Skip to content

Phase 1: ExecutionMode{Hermetic|Wet} for interpreter + claim_batch --wet - #5071

Merged
briansrls merged 11 commits into
mainfrom
session/calm-cat-275
Jun 17, 2026
Merged

Phase 1: ExecutionMode{Hermetic|Wet} for interpreter + claim_batch --wet#5071
briansrls merged 11 commits into
mainfrom
session/calm-cat-275

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Jun 17, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Summary

  • Replace interpreter dry_run: bool with ExecutionMode { Hermetic, Wet } on InterpContext, threaded through make_eval_context and a single dispatch gate (hermetic → modeled mock_response; wet → live transport).
  • claim_batch (CI claim gate) exposes the mode surface: default Wet (CI unchanged), --hermetic for mock dispatch, --wet explicit Wet (Phase 2 opt-in when default flips Hermetic).
  • gunbc run --dry-run maps to ExecutionMode::Hermetic; claim_executor and library tests pass Wet explicitly.

Test plan

  • cargo test -p v1-compiler-tests interp_dry_run — 3 passed
  • cargo check -p v1-compiler --bins — clean
  • CI green on claim_batch discovery floor (default Wet — no workflow change)

briansrls and others added 2 commits June 17, 2026 05:34
@briansrls
WIP: Phase 1 (hermetic-testing rollout): generalize interpreter dry_r…
7ebcdfc 
…un:bool
@briansrls @cursoragent
Add claim_batch --wet regression test for ExecutionMode wiring.
82118bb 
Pins the CI claim gate default (Hermetic) and opt-in wet dispatch path
introduced in Phase 1 of the hermetic-testing rollout.

Co-authored-by: Cursor <cursoragent@cursor.com>
@gunbai-bot gunbai-bot Bot changed the title Phase 1 (hermetic-testing rollout): generalize interpreter dry_run:bool into ExecutionMode{Hermetic|Wet} threaded through InterpContext + make_eval_context (cli_run.rs:741), and add a --wet flag to ci_claim_gate so the SAME .dag test fn runs hermetic (mock_response) or wet (live dispatch) chosen by Phase 1: ExecutionMode{Hermetic|Wet} for interpreter + claim_batch --wet Jun 17, 2026
@gunbai-bot
gunbai-bot Bot marked this pull request as ready for review June 17, 2026 05:40
briansrls and others added 7 commits June 17, 2026 05:42
@briansrls @cursoragent
Fix claim_batch default to Wet per Phase 1 CI constraint.
06c97fd 
Phase 1 delivers the ExecutionMode mechanism without changing CI behavior:
witnesses default to live dispatch; --hermetic opts into mock_response.
--wet remains explicit Wet for the Phase 2 polarity flip.

Co-authored-by: Cursor <cursoragent@cursor.com>
@briansrls
WIP: Phase 1 (hermetic-testing rollout): generalize interpreter dry_r…
3c1f2ab 
…un:bool
@briansrls @cursoragent
Add by-execution claim_batch --hermetic receipt test.
7f9f497 
Drives a discriminating REST witness through the claim_batch binary:
default Wet fails on unreachable endpoint; --hermetic greens via
mock_response. Complements the fast source-grep wiring check.

Co-authored-by: Cursor <cursoragent@cursor.com>
@briansrls
WIP: Phase 1 (hermetic-testing rollout): generalize interpreter dry_r…
af985c0 
…un:bool
@briansrls
Merge origin/main into session/calm-cat-275.
f58b366 
Integrate upstream before CI fix; keeps PR current with main.
@briansrls @cursoragent
Apply rustfmt to pipeline.rs after main merge.
a572bdc 
Fixes rust gate failure from long compile_dag_named call sites
introduced by the extdeps/llm split on main.

Co-authored-by: Cursor <cursoragent@cursor.com>
@briansrls
WIP: Phase 1 (hermetic-testing rollout): generalize interpreter dry_r…
36bd095 
…un:bool
@gunbai-bot

gunbai-bot Bot commented Jun 17, 2026

Copy link
Copy Markdown
Contributor Author

Addressed the §5 grep-test finding: dropped claim_batch_wires_execution_mode_flags in 36bd095. Coverage is unchanged — claim_batch_hermetic_flag_uses_mock_response_by_execution already drives the claim_batch binary with a discriminating REST witness (default Wet fails on unreachable endpoint; --hermetic greens via mock_response).

— sent from calm-cat-275

briansrls and others added 2 commits June 17, 2026 06:44
@briansrls
WIP: Phase 1 (hermetic-testing rollout): generalize interpreter dry_r…
cf4726e 
…un:bool
@briansrls @cursoragent
Store ExecutionMode directly in claim_batch ParsedArgs.
408641a 
Addresses review: avoid bool wet → ExecutionMode re-translation at run().
Also drop in-test cargo build for the hermetic receipt test; prefer an
existing release/debug claim_batch binary (CI builds release in the floor).

Co-authored-by: Cursor <cursoragent@cursor.com>
@gunbai-bot

gunbai-bot Bot commented Jun 17, 2026

Copy link
Copy Markdown
Contributor Author

Addressed the advisory items from review 29755:

  1. ParsedArgs now carries ExecutionMode directly (--wet / --hermetic set the enum in parse_args; no bool re-translation in run()).
  2. Hermetic receipt test no longer shells out to cargo build inside the test — it uses target/release/claim_batch (CI floor) or target/debug/claim_batch, with a clear fail-closed message if missing.
  3. pipeline.rs rustfmt — acknowledged; harmless merge-main churn, left as-is.

— sent from calm-cat-275

@briansrls
briansrls merged commit 6a64553 into main Jun 17, 2026
1 check passed
@briansrls
briansrls deleted the session/calm-cat-275 branch June 17, 2026 07:42
briansrls added a commit that referenced this pull request Jun 17, 2026
@briansrls
Merge origin/main (#5087 complexity budget, #5071 ExecutionMode) into…
9db78cc 
… session/eager-stag-401
briansrls added a commit that referenced this pull request Jun 21, 2026
@briansrls @claude
Widen rust gate: declare merge-landed test (lens caught it) + flag on…
2eb3261 
…e pre-existing fixture red

The coverage-completeness lens (this PR's residue) fired on its first real
input after merging main: resolve_typed_cache_equivalence_test.rs landed via
#5071 but was never declared in lib.rs, so it would silently not run — exactly
the gap the lens exists to close (fail-closed, working as designed). Declared it.

dag_emit_from_resolved_matches_compile_sources_for_v4_slice reads a removed
fixture (fixtures/v2-mvp1, absent from the tree and every git ref) — a
pre-existing red hidden by the old 3-filter allowlist, surfaced by the widening.
Flagged with a written #[ignore] reason (FLAG-DON'T-FIX); the lens keeps the
excuse reviewable.

The ~20 interp_recorded_fixture/dry_run local failures were a build-ordering
artifact (claim_batch bin not built locally); they pass once the floor's
release --bins build is present, so they are NOT ignored (ignoring them would
have been a fail-open).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@gunbai-bot gunbai-bot Bot mentioned this pull request Jun 21, 2026
Merged
briansrls added a commit that referenced this pull request Jun 22, 2026
@gunbai-bot @briansrls @claude
Widen the rust gate by construction (§1): invert the hand-picked 3-fi…
eff64e4 
…lter allowlist (29 of 792 green tests) → run-all-unless-#[ignore]d-with-written-reason; add CI-coverage-completeness so a new test is covered by default (fail-closed); measure CI-time impact before committing the full set (#5427)

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* Widen rust gate: declare merge-landed test (lens caught it) + flag one pre-existing fixture red

The coverage-completeness lens (this PR's residue) fired on its first real
input after merging main: resolve_typed_cache_equivalence_test.rs landed via
#5071 but was never declared in lib.rs, so it would silently not run — exactly
the gap the lens exists to close (fail-closed, working as designed). Declared it.

dag_emit_from_resolved_matches_compile_sources_for_v4_slice reads a removed
fixture (fixtures/v2-mvp1, absent from the tree and every git ref) — a
pre-existing red hidden by the old 3-filter allowlist, surfaced by the widening.
Flagged with a written #[ignore] reason (FLAG-DON'T-FIX); the lens keeps the
excuse reviewable.

The ~20 interp_recorded_fixture/dry_run local failures were a build-ordering
artifact (claim_batch bin not built locally); they pass once the floor's
release --bins build is present, so they are NOT ignored (ignoring them would
have been a fail-open).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Mark v4_slice ignore as draining-worklist (delete-or-restore), not permanent

Parent note: a test whose fixture is gone from every git ref is DEAD — it can
never pass, so a permanent ignore would calcify a coverage hole. Sharpened the
reason to name the resolution (delete the dead parity receipt OR restore/retarget
the fixture at a live v2 source set) and the owner routing (v2 emit slice, via
bright-stag), so the completeness lens keeps it visible as work-to-drain.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Fix completeness-lens false-positive on multi-line #[ignore] reasons

My own lens (every_ignore_carries_a_written_reason) went RED on the v4_slice
ignore I'd just added: the detector required the closing quote on the same line
as #[ignore = "..."], so a long reason wrapped with \ line-continuation left
the opening line with an unterminated string and was misread as reasonless — a
false-positive that fails-closed on legitimately-reasoned input.

Detector now recognizes a multi-line reason: opening quote + non-empty content
(sans a trailing \ continuation) is reasoned even when the string closes on a
later line; an opener with no content before the break is still empty → flagged.
Added detector_accepts_a_multiline_reason as the discriminating control (both the
valid wrap and the empty-multiline edge), fed through the same authority.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Flag the 16 execution-confirmed pre-existing reds the widening surfaces (#[ignore=failing])

The widened gate runs the self_gen8 + census families that the old 3-filter
allowlist never ran. warm-ram's opt-level run listed 32 candidates; running them
to completion (--no-fail-fast, claim_batch built) confirms only 16 actually FAIL
— the other 16 self_gen8 PASS (e.g. kernel_type_import). Ignoring warm-ram's
superset would have excluded 16 GREEN tests (fail-open); only the execution-
confirmed failures are flagged.

  • ownership_stage0_census (1): clone-census ratchet RED on main (non-emit
    .clone() 21540 > 20200+202, ~1138 over) — inert under the old allowlist while
    the seed drifted UP against "Rust shrinks toward zero". Do NOT bump the cap
    (project spirit); route to a census/substrate-migration owner.
  • self_gen8_* (15): pre-existing self-host emit regression (parametric-alias-RHS
    / reexported-type-import module resolution); route to the v2 self-host Route-A
    owner.

All draining-worklist, not permanent; the completeness lens keeps each reason
visible and reviewable. fmt-clean (rust gate runs fmt --check); parse.rs picked
up a pre-existing fmt fix from the earlier ignore conversion.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Merge origin/main (ac9a7e7) — resolve pipeline.rs: accept main's deletion of dead v4_slice test

main moved again (#5449 etc.) after my first sync. Conflict in pipeline.rs:
main DELETED dag_emit_from_resolved_matches_compile_sources_for_v4_slice (the dead
parity receipt reading the removed fixtures/v2-mvp1) — exactly the delete-or-restore
resolution I'd flagged and routed. Accepted main's deletion; my interim #[ignore]
on it is now moot and gone. The 16 self_gen8/census failing-ignores stay intact;
completeness lens green (no undeclared test file from the merge); crate compiles.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* Cluster A construction-fix (v2_layer_roots single authority) + cluster B ignore-with-reason; fix doc-lazy-continuation

27 reds surfaced by the run-all widening:
- 14 fixed-by-construction: under-scoped v2_source_roots()=[src/v2] dropped dsl, so
  v2-core's extdeps.communication.medium import was unresolvable. Funnel the 4 files'
  local copies through one helpers::v2_layer_roots()=[src/v2,dsl] authority (mirrors
  gunbc.ci_layer_roots; §6 dissolution marker to derive from the .dag fact later).
- 13 pre-existing emit/inference reds: #[ignore=failing] with written symptom + bucket
  (emit/inference/lang-go/lang-python), routed as follow-ups.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* #5427: ignore 2 more pre-existing reds the widening surfaced (#5473 + #5484 interactions)

The full run-all enumeration (761 pass / 2 fail / 89 ignored, 0 infra) surfaced two more
deterministic reds beyond the wall-clock timing test — both pre-existing on origin/main,
hidden by the old 3-test allowlist, untouched by this PR:

- source_root_ingest_manifest_host_test::manifest_entry_admission_qualified_name_is_well_formed:
  #5473 source_root admission has an absolute-vs-relative contradiction (abs path rejected by
  admission; relative path fails the file-existence check) — proven by execution. Real #5473
  API bug; routed to the source_root-tagging owner.
- wet_hermetic_equivalence_test::wet_hermetic_scaffold_roster_outcomes_agree:
  whole-tree mock-corpus precompute can't resolve #5484's doc_reachability_witness_test.dag
  (doc_graph_* fns not in scope); routed to the doc-graph reachability-lens owner.

Surgical #[ignore=failing] with routed reasons (bucket=source-root-ingest-admission /
doc-graph-wholetree-resolve). Reason-based completeness lens covers them.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* #5427: cite PR #5504 (RED#1 heal-PR) in the source-root admission ignore reason

#5504 (still-deer-248's de-fork-lane fix) grounds absolute --source-root via
repo_relative_dag_path, fixing the abs-vs-rel admission contradiction behind the
#[ignore]'d manifest_entry_admission test. #5504 is up but not yet merged, so the
interim ignore stays; reason now names #5504 as the heal-PR + un-ignore trigger.
When #5504 merges, a freshness-merge heals RED#1 and the ignore drops -> zero ignores.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* #5427: route rust gate through the cargo.Build model (§3), drop hand-typed argv strings

run_gates() now calls the modeled extdeps.cargo_build ops — cargo.Build.Fmt(["--all","--check"]),
Clippy(["--all-targets"],["-D","warnings"]), Test(["-p","v1-compiler-tests"]) — so the cargo
shell-transport argv is the single authority, not a hand-typed command String. Removes the now-dead
ci_rust_gate_{fmt,clippy,test}_command() String fns from ci_spec (they were §3 nicknames for cargo's
already-modeled interface; tools/build.dag uses the same ops). Coverage-by-construction unchanged
(run-all-except-#[ignore]d; completeness lens is the residue).

EAGAIN cold-retry is transport resilience the cargo model can't yet express (env not wired to the
shell transport); it relocates to the CI workflow step that runs the gate (follow-up: rust_tests job).

Validated by execution: floor_effect_gate_witness rust_monolith_gate_passes resolves 78 sources and
dispatches all three modeled cargo commands. Fixes the partial-snapshot break in 565a116.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Resolve merge conflict markers in rust gate cargo-modeling (ac0b732)

The auto-committer captured the merge with unresolved <<<<<<< markers in
rust_gates_ci.dag and ci_spec.dag. Resolution keeps the cargo.Build model
routing (Task 1) over origin/main's old shell-string version (which still
carried the narrow 3-filter ci_rust_gate_test_command -- exactly what this
work inverts). No code change beyond marker removal.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* WIP: Widen the rust gate by construction (§1): invert the hand-picked 3-filte

* Resolve merge conflict in cargo_build.dag; conform new .dag files to comment-ban

The auto-committer pushed an unresolved-conflict state (<<<<<<< markers in
cargo_build.dag:3,8,9) which fails to parse — fixes the BLOCKING review and the
red dsl_compile_clean_gate.

main's comment-ban sweep (#5537/#5543/#5539) left dsl/std, dsl/extdeps,
dsl/gunbc, dsl/tools, src/v2/test/claim comment-free; this PR's new/rewritten
files in those dirs re-introduced comments. Strip them to conform (code
unchanged) so we match already-merged state and avoid a future re-sweep.

ci.yml emit is byte-identical (comments don't affect output); drift gate green,
ci_spec_witnesses green, rust gate + floor_effect_gate_witness resolve clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Fix stale source-audit literal: compile.dag gate predicate (run-all surfaced)

The run-all widening (#5427) surfaced source_audit::compile_gate_keeps_infer_errors_blocking_in_stage0,
red because main refactored compile.dag's emission gate from
is_error_diagnostic to is_resolved_pipeline_typecheck_blocking (discovery-corpus
advisory typecheck) without updating this brittle string-audit literal. The
audit never ran under the 3-filter allowlist, so the drift went unseen — exactly
the class this widening exists to catch. Intent is unchanged (emission is gated
on type errors); update the literal to the current predicate so the audit stays
live and green. Fix, not #[ignore]: a stale literal is cheaply correctable, not
deep work to route.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Resolve .rs merge conflicts: keep #[ignore=reason] set, conform to .rs comment-ban

The auto-committer committed an unresolved merge (3052ecd, markers in 5 test
.rs files) bringing main's .rs comment-ban sweep (#5544). Resolution: restore my
pre-merge versions (preserving every #[ignore = "<reason>"] — the single
authority my coverage-completeness lens requires; main's bare #[ignore] would
red that lens) then strip full-line // comments to conform to the ban. Code
unchanged; 17 self_gen8 + diagnostics reasoned ignores intact; zero reasonless
#[ignore] introduced.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Re-delete orphan profiling scaffold the merge resurrected (my own lens caught it)

body_producer_infer_profile_test.rs is a manual timing-profiling scaffold
(std::time::Instant, profile_* fns, bare #[ignore], never declared in lib.rs).
My widening deliberately deleted it (26beeee); the auto-committer's merge
(3052ecd) left main's copy in the tree via the unresolved deleted-in-HEAD/
modified-in-main conflict. The new coverage-completeness lens correctly fired on
it twice — orphan file (every_test_file_is_declared_in_lib) + bare reasonless
#[ignore] (every_ignore_carries_a_written_reason) — a live proof the lens has
teeth. Re-delete to honor the original deletion; both lenses green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Flag contention-flaky wet/hermetic equivalence test (run-all surfaced)

wet_hermetic_scaffold_roster_outcomes_agree passes isolated (~110s) but fails
under the 766-test parallel load (~335s). nextest process-isolates each test, so
this is resource/timing contention on the wet-execution path, not a logic bug —
a non-deterministic test can't gate a merge (§5 fail-open-by-noise). Pre-existing
(#5276), never run under the old 3-filter allowlist. #[ignore=reason] + route to
the wet==hermetic / hermetic-testing owner; the coverage lens keeps it visible.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* cargo fmt: collapse blank lines left by .rs comment-strip

The comment-strip in f05415d left double-blank-lines where comments had been;
cargo fmt --all --check rejected them (the rust gate's Fmt op would red). Pure
blank-line removal (105 lines), zero code change, all #[ignore]s intact. Also
addresses the blank-line noise the 19:42 review flagged.

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>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@briansrls