Software GL packed points/lines and ortho fog — 2026-11-01

Both original, unchanged entrypoints passed against actual instantiated WAT software-rendered pixels:

Fixture Result Elapsed
test/test-gl-software-lines.js PASS, exit 0 3.498 s
test/test-gl-software-ortho-fog.js PASS, exit 0 3.588 s

Lines coverage includes packed points and lines, point size and invalid-size handling, horizontal/reversed/vertical/diagonal/shallow/degenerate lines, endpoint exclusion, line width and attribute restoration, smooth interpolation, flat provoking-vertex colour, additive strip joints, near-plane shortening and rejection, line depth testing, and double-valued clear depth. Fog coverage uses the production encoder under glOrtho: eye z=0 remained red [255,0,0], while z=-0.9875 produced the expected halfway blend [159,32,32].

No assertion failed and neither selected invocation crashed or trapped. These results qualify only these fixtures; they do not establish actual-game, GPU, full27, or general crash-free qualification. Accepted mipmap8/8 and texgen9/9 at ea9b18ae9 were not rerun. Held algorithms were not exercised separately.

Gate and identities

Evidence run: scratch/runs/20261101-gl-software-lines-fog/ (ignored). The bound predecessor was redline-cutscene-skip-movement-nov01, worker 01a195e4-a16f-74d0-b993-e84bdcca3ecc. Its fresh worker final/result files (06:59:30 UTC) and explicit 06:59:26 runtime/source/build RELEASE were inspected before lane acquisition and the 06:59:49 UTC source snapshot. Current main was a6dba873df94fb2b0c8a59d690cdddfcaa7555e8. Preparation before the gate was scratch only. Searching 2,735 existing run-result/verification/note files found no accepted result for either exact selected fixture.

The accepted locked Tachyon build, also accepted by the texture run, was reused: all 4,390 accepted source hashes and 122 accepted actual-read hashes matched, along with the module and generated mirror. An additional 51 foreign untracked source-tree files were pinned separately and left untouched. All 4,441 pinned files remained unchanged. No build invocation was necessary.

Execution and cleanup

The original entrypoints ran sequentially on local Node v24.18.1 under timeout --signal=TERM --kill-after=5s 120s. The existing forwarded read-only observer was copied byte-for-byte from the accepted texture run. It records file reads and actual WebAssembly instantiation bytes without replacing WAT exports or assertions. Both fixtures recorded 88 unchanged actual reads. The unchanged render helper compiles the canonical WAT closure and instantiates shared 512 MB memory; assertions inspect software DIB pixels directly.

The run contains compressed original output, both actual modules, loaded-read identities, before/after source manifests, accepted build-log copy, gate files, prior search, test results, verification, cleanup, publication, final.txt and result.json. No source, test, compiler, or repository helper was edited. Own test processes/groups and watcher were checked absent; no temporary files or box were owned (box ID/state/expiry N/A). Free disk remained above 2 GB. No further runtime work is required for this coverage assignment.