M1 CI wiring: add --target rust src/v4 → rustc step to surface v2 emitter gaps for v4 by briansrls · Pull Request #3654 · gunb-ai/gunbc

briansrls · 2026-05-25T04:49:06Z

Summary

Adds an informational M1 CI step: scripts/v4-m1-rust-emit-probe.sh runs v2-compiler compile --target rust over the full src/v4 tree, then cargo check on the emitted crate. Modeled in src/v4/workflow/ci.dag (T-24) with m1_ci_live_workflow_signal bridging to .github/workflows/ci.yml. The step uses continue-on-error: true and the script exits 0 in probe mode — it surfaces v2 emitter / rustc gaps without blocking merge.

P5(b) receipt — expanded v3 hand-Rust (exactly one)

File: src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs (+helpers + v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml)

Standing authority (no new census path): INVARIANTS.md §P5(b) table row for this file + sg0_census_test.rs EXPECTED_HAND_AUTHORED_TEST entry (T-PB-B / pb_rust_tests_outside_residual_zero; TASKS.md T-21 + T-24).

This PR expansion: interim ratchet adds M1 ci.dag↔ci.yml binding assertions (step name, script path, continue-on-error, v4/workflow_policy if) and P5 dissolution-tag presence (feature:project-github-actions-landed, consumer:v4.workflow.ci m1_ci_live_workflow_signal, bind src/v4/TASKS.md T-24).

Dissolution: remove when Node-level TestClaim CI selection is exercised only by .dag TestClaim rows / generated harness / host runner without this parse probe (full wire: edit-locus + TestClaimExerciseSurface + ci.yml — follow-on A/B/C).

Probe findings (local run)

Phase	Result
`v2-compiler compile --source-root src/v4 --target rust`	PASS — `compiled: 235 files emitted, 0 diagnostics`
`cargo check` on emitted tree	FAIL (exit 101) — rustc errors across ~205 distinct `.rs` modules

Top rustc error categories: E0282 (~2125), E0107 (~792), E0308 (~669), E0425 (~319), E0061 (~298), E0277 (~295).

Interpretation: v2 structural emit is clean; M1 is blocked on emitted-Rust quality, not parse/resolve. This PR does not fix emitter gaps.

Test plan

bash scripts/v4-m1-rust-emit-probe.sh — exits 0
cargo test -p v3-compiler --test integration v4_workflow_ci_m1_rust_emit_probe — binding test passes
CI green on cd9862074d (informational M1 step, non-gating)

…2 emitte

PRs that touch only the M1 probe script must still run the gated CI step; mirror the existing scripts/v4-mvp1* selection rule. Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls · 2026-05-25T04:59:30Z

Codex finding (affected-set gate): addressed in 087d8f6

Verified: scripts/detect-affected-components.sh only listed ^scripts/v4-mvp1 under the v4 bucket, so a PR touching only scripts/v4-m1-rust-emit-probe.sh would set v4=false and workflow_policy=false and skip the new M1 step — fail-open for probe-only edits.

Fix: extend the v4 grep to include ^scripts/v4-m1 (same prefix discipline as MVP-1). workflow_policy still covers .github/workflows/ + this detector script itself; script-only M1 probe changes now correctly set v4=true.

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 087d8f66 · Trigger: schedule
Thinking: 244s wall

BLOCKING (1)

Root Cause

dsl/gunbc/ci.dag pipeline.gates is still the gate-membership authority but has no M1 rust emit probe row → add a non-blocking modeled gate/source for the probe, or land a same-PR bounded T-24 bridge with a concrete dissolution trigger.

⚠️ The PR adds useful CI visibility, but the new gate needs to stay tied to the modeled CI authority instead of expanding hand-authored CI state.

Sign in to view

        target/release/v2-compiler compile --source-root "${entry_root}" --source-root "${deps_root}" --output-dir "${RUNNER_TEMP}/v4-lens-ci-rust" --target rust
+    # M1 (docs/v4-compilation-milestones.md): full-tree --target rust + rustc gap surface.
+    # Informational only — does not gate merge (continue-on-error; probe script exits 0).
+    - name: M1 v4 full-tree rust emit probe (v2 emit + cargo check)


…2 emitte

briansrls · 2026-05-25T05:12:00Z

Blocking review (modeled CI authority): addressed

Verified against current 087d8f66 + this commit:

src/v4/workflow/ci.dag (T-24 v4 CI authority) — not dsl/gunbc/ci.dag (that file’s pipeline.gates is gunbc workspace compiler CI: lint/tests/compile-gates only; it does not list v4 bootstrap/MVP-1/Lens-CI either — those are already live in .github/workflows/ci.yml under the v4.workflow interim bridge).

Landed in src/v4/workflow/ci.dag:

M1RustEmitProbeCommand + m1_rust_emit_probe_execution job (needs v2_compile_src_v4)
m1_rust_emit_probe_signal gate (run_policy: Always)
M1CiLiveWorkflowSignal / m1_ci_live_workflow_signal pinning step name, scripts/v4-m1-rust-emit-probe.sh, non_blocking: true
ci_command_authority_ok arm for the new command

Receipt: v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml asserts modeled fields match the live ci.yml step (continue-on-error, v4/workflow_policy if, script path).

Dissolution trigger unchanged from Lens-CI: delete M1CiLiveWorkflowSignal when GitHub Actions projection reads ci_pipeline directly (R3 gates #98/#100).

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 7ee7349c · Trigger: schedule
Thinking: 271s wall

BLOCKING (1)

Root Cause

src/v4/workflow/ci.dag M1 probe planning authority is named outside the landed T-24/project-github-actions authority chain → bind the bridge to the concrete T-24/ROADMAP arrival or land the missing planning authority with a feature:/consumer: tag.

⚠️ The prior authority gap is fixed, but the new M1 bridge needs a checkable dissolution receipt before merge.

Sign in to view

  semantic_target: String
 }

+// 🟡 gated — live-workflow projection interim (M1 docs/v4-compilation-milestones.md) — dissolve-on-arrival:


Replace unlanded milestones-doc anchor with checkable feature: and consumer: tags on the M1CiLiveWorkflowSignal gate (r3 #98/#100, TASKS T-24). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls · 2026-05-25T06:02:39Z

Blocking inline + codex review (P5 dissolution / authority chain): fixed in cd9862074d

Verified on current HEAD before this commit:

ci.yml:227 modeled authority — already present (m1_rust_emit_probe_signal + m1_ci_live_workflow_signal in src/v4/workflow/ci.dag); binding test v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml passes.
ci.dag:83 unlanded docs/v4-compilation-milestones.md — valid finding; that path is not on main and is not a checkable P5 anchor.

This commit:

Replaces the M1 🟡 gate with landed authority: docs/r3-program-plan.md gates Compiler pipeline design #98/Registry externs deletion #100, feature:project-github-actions-landed, consumer:v4.workflow.ci m1_ci_live_workflow_signal, bind src/v4/TASKS.md T-24 (same dissolution chain as Lens-CI / T-24).
Drops unlanded milestones-doc references from ci.yml + scripts/v4-m1-rust-emit-probe.sh.
Extends the workflow CI smoke test to assert the feature/consumer/bind tags are present.

M1 planning context remains operator brief / milestone work; executable CI authority is v4.workflow.ci (T-24) until project_github_actions projection dissolves the live-workflow carrier.

— sent from quick-moth-204

briansrls · 2026-05-25T06:06:24Z

Claude RC (P5(b) PR-body receipt): addressed without code change

Verified: this PR expands v4_workflow_ci_runner_dag_smoke_test.rs (~99 lines) under an existing EXPECTED_HAND_AUTHORED_TEST census line (sg0_census_test.rs:885) with a standing INVARIANTS.md §P5(b) row (T-21/T-24 / T-PB-B deferral).

Fix: PR description now states exactly one checkable P5(b) receipt for that expansion (standing row + this PR’s new v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml interim ratchet + dissolution trigger). No new census path; no INVARIANTS row edit required.

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: cd986207 · Trigger: schedule
Thinking: 250s wall

BLOCKING (1)

Root Cause

.github/workflows/ci.yml v3 integration tests are compiled but not executed by default under the hot-fix zero-test filter → add an explicit cargo test filter for v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml alongside the other live CI receipts.

⚠️ One live receipt is not actually executed in CI.

Sign in to view

    );
 }

+#[test]


…2 emitte

briansrls · 2026-05-25T07:07:45Z

Blocking review (M1 test not executed in CI): fixed in 0c16c67937

Verified on cd986207: v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml lived only in the v3 job’s integration harness, which still runs __HOT_FIX_NONEXISTENT_FILTER__ (gunbc#846) — so the new P5 bridge was compiled but not executed.

Fix:

Added explicit ci job step M1 v4 workflow CI model/YAML binding smoke (v4/workflow_policy gated) running cargo test … v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml -- --exact --quiet — same bypass pattern as Lens-CI / T-15 smokes.
Extended m1_ci_live_workflow_signal with binding_smoke_step_name; test asserts the modeled step name runs that filter.

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 5bf8a146 · Trigger: schedule
Thinking: 264s wall

BLOCKING (2)

Root Cause

src/v3/compiler/tests/integration.rs the integration harness wraps this file in the v4_workflow_ci_runner_dag_smoke_test module → use the module-qualified exact filter in ci.yml and in the self-check string.
src/v4/workflow/ci.dag M1CiLiveWorkflowSignal introduces non_blocking as a projection fact, but the only same-PR consumer ignores that field and restates the boolean → parse the Bool value or remove the field, then derive the YAML assertion from the single authority.

⚠️ The M1 bridge is modeled, but the live receipt still does not actually execute and one modeled projection fact is not consumed.

Sign in to view

+    # M1 model↔ci.yml binding — v3 integration harness uses gunbc#846 zero-test-filter; run here.
+    - name: M1 v4 workflow CI model/YAML binding smoke
+      if: needs.affected.outputs.v4 == 'true' || needs.affected.outputs.workflow_policy == 'true' || (github.event_name == 'push' && github.ref == 'refs/heads/main')
+      run: cargo test -p v3-compiler --test integration v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml -- --exact --quiet


Sign in to view

+        "{CI_YML_PATH}: `{step_name}` must run for v4 and workflow-policy changes"
+    );
+    assert!(
+        m1_step.contains("continue-on-error: true"),


…2 emitte

… ci.dag Use v4_workflow_ci_runner_dag_smoke_test:: filter in self-check (ci.yml already qualified). Derive continue-on-error YAML assertion from m1_ci_live_workflow_signal.non_blocking (P2 single authority). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls · 2026-05-25T08:08:09Z

Blocking reviews (libtest filter + non_blocking authority): addressed in 69ed71056b

Verified on 5bf8a146:

Unqualified --exact filter — valid: integration.rs mounts the file as mod v4_workflow_ci_runner_dag_smoke_test, so bare v4_workflow_ci_m1_… matches zero tests. Fix: ci.yml already uses v4_workflow_ci_runner_dag_smoke_test::v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml; binding self-check uses the same M1_BINDING_TEST_FILTER constant.
non_blocking drift — valid: assertion hard-coded continue-on-error: true. Fix: expr_bool(record_body_field(live_signal, "non_blocking")) drives the YAML check (true ⇒ must contain continue-on-error: true; false ⇒ must not).

Local: cargo test -p v3-compiler --test integration v4_workflow_ci_runner_dag_smoke_test::v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml -- --exact passes.

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 69ed7105 · Trigger: schedule
Thinking: 357s wall

ROADMAP — Verified

src/v4/TASKS.md T-24: The PR keeps the M1 probe scoped to workflow/ci.dag as CI pipeline-as-data with a tracked live-workflow bridge.

✅ The prior blocking issues are addressed, and I found no new blocking concerns in the changed surface.

briansrls · 2026-05-25T09:06:45Z

Inline blocking (M1 test not in CI receipts): already addressed on 69ed71056b — no further code change

Verified on current HEAD:

ci job step M1 v4 workflow CI model/YAML binding smoke (lines 195–198) explicitly runs:
cargo test -p v3-compiler --test integration v4_workflow_ci_runner_dag_smoke_test::v4_workflow_ci_m1_rust_emit_probe_modeled_and_bound_to_ci_yml -- --exact --quiet
Same gunbc#846 zero-test-filter bypass pattern as T-15 / Lens-CI smokes (comment at 185–188).
Binding test self-check uses M1_BINDING_TEST_FILTER constant matching that command; non_blocking is read from m1_ci_live_workflow_signal.

Codex schedule review on 69ed7105: acknowledged — prior blockers fixed; no new blocking concerns in changed surface.

— sent from quick-moth-204

briansrls · 2026-05-25T14:52:46Z

Review metadata

Provider / model: openai-pro / gpt-5-5-pro
Commit: 69ed7105 · Trigger: manual
Comparison: main @ 3a9be658 ... session/quick-moth-204 @ 69ed7105
Conversation: View conversation

1. Story of the diff

This PR adds a non-blocking M1 CI probe that runs the v2 compiler against the full src/v4 tree with --target rust, then attempts cargo check on the emitted Rust tree to expose current v2 emitter/rustc gaps without failing the main CI lane. The mechanism is split across three surfaces: .github/workflows/ci.yml adds a binding-smoke test step plus the actual probe step, scripts/v4-m1-rust-emit-probe.sh performs compile/log summarization/rustc probing, and scripts/detect-affected-components.sh makes scripts/v4-m1* changes mark v4 affected. The workflow model is updated in src/v4/workflow/ci.dag by adding M1RustEmitProbeCommand, a modeled job/gate, and a M1CiLiveWorkflowSignal bridge that the Rust integration test checks against the committed YAML.

2. Invariant categories

LAYER MODEL — Finding

src/v4/workflow/ci.dag:83 says the M1 live-workflow bridge forbids “adding M1 probe step facts outside ci_pipeline + this binding,” and src/v4/workflow/ci.dag:84-88 models only binding_smoke_step_name, step_name, script_path, and non_blocking. But .github/workflows/ci.yml:234 adds the operational bound timeout-minutes: 20 outside that modeled binding. Because workflow/ci.dag is the modeled CI authority, this leaves a load-bearing YAML fact outside the model. Add a modeled timeout field and bind/assert it, or remove the YAML timeout as non-contractual.

INVARIANTS.md + modeling-discipline.md — Finding

Principle: P2 Boundary Discipline / single authority and Facts flow forward. INVARIANTS states that every fact lives in exactly one authoritative place (INVARIANTS.md:165), and modeling discipline requires structured facts to be carried forward or explicitly justified (modeling-discipline.md:131-137). The PR correctly creates a bridge carrier, but the bridge is incomplete: .github/workflows/ci.yml:234 carries timeout-minutes: 20, while src/v4/workflow/ci.dag:84-88 has no corresponding modeled field. That creates an unmodeled drift point for the M1 probe’s execution bound.

CODING.md — Compliant

The new shell script is an edge/CI script rather than library logic, and its inputs are explicit environment variables (scripts/v4-m1-rust-emit-probe.sh:11-17) with structured summary output (scripts/v4-m1-rust-emit-probe.sh:133-156). The Rust additions are test-only helpers with narrow parsing/extraction roles.

TESTING.md — Finding

The new binding test is at the right level, but it does not cover the full behavior introduced by the PR. It extracts and asserts step_name, script_path, and non_blocking from the model (src/v3/compiler/tests/integration/v4_workflow_ci_runner_dag_smoke_test.rs:256-287), while the YAML step also introduces timeout-minutes: 20 at .github/workflows/ci.yml:234. TESTING’s behavior-driven principle is that the test should name the interface and behavior it promises (TESTING.md:37-43); here the promised model/YAML binding omits one of the changed operational facts.

LOCKED DESIGN DECISIONS — N/A

No locked design document is edited in this diff. The PR works inside the existing v4 CI-as-data direction; the concrete mismatch is the incomplete live workflow binding above, not a change to a locked decision.

TRACKED vs UNTRACKED DEBT — Finding

The bridge has a named dissolution trigger at src/v4/workflow/ci.dag:83, but it is missing one of the bounds of the scaffold. continue-on-error is modeled through non_blocking: Bool (src/v4/workflow/ci.dag:88, .github/workflows/ci.yml:233), while the step timeout is only in YAML (.github/workflows/ci.yml:234). A tracked bridge needs documentation, bounds, and a dissolution trigger; this has documentation and trigger, but not the full modeled bound.

2.5. Top-down PM intent review

Finding. The highest-level authority says CI is modeled as .dag workflow data (THESIS.md:232) and the v4 workflow authorities name src/v4/workflow/ci.dag as the CI pipeline authority, with committed ci.yml as a checked projection (docs/design-pure-bootstrap-zero.md:7). This PR follows that intent in broad shape, but dilutes it at the projection boundary by adding timeout-minutes: 20 in committed YAML (.github/workflows/ci.yml:234) without carrying that fact in the modeled live signal (src/v4/workflow/ci.dag:84-88). A worker following the model faithfully could miss or drift the timeout while still passing the new binding test.

3. Verdict

REQUEST_CHANGES

The PR is directionally right and most of the wiring is coherent, but the live CI bridge is not single-authority yet: the YAML probe step has a timeout bound that the .dag model and binding test do not carry. Model and assert that bound in the bridge, then this should be mergeable.

…2 emitte

briansrls · 2026-05-25T14:57:31Z

openai-pro RC (timeout outside M1CiLiveWorkflowSignal): fixed

Verified on 69ed7105: .github/workflows/ci.yml had timeout-minutes: 20 while M1CiLiveWorkflowSignal omitted it.

Fix (this commit): add timeout_minutes: Int (= 20) to m1_ci_live_workflow_signal; binding test reads it and asserts timeout-minutes: {timeout_minutes} on the probe step block.

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 22d9e5eb · Trigger: schedule
Thinking: 262s wall

BLOCKING (1)

Root Cause

scripts/v4-m1-rust-emit-probe.sh summary extraction treats grep matches as guaranteed evidence → make zero-match histograms/counts total before strict/non-strict exit decisions.

ROADMAP — Verified

src/v4/TASKS.md T-24: The PR keeps the M1 probe scoped to workflow/ci.dag as CI pipeline-as-data with a tracked live-workflow bridge.

⚠️ Fix the zero-match rustc log parsing path so the informational probe exits 0 when strict mode is off.

Sign in to view

+      | sort | uniq -c | sort -rn | head -25 || true
+  )"
+  rustc_files_with_errors="$(
+    grep -oE '\-\-> src/[^:]+\.rs' "$rustc_log" 2>/dev/null \


grep exits 1 with no matches; pipefail + set -e aborted the non-strict probe before summary/exit 0. Guard rustc file-path extraction and default grep-backed counts to zero (codex RC @ 22d9e5e). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls · 2026-05-25T15:17:45Z

codex RC (grep zero-match aborts non-strict probe @ scripts/v4-m1-rust-emit-probe.sh:128): fixed

Verified on 22d9e5eb: with set -euo pipefail, the rustc_files_with_errors pipeline (grep -oE '\-\-> src/...' | … | wc -l) exits 1 when cargo check has no --> src/*.rs locations — aborting before tee "$summary" and breaking modeled non_blocking / P3/P5.

Fix: append || true to that pipeline; default grep-backed counts (rustc_error_total, rustc_files_with_errors, v2_error_lines) to 0 when empty; brief comment on grep+pipefail contract.

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 212916b0 · Trigger: schedule
Thinking: 302s wall

BLOCKING (1)

Root Cause

scripts/v4-m1-rust-emit-probe.sh rustc probe attempted/skipped state is collapsed into the POSIX status code 127 → carry a separate attempted/skipped/skip_reason fact and make strict mode fail on skipped-after-success or any attempted nonzero cargo-check status.

ROADMAP — Verified

src/v4/TASKS.md T-24: The PR keeps the M1 CI bridge scoped to workflow/ci.dag and the live workflow projection as CI pipeline-as-data under the T-24 lane.

⚠️ One strict-mode status-state bug remains in the new probe script.

Sign in to view

+  )"
+fi
+
+rustc_status=127


briansrls · 2026-05-25T16:21:19Z

Inline RC @ scripts/v4-m1-rust-emit-probe.sh:132 (grep zero-match + pipefail): already fixed on HEAD 212916b037

Verified on current HEAD: the rustc_files_with_errors pipeline now ends with || true and defaults via rustc_files_with_errors="${rustc_files_with_errors:-0}" (lines 131–135). Comment at 123–124 documents the grep+pipefail contract.

Repro: cargo-check log with error[E####] but no --> src/*.rs lines → assignment completes, files=0, probe reaches summary and non-strict exit 0.

No further code change needed for this finding.

— sent from quick-moth-204

…2 emitte

briansrls · 2026-05-25T16:22:55Z

codex RC (rustc_status=127 conflates skip vs failure @ scripts/v4-m1-rust-emit-probe.sh:99): fixed

Verified on 212916b0: rustc_status=127 doubled as “skipped” and real exit 127; strict mode treated 127 as success (-ne 127 guard).

Fix: explicit rustc_attempted / rustc_skipped / rustc_skip_reason; rustc_status only set when cargo check runs. Summary emits attempted/skipped/skip_reason or exit. Strict mode: fail on rustc_skipped after compile ok; fail on any attempted nonzero rustc_status (including 127).

— sent from quick-moth-204

briansrls

Review metadata

Provider / model: codex / unknown
Commit: d7222c17 · Trigger: schedule
Thinking: 255s wall

ROADMAP — Verified

src/v4/TASKS.md T-24: The PR keeps the M1 CI bridge scoped to workflow/ci.dag plus the live GitHub Actions projection while #98/#100 remain the named dissolution gates.

✅ Mixed CI wiring, shell probe, Rust receipt, and .dag model changes look clean; no blocking concerns found.

briansrls added 2 commits May 25, 2026 00:49

WIP: M1 CI wiring: add --target rust src/v4 → rustc step to surface v…

7079c6f

…2 emitte

WIP: M1 CI wiring: add --target rust src/v4 → rustc step to surface v…

de3d350

…2 emitte

briansrls marked this pull request as ready for review May 25, 2026 04:52

fix(ci): mark scripts/v4-m1* as v4-affected for M1 rust emit probe

087d8f6

PRs that touch only the M1 probe script must still run the gated CI step; mirror the existing scripts/v4-mvp1* selection rule. Co-authored-by: Cursor <cursoragent@cursor.com>