PR A: one release-profile v1-compiler build per CI build job (align features, gate after release)

[briansrls]

Summary

One release-profile v1-compiler build per CI build job: align ci_release_features with v1-compiler-tests (text_lookup_work_counter,interp_test_witness,test_hooks), run cargo test --release --no-run after the release build so the gate reuses the same artifact, and default-off interp_test_witness arming so release bins compile without per-call chain-walk instrumentation.

Admission removed: the lane brief step 3 asked for change-sensitive admission. We built it (compile_contract_admission_witness, receipt, five unit tests). Measurement showed it costs 84–133s (resolve_entry_graph_shared for one git diff; the diff itself is 16–18ms) to decide whether to skip an 18s gate compile — a net loss on every PR, including docs-only skips. Removed rather than kept because it was requested.

Structural claim (proven — log, not wall-clock)

One release-profile v1-compiler compile per build job. Release step logs Compiling v1-compiler once; the gate step compiles only v1-compiler-tests against that artifact. Feature union closes the artifact fork; default-off arming keeps the shared interp_test_witness feature inert in release bins (one relaxed load per eval_call until a test calls arm_call_env_depth_witness_for_test).

Gate command benefit (controlled local comparison)

CI wall-clock on this fleet is a per-slot workspace lottery: self-hosted runners keep _work/target/ between jobs, so the same gate command spans 18s–4m30 depending on which slot you draw — not on the change. actions/cache (~250 MB) is registry/index only; sccache deltas of hit 0 / miss 1 appeared on both fast and slow samples. No single CI run measures the PR; collecting more CI samples samples the lottery.

Controlled comparison on this worktree (RUSTC_WRAPPER= unset, CARGO_INCREMENTAL=0):

	Command	Wall time
old cold	cargo check -p v1-compiler-tests --tests (dev profile; cleans v1-compiler + v1-compiler-tests)	63.6s
old warm	same	0.31s
new cold	cargo test -p v1-compiler-tests --release --no-run (release v1-compiler already built; clean v1-compiler-tests only)	18.2s
new warm	same	0.33s

New-cold beats old-cold; warm paths tie. That is the gate-command claim this PR makes.

CI observations (illustrative only — slot-dependent)

Run	Host	Gate step	Notes
31045643507	srv2-04	~18s cargo	warm slot
31062435170	srv2-01	2m41 gate step	current head; admission gone
31056301056	srv1-04	~4m30 gate cargo	broken-host OOM / sccache placement (separate known defect)

Do not read any one row as steady-state CI cost.

Merge bar

Bar	Status
Expensive v1-compiler compilations = 1	Proven (release-step log)
Build job materially faster	Gate command: local controlled comparison above; CI wall: slot lottery — cannot certify from one sample
Floor wall non-regressing	Default-off arming; depth tests arm explicitly

Open (operator): one shared artifact vs literally-zero per-call instructions vs full compile-coverage.

Test plan

• Compile gate witness (ordering, feature alignment, unconditional cargo command)
• Gate log artifact-reuse proof
• Controlled local old-vs-new gate timing (four samples above)
• Admission removal verified (zero compile_contract in cli_run.rs; witness bin gone)

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[gunbai-bot]

Addressed review 49123 in 52c748cbf:

• ci_release_features is now text_lookup_work_counter only; interp_test_witness and test_hooks are dropped from the release build line (ci.yml/falsifier.yml updated to match).
• Default-off arming instead of compile_contract_surface: record_call_env_depth is behind interp_test_witness and returns immediately unless arm_call_env_depth_witness_for_test() was called. Release bins omit the feature, so claim_executor/gunbc do not contain the hook at all.
• The WIP 4c8ab445c path (which introduced compile_contract_surface and failed cargo fmt) is reverted/superseded.

Warm compile-gate locally after cached test graph: ~0.27s (vs ~173s first compile). CI re-running on this head.

— sent from fierce-heron-177

[gunbai-bot]

review 49198 (Composer REQUEST_CHANGES): Verified — v1_compiler_tests_typecheck_module_callers_hand_rust_dissolve_trigger still cited cargo check -p v1-compiler-tests --tests while the build job now enrolls cargo test -p v1-compiler-tests --release --no-run after the release build (ci_v1_compiler_tests_compile_gate_emit / ci_v1_compiler_tests_compile_gate_witness_test.dag). Updated the scaffold line to match (a594292).

— sent from fierce-heron-177