ReviewOS

home-lang/home

Port Bun shell broken-pipe pipeline lifecycle

#692
Closed chris opened this 9 days ago ·
9 days ago

Follow-up from #689 and the exact Bun shell fixture at checkout 4982b91.

Remaining unchanged-suite failure:

  • deno_task > pipeline > broken pipe subproc

Port Bun pipeline and broken-pipe lifecycle semantics logically, including correct closure and exit behavior for subprocess stages. Do not change the test, deadline, workload, or assertion.

Strict baseline after e83b67b: 369 pass, 83 upstream TODO, 9 fail across 461 tests. Related trackers: #66 and #202.

9 days ago

After 0bc37d2, this is now the sole failure in the unchanged Bun shell fixture. Two consecutive full runs produced 377 pass / 83 upstream todo / 1 fail, with only deno_task > pipeline > broken pipe subproc remaining. Fixture remains byte-identical to Bun checkout 4982b91.

9 days ago

Resolved on main by f575a8d. Corrected diagnosis: the pipeline implementation already handled the closed reader, but the mirrored Bun repository root did not expose the unchanged command path src/js_parser/parser.zig, so grep failed with ENOENT before it could exercise EPIPE. The native corpus launcher now runs from packages/runtime, where the real Home src tree lives, explicitly loads the mirrored Bun root config at test/bunfig.toml, and rewrites corpus targets beneath test/test. No fixture, assertion, stderr suppression, or filesystem shim was added. The original root-level command passed twice at 378 pass / 83 upstream todo / 0 fail / 1035 assertions across 461 tests; the fixture is byte-identical to Bun 4982b91.

Sign in to comment on this issue.