Repository navigation
Ground emit-path variant-owner determinism in .dag authority - #5913
Merged
Merged
Conversation
…grounding) Port #5879's variant-owner disambiguation back into .dag authority and hand-sync the v1 Rust seed: imported_variants (source-module local items), alpha-sorted variant_fold bindings, sorted export-set map_keys, and field-type-hint struct disambiguation. Regen in #5873 overwrote these fixes; this re-lands them on current main (#5899) without disturbing expected_type_override_enum (#5891). Co-authored-by: Cursor <cursoragent@cursor.com>
Re-land imported_variants + sorted variant_fold on top of #5899's func_sigs/all_declared_sigs path (not the reverted imported_sigs fork). Restores compile-clean hand-sync; variant_owner_disambiguation 3/3 green. Co-authored-by: Cursor <cursoragent@cursor.com>
…e regen. Fix imported_variants map type so regen emits valid Rust (Map<String,String> for unambiguous owners only), then regen stage0 to match .dag fixed point. Co-authored-by: Cursor <cursoragent@cursor.com>
gunbai-bot
Bot
changed the title
3 tasks done
briansrls
added a commit
that referenced
this pull request
Jun 29, 2026
Close FLAGS 1/2/5 as decided; keep 2 genuine operator FLAGS. Adopt DeterminismAxis one-core-two-refinements (§3 probe). State additive relationship to #5913 as first roster construction-instance. Co-authored-by: Cursor <cursoragent@cursor.com>
1 of 3 tasks
2 of 4 tasks
briansrls
added a commit
that referenced
this pull request
Jun 29, 2026
* docs: §5 determinism mechanism design (shape-sign draft) Model how v2.std.determinism graduates from inert carrier to a signature-derived classification axis orthogonal to EffectShape, with authority boundaries, phasing, and operator FLAGS — no std/lens edits yet. Co-authored-by: Cursor <cursoragent@cursor.com> * docs: sharpen determinism design per manager review Close FLAGS 1/2/5 as decided; keep 2 genuine operator FLAGS. Adopt DeterminismAxis one-core-two-refinements (§3 probe). State additive relationship to #5913 as first roster construction-instance. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(ci): wire determinism design doc into doc-graph + DESIGN authority Link docs/plans/determinism-mechanism-design.md from DESIGN.md open threads (doc reachability) and sync design_document.dag (generated-artifact drift gate). Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Cursor <cursoragent@cursor.com>
briansrls
pushed a commit
that referenced
this pull request
Aug 28, 2026
… run to run, and a two-draw control could not have told the difference (#9496) * Retract a causal story I put on main: the emitter's pub use ordering varies run to run, my preamble-reorder diagnosis was never established, and a two-draw control could not have told the difference #9439 landed an annotation on emit_rust's preamble asserting that factoring the preamble moved emitted bytes, that "the emitter is stable given its source and NOT invariant under this reordering", and that restoring the original order restored byte-identity. THE FIRST HALF OF THAT SENTENCE IS FALSE AND THE REST IS UNSUPPORTED. Prose on main asserting a mechanism nobody established is premise contamination, and the next person to touch that preamble would have found a confident causal story and planned against it. WHAT IS ACTUALLY HAPPENING, one binary compiling one unchanged corpus six consecutive times (scoped emit of src/v2/compiler/00_compile.dag, 175 files): the differing-file count VARIES BY RUN -- 2, 0, 1, 0, 2. Exactly two files ever differ (v2_lens_enforcement_vocab.rs, v2_std_cross_tree_resolution.rs), each with exactly two distinct outputs; sorted lines are IDENTICAL in every differing pair, so this is REORDERING and not value nondeterminism; every changed line is a `pub use` line (2 of 2, 4 of 4); and after rustfmt both files are NORMALIZED-IDENTICAL. That is the known import-set ordering class (#5913; measured again on 03_ingest 2026-08-22), reported independently by another lane on main at 38a127b naming THESE TWO FILES with no contact between lanes. THE CENTRE OF THE REWRITE IS THE LESSON, NOT THE FINDING: a control over a probabilistic subject needs a stated sample size before it concludes anything, and two agreeing draws are not determinism. The same-source control was run twice, agreed twice, and was read as proof of determinism -- against a flip with roughly those odds it agrees about half the time, so it could not have detected the thing it was controlling for. That generalises past this file; the pub use finding does not. TWO CORRECTIONS STATED IN THE TERMS THAT MATTER. The reordering was never shown to move a byte, and was never shown innocent either -- so keeping the original binding order is NOT justified by the specimen given for it, and the annotation says so rather than quietly keeping the conclusion. And the claim that this undermines every byte-comparison gate including regen's fixed point is true in general and FALSE of this mechanism against that gate: the compared population is normalized, and normalization is exactly what removes pure use-statement reordering. WHAT THIS DOES NOT DO: it offers no theory for why a site grounded in June (#5913) varies again in August. That is unexplained, is stated as unexplained, and a correct retraction must not become a second causal story. The contribution is the localisation -- two named files, pure `pub use` order, two outputs each. ALSO IN THIS PR, from review 56672's non-blocking note on #9439: reference_derived_census now counts through one fold that dispatches on the coproduct instead of four filters over the rendered disposition NAME. Adding a fifth arm now breaks this function rather than being silently uncounted -- which, in a change whose subject is a population that goes uncounted in silence, was that defect reintroduced one level up. It also makes `candidates` the sum of the arms by construction, and the witness pins that. VERIFIED: required-regen first_generation_equal=true planned=138 executed=138 with NO drift against the committed mirror; all five witness rows PASS. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * Take #9537's repair instead of carrying it here: drop the three mirrors this branch did not originate The merge regenerated five stage0 mirrors -- the two this branch owns, plus v1_compiler_emit_core_support.rs, v1_compiler_emit_go.rs and v1_compiler_emit_python.rs, whose drift a pristine-main control showed to be main's and not this branch's. A dedicated PR (#9537) now carries exactly those four files at the same base, so carrying them here too would be two independent repairs of one drift -- the conflict class this branch spent the evening resolving. Also restores src/v1/stage0/src/bin/claim_executor.rs and src/v1/stage0/src/cli_run.rs to main's bytes: the regeneration dispatch tarred the whole stage0 source directory back from a runner whose checkout predated this merge, so those two hand-maintained files returned as pre-merge copies and silently dropped main's content. * Complete the census sum assertion over all six arms (review 57169) The #9466 merge extended the disposition coproduct to six arms and this file's fixture to six rows, and did not extend the sum clause. It read candidates == survived + own_module + registry_absent + export_proof_failed against a six-row fixture, so it asserted 6 == 4 and evaluated FALSE -- and it asserted the opposite of the property it exists to check, that the two variant arms are not part of the total. Nothing caught it because nothing ran it. The fold compiled, the per-arm equalities were all correct, the mirror regenerated, and required-regen reached first_generation_equal with fixed-point 0 -- six green signals, none of which evaluates a witness assertion. The regen gate proves the mirror matches the authority; it says nothing about whether the authority is right. Verified by execution rather than by inspection this time: all five rows run green against the emitted mirror (2 passed, 0 failed). --------- Co-authored-by: gunbc-ci-auto-heal <gunbc-ci-auto-heal@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Grounds the structurally-nondeterministic emit-path variant-owner pick and import-set ordering in the
.dagauthority (not a hand-synced seed restore of #5879).imported_variantsdisambiguation (04_infer.dag): per-variant owner lookup uses the source module's locally-defined items, not transitivetype_env.bindings. Map stores only unambiguous owners (Map<String, String>) so regen emits valid Rust.variant_fold: alpha-sortenv.bindingsbefore fold so shared-variant owner selection is stable across runs.05_emit_rust.dag): sortmap_keys(export_sets)at three call sites for stablepub useordering.find_struct_name_by_fields): field-type hints filter candidates when multiple structs share field names.Preserves #5891
expected_type_override_enum(per-site expected-type override).Proof (execution-grounded)
regen_stage0 --verify: green (seed is pure regen artifact of.dag)gunbc compile --source-root dsl --target rustx2,diff -rq: 0 (was 36 on main)cargo test -p v1-compiler-tests variant_owner_disambiguation: 3/3 greenTest plan
regen_stage0 --verifygreenvariant_owner_disambiguation3/3