Repository navigation
Rebase #5135 (fix/infer-perf-body-producer) onto main + resolve the §3/§7 REQUEST_CHANGES: model substitute_generics_apply identity short-circuit in 04_infer.dag + regen seed - #5146
Merged
Conversation
…esses Root cause for 244s body_producer resolve was decode_value_expression_projection_bundle (15-deep bind_outcome in target_model, ~287s analyze_item), not body_producer itself. Shard into one-bind chain steps (244s→~23s). Execution instrumentation confirms H2a: substitute_generics 13446 calls / 169 distinct sites (ratio 79.6) — memo deferred to merry-owl cache-kernel consolidation per §3. Adds §5 witness (bounded resolve + wrong-type RED). Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Accidental removal of Commands::Ci and extdeps_shape_transport_policy_project broke CI (`gunbc ci` unrecognized). Restore both alongside generic-inst instrumentation. Co-authored-by: Cursor <cursoragent@cursor.com>
Keep generic_instantiation_instrument and interface_elaboration_instrument modules from the perf-fix branch. Co-authored-by: Cursor <cursoragent@cursor.com>
…ducers Empty `diagnostics: []` failed resolve (Diagnostics is None | Some, not List). Co-authored-by: Cursor <cursoragent@cursor.com>
Remove generic_instantiation_instrument and interface_elaboration_instrument (hand-Rust scaffolding with no dissolution receipt). H2a localization and the structural perf fixes are proven; hooks added per-call overhead on substitute_generics/build_type_env/typecheck_module. Guard enrollment tracks in follow-on #5139. Co-authored-by: Cursor <cursoragent@cursor.com>
substitute_generics uses map_is_empty(subst) in 04_infer.dag; without builtin_function_registry entry it fell through to unit_type() and diverged from hand-synced Rust. Wire .dag authority + seed + interpreter. Co-authored-by: Cursor <cursoragent@cursor.com>
4 tasks done
… 04_infer.dag Resolves the §3/§7 REQUEST_CHANGES: the seed's substitute_generics_apply preserved Node identity via a short-circuit (Rc::ptr_eq on the resolved inferred node + rc_vec_ptr_eq on children/params) that the .dag never modeled and could not emit — a second authority for infer behavior in a "do not edit" generated seed. Model it in 04_infer.dag as a SOUND IDENTITY HINT (true => structurally equal): recurse children/params/inferred, return the original n when all are unchanged, else rebuild. Surface the primitives as ordinary builtins with OWNED Rc signatures (cheap refcount bump preserves inner pointers), so they emit through the normal clone path — no all-args-by-ref emit rule and no emit-logic change: - runtime_rust.dag: add rc_ptr_eq, flip rc_vec_ptr_eq to owned + contract - extdeps/languages/rust/emit.dag: register map_is_empty, rc_ptr_eq, rc_vec_ptr_eq in rt_function_registry (map_is_empty also never round-tripped before — fixed here) - 04_method.dag: register rc_ptr_eq/rc_vec_ptr_eq result types - v1_interpreter.rs: value-equality realization of the same contract Seed regenerated via regen_stage0 from the .dag (round-trip proven: the emitter now produces v1_rt::rc_ptr_eq(new_rt.clone(), rt.clone()) etc.). Narrow seed update only — a blanket regen also rewrites unrelated files (emit_rust.rs wire_variant_tag_for_policy, extdeps_cargo.rs) byte- identically on main too, a pre-existing emit non-fixpoint tracked separately; adopting it would regress the seed (broken /* Type */ placeholder), so it is out of scope here. Also fixes pre-existing PR-branch CI dirtiness: cargo fmt on build_type_env + the perf witness/profile tests, and #[cfg(test)] on the profile test module (the lone module missing it, which failed clippy --all-targets). Proof: crate builds, clippy --all-targets -D warnings green, fmt clean, body_producer_infer_perf_witness green (clean resolve + wrong-type RED). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
briansrls
force-pushed
the
work/5135-s3-fix
branch
from
June 18, 2026 11:51
d9ef67e to
3d0aa5f
Compare
Contributor
Author
§3/§7 resolution + stash-regen control (why narrow, not blanket-regen)The §3/§7 finding is resolved: This is a narrow seed update, not a blanket Control (proves the collateral is pre-existing, not induced by this change):
Proof bar: crate builds; |
This was referenced Jun 18, 2026
Merged
briansrls
added a commit
that referenced
this pull request
Jun 19, 2026
…nblock) resolve_expr_types double-resolved every child in its variable-arity arms (ExprCall/RecordLit/MethodCall/Match/StringInterp): one traversal built the resolved children, a sibling traversal re-resolved the SAME children to re-collect diagnostics. A nested constructor's tail child is itself such a node, so each level doubled the recursion -> O(2^depth) (#5146-class). This was the N_v2 self-host blocker: resolve_entry_graph hung >800s on the generated host_source_root_ingest_manifest, whose host_compiler_closure_admission .imports is a ~20-deep Cons chain (195s, constant in record count) and whose host_source_root_ingest is a 59-deep chain. Fixed in BOTH the .dag authority (src/v1/04_resolve.dag) and the v1 seed (v1_compiler_infer_resolve.rs): bind the per-child result ONCE, then split map(.expr)/flat_map(.diagnostics) — the shape the in-file ExprListLit/ExprBlock siblings already used. Proof by execution: N_v2 substrate resolve 279s (was >800s exit-124); scaling probe linear (n=20 hang -> 87s); purity oracle green (resolve_typed_cache_ equivalence, identical self_gen8 pass/fail split pre/post) — result-identical, only faster, since resolve_expr_types is pure (resolving once == twice). Guard: resolve_expr_types_retraversal_guard_test source-bridges the real 04_resolve.dag Node and flags any arm with >=2 sibling self-calling traversals (RED on the real pre-fix source: 5 arms; GREEN post-fix). Enrolled in the CI rust gate via ci_spec. The 9 nv2_gate_resolve_bisect scaffolds dissolve per their root-caused trigger; a cleaned scaling probe remains as the perf harness. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
briansrls
added a commit
that referenced
this pull request
Jun 19, 2026
…nblock) (#5258) * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * Fix O(2^depth) resolve_expr_types: bind-once child resolution (N_v2 unblock) resolve_expr_types double-resolved every child in its variable-arity arms (ExprCall/RecordLit/MethodCall/Match/StringInterp): one traversal built the resolved children, a sibling traversal re-resolved the SAME children to re-collect diagnostics. A nested constructor's tail child is itself such a node, so each level doubled the recursion -> O(2^depth) (#5146-class). This was the N_v2 self-host blocker: resolve_entry_graph hung >800s on the generated host_source_root_ingest_manifest, whose host_compiler_closure_admission .imports is a ~20-deep Cons chain (195s, constant in record count) and whose host_source_root_ingest is a 59-deep chain. Fixed in BOTH the .dag authority (src/v1/04_resolve.dag) and the v1 seed (v1_compiler_infer_resolve.rs): bind the per-child result ONCE, then split map(.expr)/flat_map(.diagnostics) — the shape the in-file ExprListLit/ExprBlock siblings already used. Proof by execution: N_v2 substrate resolve 279s (was >800s exit-124); scaling probe linear (n=20 hang -> 87s); purity oracle green (resolve_typed_cache_ equivalence, identical self_gen8 pass/fail split pre/post) — result-identical, only faster, since resolve_expr_types is pure (resolving once == twice). Guard: resolve_expr_types_retraversal_guard_test source-bridges the real 04_resolve.dag Node and flags any arm with >=2 sibling self-calling traversals (RED on the real pre-fix source: 5 arms; GREEN post-fix). Enrolled in the CI rust gate via ci_spec. The 9 nv2_gate_resolve_bisect scaffolds dissolve per their root-caused trigger; a cleaned scaling probe remains as the perf harness. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
briansrls
added a commit
that referenced
this pull request
Jun 19, 2026
…boration hangs resolve_entry_graph on the v2 compiler closure (the N_v2 interpreter blocker; #5146-class), then fix the root (#5266) * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor * N_v2 gap-2: relax dag comma-list to optional inter-item comma (mirror v1) The v2 dag LanguageModel's shared list combinator dag_grammar_comma_list_expr required a MANDATORY comma between items, forking from the v1 single authority parse_field_list_acc (src/v1/stage0/src/v1_compiler_parse.rs) which separates items on a comma OR bare juxtaposition (comma optional; newlines are whitespace). The fork left every newline-separated multi-field record/type body unparseable (parse_g0_tokens_remain) — the next located N_v2 self-host parse blocker after the caret lex gap, pinned by prefix-bisection of 00_compile.dag to `type Validated`. Fix: make the inter-item comma optional in the one shared combinator — item (comma? item)* comma? — a strict superset (comma lists still parse) that fixes all 13 callers (field decls, param lists, generic/type-arg lists, field init/pattern, arg lists) at once, mirroring v1's uniform optional-comma. Floor witnesses (green by execution): newline-separated multi-field record + variant record payload Accept; one-field + comma-separated still Accept (superset); and a DISCRIMINATING no-fail-open control — a field missing its colon still Rejects, proving the relaxation made the COMMA optional, not the field SHAPE permissive. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: Resolve-perf bisect+fix: pin which direct-import's full-interface elabor --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (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.
Auto-opened by session-dashboard for session
crisp-wolf-348.Pushing to
work/5135-s3-fixadvances this PR.Worker attestation
Before flipping this PR to ready for review, confirm each item:
npm test,cargo test) and the result.Closes #Ndirective.Summary
TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.
Test plan