[GH-ISSUE #6675] [Bug] Main PTX boundary fixture expects partial coverage for pointer qualifiers #4054

New issue

Closed

opened 2026-10-31 00:46:25 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-31 00:46:25 +00:00

Owner

Originally created by @gensokyo-dev on GitHub (Oct 23, 2026).
Original GitHub issue: https://github.com/morluto/rea/issues/6675

Before submitting

Searched open and closed issues for related PTX/fixture/partial reports, including #6606.
Removed credentials and private target data.

Area

Native analysis or decompilation / CI tests

Observed behavior

On clean main 99a5ffae3bf3d73f11bca3d9b8893347964b10b9, tests/boundary/gpu/codeObjects.test.ts has 51 passing cases and one failing case: marks unknown PTX parameter declarations partial while retaining known facts. Line 440 expects partial, but the provider returns complete. The same single failure occurs with Vite 8.3.5 in #6416; main uses Vite 8.3.2.

Reproduction

With Node 24.22.1 / npm 11.16.0 on Linux x64, in a clean checkout of the main SHA above:

npm ci --ignore-scripts --include=dev --no-audit --no-fund
node node_modules/vitest/vitest.mjs run --project boundary --maxWorkers 1 --no-file-parallelism tests/boundary/gpu/codeObjects.test.ts

Executed once on clean main and once on the updated dependency tree: both produce the same 51 pass / 1 fail result. Clean-main run: 2.799 s elapsed, 2.735 s CPU, 379.6 MiB sampled process-family RSS. No real GPU execution or public CLI/MCP reproduction was performed.

Expected behavior

Keep the fixture contract aligned with supported PTX syntax, while preserving a genuinely unsupported-parameter case that verifies partial facts, retained PTX and limitations. Do not merely change the coverage assertion to make CI green.

Impact and scope

Observed CI failure in #6416 shard 1: one failure, 1,749 pass, 19 skipped. Matrix fail-fast then cancelled three other shards. This is independently reproducible on main and predates the Vite update. Product impact beyond the fixture is NOT_ENOUGH_INFO.

REA version

Source 99a5ffae3bf3d73f11bca3d9b8893347964b10b9; CI merge f6c18a9ff55cdd91daf7b1f1c29e5643a70f3504.

Runtime and platform

Node 24.22.1, npm 11.16.0, Linux x64; one CPU and one worker for the local reproduction.

Target and artifact context

Existing synthetic PTX fixture: replace .u64 output with .u64 .ptr .global output.

Interface

Other: production-provider boundary test.

Earliest failure stage

Native analysis or reconstruction / coverage assertion.

Relevant output or evidence

AssertionError: expected complete to be partial at tests/boundary/gpu/codeObjects.test.ts:440. The main and PR CI merge have identical Git objects for src/gpu, src/domain/gpu, this test file and tests/fixtures/gpu/codeObjects.ts.

Analysis or suspected cause

A stale unsupported-syntax fixture following pointer-qualifier support (#6606) is a hypothesis, not a completed diagnosis. Inspect the parser owner and retained unsupported-layout coverage before choosing the correction. This is tracked separately from #6480 / #6416; no GPU implementation or test oracle is changed by that dependency update.

Originally created by @gensokyo-dev on GitHub (Oct 23, 2026). Original GitHub issue: https://github.com/morluto/rea/issues/6675 ## Before submitting - [x] Searched open and closed issues for related PTX/fixture/partial reports, including #6606. - [x] Removed credentials and private target data. ## Area Native analysis or decompilation / CI tests ## Observed behavior On clean main `99a5ffae3bf3d73f11bca3d9b8893347964b10b9`, `tests/boundary/gpu/codeObjects.test.ts` has 51 passing cases and one failing case: `marks unknown PTX parameter declarations partial while retaining known facts`. Line 440 expects `partial`, but the provider returns `complete`. The same single failure occurs with Vite 8.3.5 in #6416; main uses Vite 8.3.2. ## Reproduction With Node 24.22.1 / npm 11.16.0 on Linux x64, in a clean checkout of the main SHA above: ```sh npm ci --ignore-scripts --include=dev --no-audit --no-fund node node_modules/vitest/vitest.mjs run --project boundary --maxWorkers 1 --no-file-parallelism tests/boundary/gpu/codeObjects.test.ts ``` Executed once on clean main and once on the updated dependency tree: both produce the same 51 pass / 1 fail result. Clean-main run: 2.799 s elapsed, 2.735 s CPU, 379.6 MiB sampled process-family RSS. No real GPU execution or public CLI/MCP reproduction was performed. ## Expected behavior Keep the fixture contract aligned with supported PTX syntax, while preserving a genuinely unsupported-parameter case that verifies partial facts, retained PTX and limitations. Do not merely change the coverage assertion to make CI green. ## Impact and scope Observed CI failure in [#6416 shard 1](https://github.com/morluto/rea/actions/runs/39946346209/job/120390449341): one failure, 1,749 pass, 19 skipped. Matrix fail-fast then cancelled three other shards. This is independently reproducible on main and predates the Vite update. Product impact beyond the fixture is NOT_ENOUGH_INFO. ## REA version Source `99a5ffae3bf3d73f11bca3d9b8893347964b10b9`; CI merge `f6c18a9ff55cdd91daf7b1f1c29e5643a70f3504`. ## Runtime and platform Node 24.22.1, npm 11.16.0, Linux x64; one CPU and one worker for the local reproduction. ## Target and artifact context Existing synthetic PTX fixture: replace `.u64 output` with `.u64 .ptr .global output`. ## Interface Other: production-provider boundary test. ## Earliest failure stage Native analysis or reconstruction / coverage assertion. ## Relevant output or evidence `AssertionError: expected complete to be partial` at `tests/boundary/gpu/codeObjects.test.ts:440`. The main and PR CI merge have identical Git objects for `src/gpu`, `src/domain/gpu`, this test file and `tests/fixtures/gpu/codeObjects.ts`. ## Analysis or suspected cause A stale unsupported-syntax fixture following pointer-qualifier support (#6606) is a hypothesis, not a completed diagnosis. Inspect the parser owner and retained unsupported-layout coverage before choosing the correction. This is tracked separately from #6480 / #6416; no GPU implementation or test oracle is changed by that dependency update.

bsakaguchi

2026-10-31 00:46:25 +00:00