Regime-2 emission: collapse yaml/gitignore/runner_deploy serializers into one std.layout render fold

[gunbai-bot]

Regime-2 emission: one shared fold over a std layout IR

Collapses the three boutique pure-projection (regime-2, emit-only) serializers into one fold over a shared layout IR, per docs/plans/regime2-shared-emission-fold.md. DESIGN refs: §2 (no duplicated layout logic — one concept, every format), §3 (single layout authority in std), §4 (emit = ingestion⁻¹, the forward-only subset), §5 (emit-only → no round-trip oracle; honesty at the boundary), §6 (DFS before minting).

The shape

expected_X = render(project_X_to_doc(authority), X_protocol)

• dsl/std/layout.dag (new) — the emission model: Doc = DocEmpty | DocText | DocLine | DocConcat | DocNest, LayoutProtocol { indent_unit }, and one render fold. The renderer inserts indentation lazily (before the first text of each line) so it reproduces yaml's line-prefix indentation byte-for-byte with no trailing-space artifacts.
• The discriminator holds: render has zero format-specific branches (5 structural cases). All format-specifics live in project_X_to_doc (quoting/escaping/comment-prefixes baked into DocText) and the thin LayoutProtocol. Asserted by layout_render_witness_test (one fold drives flat + nested; nest/indent perturbations go RED).

The three projections

• gitignore → project_gitignore_to_doc (dsl/gunbc/gitignore_emit.dag), indent_unit: "".
• runner_deploy → project_runner_deploy_to_doc (dsl/gunbc/runner_deploy_emit.dag), indent_unit: "".
• yaml → project_yaml_to_doc (dsl/extdeps/formats/yaml.dag), indent_unit: " ". Scalar quoting/escaping/flow-seq kept as text builders, baked into DocText; nesting/block-seq/multiline via DocNest+DocLine. The old prefix_block_sequence_mapping empty-line-skip is reproduced by construction (multiline blocks emit no wrapper blank).

Public entry signatures preserved (expected_gitignore / expected_ci_yml / expected_runner_deploy_manifest) — gentle-ibex's universal generated-artifact gate (#5568) is unaffected.

Faithfulness — byte-identical (verified by execution)

• .gitignore (3280 B) and .github/workflows/ci.yml (11723 B): regenerated output is byte-identical to committed (git diff empty). Runner manifest: no committed file (computed; drift-gated) — its witness passes unchanged.
• All existing witnesses green: ci_yaml_serializer_keystone_holds, gitignore_witnesses, runner_placement_holds. New: layout_render_keystone_holds, yaml_quoting_keystone_holds (sharpening B — reserved/double-quoted/escaped/empty scalar paths that ci.yml does not exercise).

§3 single-authority cleanup (sharpening A)

Removed the dead flat layout IR so std has one layout authority: std.types Document/DocumentLine/DocumentSection/DocumentFile, the shared.dag_util document builders/render_document, and the seed std_types.rs Document/DocumentLine/DocumentSection/TextFile(wrapping Document) cluster — all verified zero-caller across dsl/, src/v1, src/v2. (The removed seed TextFile is distinct from the live v1.std.core.TextFile = path+content.)

Scaffold honesty + dissolution

std.layout Doc + render are seed-realized scaffold; the dissolution trigger is the v2 TargetModel grammar-row inverse (serialize_target) subsuming them at self-host — the Doc/protocol spellings are a forward-only subset of those rows. (This note lives here, not in-source: .dag comments are banned tree-wide post-#5567 parser-wall flush.) Regime-2 is emit-only — no round-trip oracle is claimed; faithfulness rests on the authored projection (§5).

Out of scope (untouched): std.markdown / roadmap_emit (bright-stag's regime-1 lane).

🤖 Generated with Claude Code

[gunbai-bot]

Thanks for the review. Addressing the items:

Stale prose (docs/plans/emission-ingestion-inverse.md:218 names the deleted emit_yaml_value/emit_mapping_body) — valid; non-blocking, and the head is currently frozen for approval accrual (per my manager session), so I'm tracking it as a post-merge follow-up rather than resetting reviews/CI with a doc-only push.

One additional item I found in my own pre-review (disclosing for transparency, also post-merge follow-up): emit_inline_scalar_text dropped the old emit_scalar_string multiline |-block branch, so a multiline YamlString outside mapping-value position (a bare sequence element or top-level scalar) now emits embedded raw newlines instead of a block literal. It is latent / zero-trigger: serialize_yaml's only consumer is ci.yml, whose projection only ever places multiline strings as mapping values (run:/path:/restore-keys:), which take the preserved multiline_entry_doc path. ci.yml + .gitignore stay byte-identical and all witnesses are green. The follow-up will either route those arms through the block-literal too, or fail-closed per §5 rather than emit broken YAML.

— sent from tidy-crane-536

[gunbai-bot]

Update — both items fixed in 749fd76 (not post-merge after all):

1. Multiline |-block regression — serialize_yaml now emits a block literal for multiline strings in all positions (mapping value, bare sequence element, top-level scalar) via a shared multiline_block_doc helper, and all_scalars_flowable now excludes multiline strings so they take the block path rather than a broken flow […]. New yaml_multiline_block_witness_test asserts the block literal in each position and goes RED on the embedded-newline form. ci.yml + .gitignore remain byte-identical (the mapping-value path is unperturbed).
2. Stale prose — docs/plans/emission-ingestion-inverse.md:217 repointed off the deleted emit_yaml_value/emit_mapping_body; it now describes serialize_yaml rendering project_yaml_to_doc over the shared std.layout Doc fold (the paragraph's round-trip-oracle argument is preserved).

— sent from tidy-crane-536