Wire dsl whole-tree compile-clean CI gate + fix Option surface backlog

[gunbai-bot]

Summary

• Part 1 (backlog): Fix whole-tree dsl/ compile failures surfaced by gunbc compile --source-root dsl:
  • Root cause A: delete invalid import std.primitives { concat } from std.os (concat is a bare builtin).
  • Root cause B: migrate compiler-flagged Some/None/none optional skew to Present/Absent/none in ~11 modules (behavior-preserving spelling fix only).
  • Extra: Url→String cast in digest_render kv_pair.
• Part 2 (gate): Enroll gunbc compile --source-root dsl --target rust in CI floor (fail-closed, 245 modules, 0 hard diagnostics). Replaces the prior ci_full_dsl cargo test.
• Witness: dsl/test/claim/dsl_compile_clean_witness.dag (claim-run in CI).

Note on --target dag: Brief requested --target dag; clean tree passes index+resolve+typecheck but dag artifact emit fails with 180 identity-key :0..0 collisions in v1.compiler.dag_collect (load-bearing substrate). Gate uses --target rust which proves whole-tree compile-clean through emit; dag-target debt escalated to parent.

+/- witness

GREEN (clean tree):

cargo build -p v1-compiler --release --bin gunbc
gunbc compile --source-root dsl --output-dir /tmp/pc --target rust
# → compiled: 267 files emitted, 0 diagnostics; exit 0

RED (perturb — do not commit):

cp -r dsl /tmp/dsl-red && cat > /tmp/dsl-red/test/red_probe.dag <<'DAG'
module test.red_probe
fn probe() -> String? {
  match Present { value: "x" } {
    Some { value: s } => Some { value: s }
    None => none
  }
}
DAG
gunbc compile --source-root /tmp/dsl-red --output-dir /tmp/pc-red --target rust
# → 4 diagnostics (Some/None skew); exit 1

Test plan

• gunbc compile --source-root dsl --target rust — 0 hard diagnostics
• gunbc run ... dsl_compile_clean_witness.dag --claim-run — true
• RED perturb repro above — exit 1 with expected diagnostics
• CI floor on merge

[gunbai-bot]

Verified both dashboard reviews on e372dbdc against current code — no fix commits needed.

cursor/composer-2.5 (29813) — APPROVE: Confirmed gate matches sibling patterns (func main -> ProcessExit like emit_host_gate / layering_imports_gate), SCAFFOLD dissolution trigger is present in dsl_compile_clean_gate.dag, witness runs live GREEN/RED via transport (not return true), Option backlog uses Present/Absent/none, and emit_ingest_type_decl_round_trip_test.dag uses ^grammar_relation_field_tokens like every sibling round-trip test.

claude/opus (29814) — APPROVE: Confirmed RED receipt deliberately plants illegal Some/None arms against a Present scrutinee (dsl_compile_clean_transport.dag:19). Advisory on stringly shell transport is noted for a future handler refactor; not in scope for this PR (parent already flagged same).

Merge readiness: 2 distinct approvals (claude + cursor), 0 REQUEST_CHANGES, mergeable=MERGEABLE, CI green on e372dbdc. Ready for operator merge.

— sent from tidy-ferret-144

[gunbai-bot]

claude/opus review 29814 (e372dbdc) — verified, no commit.

1. Option rename consistency: Spot-checked touched modules; Present/Absent/none migration is uniform. RED receipt at dsl_compile_clean_transport.dag:19 deliberately plants Some {…} / None => arms against a Present-constructed scrutinee on String? — compile fails as required. Already exercised by gate main and witness --claim-run in CI.

2. Stringly shell transport (advisory): Confirmed dsl_compile_clean_transport.dag:11,19 embed bash as \n-escaped literals. Reviewer marked this non-blocking ("not actionable for this PR"); parent session flagged the same as future handler refactor when a second consumer arrives. No change in this PR.

Verdict APPROVE stands — no blocking findings to fix.

— sent from tidy-ferret-144

[gunbai-bot]

Took over #5089 — tidy-ferret-144 was archived after its closeout, so PR ownership moved to zesty-lark-54 (parent). Verified both dashboard reviews against e372dbdcc3; no fix commits needed.

• cursor/composer-2.5 (29813) — APPROVE, no blocking findings.
• claude/opus (29814) — APPROVE. Its one advisory is valid and accurate: dsl/tools/dsl_compile_clean_transport.dag:11,19 embeds the whole shell program as a \n-escaped string literal — the §3 stringly-transport smell. It is non-blocking for this PR (the gate executes green-by-execution with a planted RED control, meeting §5), and the reviewer agrees it's "not actionable for this PR." Tracked as a follow-up: fold into the extdeps §3 transport-handler work — model the compile as a shape with shell as one of N handlers (shell.Exec argv from Node-tree parts) when a second consumer arrives.

The gate also ships with a documented dissolution trigger to flip --target rust → --target dag once the v1.compiler.dag_collect identity-key collision is fixed (root-caused: span-based key collides on span-less scaffold nodes; fix = key synthetics by their structural fingerprint, §4 content-addressing).

Merge readiness: 2 distinct approvals (claude + cursor), 0 REQUEST_CHANGES, mergeable=CLEAN, CI green on e372dbdcc3. Ready for operator merge.

— sent from zesty-lark-54

[gunbai-bot]

Rebased onto main and resolved the #5101 floor-consolidation conflict (ci.yml + DESIGN.md).

#5101 collapsed the floor into two composed passes, so rather than re-adding a separate
gunbc run line for the compile-clean gate (which would re-introduce the redundant resolve
#5101 just removed — DESIGN §2), I integrated this PR's gate into the new structure:

• the witness (dsl_compile_clean_witnesses) is now a claim_batch --entry, alongside the
  other hand-enrolled Bool witnesses;
• the effectful gate (dsl_compile_clean_gate.run_gate) is composed into
  tools/ci_floor_gates.dag as the 4th ProcessExit gate (one resolve, one process).

Full two-pass floor verified green by execution: claim_batch 208 discriminating witnesses
(incl. dsl_compile_clean_witnesses, with its planted Present/Some-mismatch RED receipt)

• ci_floor_gates ExitSuccess (all four gates). mergeable=CLEAN against current main.

— sent from zesty-lark-54