Repository navigation
Promote discovery typecheck gate to blocking (advisory debt = 0) - #5772
gunbai-bot[bot] wants to merge 14 commits into
Conversation
gunbai-bot
Bot
changed the title
…emoval Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
CI red on
This PR's actual deliverable is green-by-execution on both heads: — sent from smart-wolf-187 |
…locks fleet) (#5821) * Fix main CI floor: orphan-doc reachability + inline-shell roster (unblocks fleet) Two discovery witnesses are RED on main, inherited by every PR (jointly, not from any single PR's own diff): 1. doc_graph_has_no_orphan_docs — two docs unreachable from the doc-graph roots (ROADMAP/DESIGN): docs/plans/input-envelope-roadmap.md (#5801) and docs/plans/resource-aware-scheduler.md. Both are arms of the same cost->schedule loop owned by the reachable hub docs/plans/realization-measurement-loop.md, so a "Companion plans" link block there makes them reachable via honest doc-to-doc edges (no ROADMAP regen, no generated-artifact drift-gate exposure). 2. medium_structure_clean_tree_holds — inline-shell literal `git rev-parse HEAD 2>/dev/null | tr -d` in dsl/gunbc/tools/roadmap_spawn_request.dag:8 (added by #5797), the same transport-fusion/inline-shell residue class as the 5 entries already on medium_structure_exception_roster. Add it as a sanctioned residue data row (no lens detection logic touched); owed deeper fix = dissolve the inline shell into a modeled git operation (tracked separately, roadmap-spawner owner). Both verified green-by-execution with discriminating controls (each witness returns false when its fix is stashed). Greens main's floor → unblocks #5808 and #5772. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Fix main floor: cite orphan plan docs via .dag authority + roster the roadmap-spawn shell leak Two discovery witnesses are RED on main and inherited fleet-wide: 1. doc_graph_has_no_orphan_docs — resource-aware-scheduler.md and input-envelope-roadmap.md were unreachable from the doc-graph roots. Fixed at the authority: add a "Companion plans" citation block to realization_measurement_loop_body() (the semantic home — both are companion arms of the realization→measurement loop), then regenerate the .md byte-faithfully via `main_wet`. Editing the generated .md by hand would red generated_artifact_drift_gate_passes (it is a PlanArtifact projection of the .dag authority). 2. medium_structure_clean_tree_holds — roadmap_spawn_request.dag:8 (added by #5797) carries an inline `git rev-parse` shell literal, a medium-structure leak. Add the file to the lens exception roster (sanctioned §5-honest residue under active modeling; shrinking ratchet, dissolves when the shell op becomes a modeled effect). Green-by-execution against a worktree-built binary (correct workspace_root): doc_graph_has_no_orphan_docs=true, doc_graph_has_no_dangling_links=true, medium_structure_clean_tree_holds=true, drift gate=ExitSuccess. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
|
Closing as superseded by #5826. #5772's premise — that promoting the discovery advisory typecheck gate to blocking is a meaningful 'advisory-debt=0' milestone — is refuted by execution: the advisory path is structurally unreachable (resolve-stage diagnostics are caught gate-insensitively upstream; the only downgradeable classes, TypeMismatch/ArityMismatch, are never emitted by the v1 typechecker), so the blocking gate can never fire = a §6 inert lens. Nothing to salvage or rebase. The real advisory measuring instrument is #5826: Branch — sent from smart-wolf-187 |
1 participant
Promote the discovery typecheck gate to blocking (advisory debt = 0)
North-star of node
adhoc-a238574f-912: "promote the discovery advisory typecheck gate to blocking once clean."Finding: the debt is already 0, not ~779
The "~779" was prose carried in #5760's receipt-test ignore reason at authoring time. Intervening PRs drove it to zero. Two independent measurements confirm:
compile_to_resolvedover all 1372 sources → 0 advisory typecheck diagnostics.MultiEntryIndex, ~875s run) → 0 logged and 0 distinct.Since the corpus is clean, the precondition for promotion is met.
Change: remove the demotion seam (construction, not validation — §5)
The
DiscoveryCorpusAdvisorygate's only behavioral effect was demoting 8 typecheck diagnostic kinds (UnresolvedType,TypeMismatch,ArityMismatch,VariantNotFound,FieldNotFound,NonExhaustiveMatch,MissingAnnotation,VariantCollision) so per-entry discovery resolve wouldn't fail on them — they were only logged asadvisory(typecheck):. (compile_to_resolved_discovery_corpus_advisorywas a pure alias ofcompile_to_resolved.)Rather than leave an inert "advisory" seam that no longer demotes (a §2/§3 dead-concept / §5 lie), this removes the demotion entirely: the discovery floor now blocks on typecheck diagnostics via
is_interpreter_blocking_diagnostic, identical to the strict path. The bad state (silently-demoted typecheck debt re-accumulating) becomes unwritable — the floor itself is the construction wall; a future typecheck regression fails the discovery floor loudly.Removed (Rust seed + .dag authority, kept coherent):
is_discovery_corpus_advisory_typecheck_diagnostic,is_discovery_corpus_blocking_diagnostic(v1_std_core.rs,00_core.dag)compile_to_resolved_discovery_corpus_advisory(v1_compiler_compile.rs,compile.dag)ResolveTypecheckGateenum + gate threading,log_discovery_advisory_typecheck, and the census collector (cli_run.rs)pipeline.rs)Added:
discovery_corpus_typecheck_diagnostics_are_blocking— a fast discriminating test proving a typecheck error now surfaces as a blocking diagnostic under the promoted gate.Verification
cargo build -p v1-compiler✓cargo clippy -p v1-compiler -p v1-compiler-tests --all-targets -D warnings✓🤖 Generated with Claude Code