_on_error=1 ./build-asan-score/test/test_pic_preallocation`, `UBSAN_OPTIONS=halt_on_error=1:print_stacktrace=1 ./build-ubsan-score/test/test_pic_preallocation`, and `TSAN_OPTIONS=halt_on_error=1 ./build-tsan-score/test/test_pic_preallocation` all pass. The workflow removes only `test_pic_preallocation` from the ASan / UBSan / TSan `EXCLUDE` regexes; `test_cli_parse` and `test_predict` remain tracked. |
+| **T-SANITIZER-PIC-PREALLOCATION-CLEAN — `test_pic_preallocation` no longer needs sanitizer deselect** — the test was bundled into the broad T-SANITIZER-DEFECTS-REVEALED-758 exclusion after PR #758 made the sanitizer matrix enumerate the real C test set. PRs #765 (PREV_REF refcount leak in batch func) and #769 (dict leak via is_initialized guard in flush_context_threaded) together fixed all underlying bugs. All 8 sub-tests now pass cleanly under ASan+LSan, UBSan, and TSan. | DONE — PR this branch (`fix/dev-container-cuda-passthrough`, 2026-06-06). `test_pic_preallocation` removed from all 5 EXCLUDE regexes across `.github/workflows/sanitizers.yml` (2 lanes) and `.github/workflows/tests-and-quality-gates.yml` (3 lanes). | — (stale sanitizer deselect cleanup; no ADR per CLAUDE §12 r8) | 8/8 sub-tests pass: `ASAN_OPTIONS=detect_leaks=1:halt_on_error=1:abort_on_error=1 ./core/build-asan/test/test_pic_preallocation`, `UBSAN_OPTIONS=halt_on_error=1:print_stacktrace=1 ./core/build-pic/test/test_pic_preallocation`, `TSAN_OPTIONS=halt_on_error=1 ./core/build-tsan/test/test_pic_preallocation`. GCC debug build: 8/8. 20 consecutive GCC runs: all pass. |
 | **SAN-FLOAT-MS-SSIM-MIN-DIM-LEAK** — `test_float_ms_ssim_min_dim::invoke_init` was excluded from the ASan deselect list based on a reported 240-byte / 6-allocation leak. Re-verification under `ASAN_OPTIONS=detect_leaks=1` (2026-05-13) shows zero leaks: `invoke_init` already calls `fex->close(fex)` + `free(priv)` on every code path (both early-reject and success). The exclusion was never needed after the teardown was added to the test body. | `ASAN_OPTIONS=detect_leaks=1 ./build-asan-test/test/test_float_ms_ssim_min_dim` → `3 tests run, 3 passed`, no leak report. | — | Removed `test_float_ms_ssim_min_dim$` from `EXCLUDE=` in `.github/workflows/tests-and-quality-gates.yml` (ASan lane). |
 | **T-VMAFTUNE-RECOMMEND-FROM-CORPUS-FILTER — `vmaf-tune recommend --from-corpus` bypassed the library row filter** — The programmatic `recommend()` API dropped rows with `exit_status != 0`, missing / non-finite `vmaf_score`, and non-matching `encoder` / `preset`, but the CLI `--from-corpus` path called `pick_target_vmaf` / `pick_target_bitrate` directly. A failed encode row with high VMAF, a `NaN` score row, or a row for a different encoder could therefore win from the CLI even though the library API rejected it. | PR #781 (`fix/backlog-gap-pass-7-2026-05-14`) | — (bug fix; no ADR per CLAUDE §12 r8 — only-one-way fix) | `_run_recommend_from_corpus` now builds a `RecommendRequest` and delegates to `recommend()`. Regression tests cover failed-row filtering, `NaN` filtering, and encoder filtering. Smoke: `PYTHONPATH=tools/vmaf-tune/src .venv/bin/python -m pytest tools/vmaf-tune/tests/test_recommend.py -q` — 20/20 passed. |
 | **T8-1b — Metal (Apple Silicon) backend runtime ([ADR-0420](adr/0420-metal-backend-runtime-t8-1b.md))** — replaces the T8-1 scaffold's `-ENOSYS` C stubs with three Obj-C++ `.mm` TUs (`common.mm`, `picture_metal.mm`, `kernel_template.mm`) driving `Metal.framework` via Obj-C++ ARC. `MTLCreateSystemDefaultDevice` / `MTLCopyAllDevices`, Apple-Family-7 gate, `MTLResourceStorageModeShared` zero-copy buffers, private MTLCommandQueue + two MTLSharedEvent handles per consumer. | PR #764 (`feat/metal-runtime-t8-1b`, 2026-05-11). | [ADR-0420](adr/0420-metal-backend-runtime-t8-1b.md) | macOS Metal CI lane green; subsequent PRs #765 (kernels T8-1d–j), #766 (CLI selectors, ADR-0422), #767 (IOSurface zero-copy import, ADR-0423) layered on top. Homebrew tap flipped from MoltenVK to native Metal ([formula](https://github.com/lusoris/homebrew-tap/blob/master/Formula/libvmaf.rb)). |