Emitter: declare libc in emitted Cargo.toml (fresh-emit cargo-green)

[briansrls]

Re-earns the post-#6243 fresh-emit re-measure receipt

The prior re-measure receipt was never merged to main — it lives only on session/neat-hawk-785 (an older pre-fork branch). So the receipt was still owed on main; this re-earns it and closes the last emitter gap.

Re-measure (non-destructive regen_stage0 --emit-fresh + cargo build, on main tip)

	rustc errors
Original receipt (2026-07-05)	1667
Fresh emit today	3 — all E0433, one root: libc
Fresh emit + libc	0 — cargo-green

The emitter restoration already on main (the field_access_field_is_boxed / shared_types predicate) closed ~all of the 1667. The sole remaining gap: emit_cargo_toml omitted libc from the emitted [dependencies], but the emitted crate bundles the copied host-physics v1_interpreter.rs, which uses libc::clock_gettime(CLOCK_THREAD_CPUTIME_ID) for the fast-lane eval budget — so a fresh crate is cargo-RED without it. (The committed stage0 Cargo.toml already carries libc directly; this brings the emitted manifest into parity.)

The fix

One dep row in emit_cargo_toml's base_deps (src/v1/05_emit_rust.dag):

+    emit_cargo_dep(name: "libc", version: "0.2", features: [])

Verified by execution (not deduction)

• gen-1: the freshly-emitted v1_compiler_emit_rust.rs now emits the libc row (absent in baseline).
• gen-2: running emit_cargo_toml through the interpreter over the live .dag yields a Cargo.toml containing libc = "0.2".
• New execution witness dag/test/claim/emit_cargo_toml_libc_witness_test.dag (PASS) asserts this — RED control: drop the libc row → witness reds.

Not included (operator-gated)

The full write-mode cutover — regenerating all ~40 seed files (~4.3k lines of accumulated drift) and un-ignoring the witnesses — is a separate, heavier decision. This PR lands only the targeted emitter fix + its witness.

🤖 Generated with Claude Code

[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.

[briansrls]

What this change does

Adds one dependency row — libc = "0.2" — to the emitted crate's Cargo.toml, authored in emit_cargo_toml's base_deps (src/v1/05_emit_rust.dag).

Why: the emitted crate bundles the copied host-physics v1_interpreter.rs, which calls libc::clock_gettime(CLOCK_THREAD_CPUTIME_ID) for the fast-lane eval-budget clock. The emitted manifest didn't declare libc, so a fresh-emit crate was cargo-RED with 3× E0433. The committed stage0 Cargo.toml already carries libc directly; this brings the emitted manifest into parity.

Impact: this was the sole remaining gap in the fresh-emit re-measure — 1667 → 3 → 0 errors (cargo-green). The rest of the original 1667 was already closed on main by the emitter restoration (field_access_field_is_boxed / shared_types predicate).

Verified by execution: gen-1, the freshly-emitted v1_compiler_emit_rust.rs now emits the libc row; gen-2, running emit_cargo_toml through the interpreter over the live .dag yields a Cargo.toml containing libc = "0.2". The new witness emit_cargo_toml_libc_witness_test.dag asserts this and reds if the row is dropped.

Scope is just the emitter fix + witness — not the full write-mode cutover.

[briansrls]

Follow-up: regen fixed-point (CI regen_verify_gate)

The first push redded regen_verify_gate_passes — as expected for a .dag emitter change: the fresh self-compile of emit_cargo_toml now emits the libc row, but the committed seed didn't have it yet (the two-generation regen lag). Fixed by regenerating the stage0 seed (regen_stage0):

• Exactly one generated file changes — src/v1/stage0/src/v1_compiler_emit_rust.rs (the emitted emit_cargo_toml gains the + emit_cargo_dep("libc", "0.2", ...) line).
• regen_stage0 --verify → regen_divergence_count=0 (committed seed now matches fresh self-compile).
• v1-compiler builds clean. (The v1-stage0-std-core / std_currency build error some targets show is pre-existing on main, unrelated to this change — verified against the un-patched baseline.)

Commit ddd1c09.

[briansrls]

Follow-up 2: floor witness source-root fix

The ci floor redded dag_compile_clean_gate — my new emit_cargo_toml_libc_witness_test.dag imported v1.compiler.emit_rust, but the floor compiles dag/test/claim with source roots dag + src/v2 only (not src/v1), so the import was unresolvable. (I'd resolved it locally with --source-root src/v1, which isn't the CI path.)

Fix (commit d7afe8c):

• Removed the import-based witness.
• Folded the guard into the existing v1_source_audit_witness_test.dag, which reads v1 source as text via filesystem_io (ReadsLiveTree) — the floor-compatible pattern for auditing v1.compiler source. emit_generates_cargo_toml now also asserts the libc dep row (RED control verified: drop the row → witness reds).

Execution coverage of the emit itself stays with regen_verify_gate (the .dag emitter → committed seed, byte-exact — green after the regen in follow-up 1).

[briansrls]

Follow-up 3: remaining ci red is a pre-existing main-red, not this PR

The ci run redded on non_fold_residue_no_unrostered_or_stale (src/v2/lens/non_fold_residue_test.dag). Investigated — it is not caused by this PR:

• The lens passed on my branch built from this PR's base, but the pull_request job tests merge-with-current-main. Main advanced 11 commits since my base, including PR-time detectors: CommitWitnessClaim roster-resolvability check + always-on non-fold-residue receipt, with RED controls #7085 "always-on non-fold-residue receipt", which rewrote the census (cli_run.rs +467) and made it gating.
• Reproduced locally by merging current main + rebuilding: census fails. Reverting all of this PR's .dag edits to main's versions → still fails. The census scans only .dag files (not my regenerated seed .rs).
• Decisive: ran the census against a pristine checkout of current main (1ee1221c) → FAIL. Main itself is red on this gate.
• The flagged unrostered sites are pre-existing dag/extdeps/** wildcard-arm witnesses (container/oci/ctrl_session_witness.dag, formats/elf/hello_static_witness.dag, cron/schedule_model.dag, …) — none touched by this PR, and unchanged across the 11 main commits. PR-time detectors: CommitWitnessClaim roster-resolvability check + always-on non-fold-residue receipt, with RED controls #7085 made the census gating while these sites were unrostered.

So this PR's own changes are green (emitter fix + regen fixed point + floor-compatible witness); the red is inherited from main and will clear once main's residue roster is reconciled (owner of #7085), independent of this PR.