Repository navigation
Conversation
…compiling at all
`src/v2/workflow/floor_route_gap.dag` does not parse on main. The seed refuses
it with `expected expression, found Newline` at byte 93168 and then, because the
parse never recovers, reports every subsequent annotation as having no subject:
343 hard diagnostics from one malformation. The module is imported by
gunbc.witness_v2_native_route, so `gunbc test //gunbc/instruments:self-host`
refuses outright -- the seed cannot emit the v2 closure at this revision.
The cause is a spelling, not a semantic change. Seven sites in
floor_route_gap_expectation_chunk_20 write a bare `tail:` with its value on the
following line; the same file writes `tail: Cons {` inline 274 times and
`tail: Empty {}` 27 times. The repair makes those seven match the other 301.
Nothing else changes: no row is added, removed or reordered, and the six
insertions against twelve deletions are whitespace joins.
Evidence, both arms with the same command
(`gunbc compile --source-root dag --source-root src/v2 --entry
src/v2/workflow/floor_route_gap.dag --target dag`):
before, 343 hard diagnostics headed by the byte-93168 refusal; after, the
refusal and its entire cascade are gone and no diagnostic names this file.
What this does NOT fix, stated so it is not read as a clean main: 26 blocking
errors remain, all of one different class -- `source annotation sits inside a
declaration body` -- across six files (host_capture_historical_binding,
target_binding, namespace_reference_derived_residency_qualification,
floor_memory_demand, mtcollins1_census_image_local_wet_test,
runner_microvm_boot_probe). Open PRs already cover three of those files and
should be widened to the rest.
Both defects landed behind a required check that concludes success while the
floor refuses, which is why neither was visible at merge time.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Closing as a duplicate: #11896 (opened 19:33, two and a half hours before this) makes the identical seven-site repair, and #11907 covers it plus the annotation refusals. Carrying the diagnostic evidence across so it is not lost with this PR. The defect, independently reproduced. The seed refuses Both arms, same command ( What remains after this file is repaired, for whoever lands #11907: 26 blocking errors, all Both defects landed behind a required check that concludes success while the floor refuses. |
1 participant
main does not compile at
87c6658641esrc/v2/workflow/floor_route_gap.dagdoes not parse. The seed refuses it withexpected expression, found Newlineat byte 93168, and because the parse never recovers it then reports every subsequent annotation as having no subject — 343 hard diagnostics from one malformation. That module is imported bygunbc.witness_v2_native_route, sogunbc test //gunbc/instruments:self-hostrefuses outright: the seed cannot emit the v2 closure at this revision.The cause is a spelling
Seven sites in
floor_route_gap_expectation_chunk_20write a baretail:with its value on the following line:The same file writes
tail: Cons {inline 274 times andtail: Empty {}27 times. This makes the seven match the other 301. No row is added, removed or reordered — the six insertions against twelve deletions are whitespace joins.chunk_20arrived in 4f04146 (#11731, today 18:25Z).Evidence — both arms, same command
gunbc compile --source-root dag --source-root src/v2 --entry src/v2/workflow/floor_route_gap.dag --target dagReproduced independently in a worktree carrying none of my other changes, which is how the defect was attributed to main rather than to a local edit.
What this does NOT fix
Stated so this is not read as a clean main. 26 blocking errors remain, all of one different class —
source annotation sits inside a declaration body— across six files:host_capture_historical_binding·target_binding·namespace_reference_derived_residency_qualification·floor_memory_demand·mtcollins1_census_image_local_wet_test·runner_microvm_boot_probeOpen PRs already cover three of those files and should be widened to the rest.
Why neither was visible at merge time
Both landed behind a required check that concludes success while the floor refuses. This is that failure mode's second confirmed instance today.
🤖 Generated with Claude Code