Repository navigation

Declaration grafting: type/data decls never become Named edges — the emit-derisk red and XL-0's field gap are one defect - #11574

Merged

gunbai-bot[bot] merged 70 commits into

mainfrom

session/neat-moth-237

Sep 20, 2026

Commits on Sep 18, 2026
- Rehome type and data declarations onto Named graft edges.
  
  Brian Searls
  and
  cursoragent
  committed
  c0d77bc
- Use match for type/data unit dispatch so the optional Edge unifies.
  
  Brian Searls
  and
  cursoragent
  committed
  c7eb597
- Do not treat a type/data shell without a captured child as a declaration name.
  
  Brian Searls
  and
  cursoragent
  committed
  f8b6587
- Pair record field names with their types from user atoms under the field block.
  
  Brian Searls
  and
  cursoragent
  committed
  c773382
- Probe why record construction still refuses after type grafting.
  
  Brian Searls
  and
  cursoragent
  committed
  fcec4c2
- Keep only user-named flatten members so record field blocks are not module mentions.
  
  Brian Searls
  and
  cursoragent
  committed
  7139336
- Restore the record-construct assemble hold after the flatten filter.
  
  Brian Searls
  and
  cursoragent
  committed
  8bfb5c8
- Take the type name as the first user atom in the declaration, not sequence-left.
  
  Brian Searls
  and
  cursoragent
  committed
  80feda1
- Rehome unwrapped record types from ident + field-block sequences.
  
  Brian Searls
  and
  cursoragent
  committed
  b83df09
- Drop the duplicate Box record source and unused assemble-reason helper.
  
  Brian Searls
  and
  cursoragent
  committed
  267bc60
- Bind a pending type name to the next field-decl block in graft order.
  
  Brian Searls
  and
  cursoragent
  committed
  295036c
- Name record types from the type keyword then ident, and keep flatten only for functions.
  
  Brian Searls
  and
  cursoragent
  committed
  b65254d
- Keep flatten members whose targets are types or functions, not bare field atoms.
  
  Brian Searls
  and
  cursoragent
  committed
  50e87ba
- Discriminate the record-graft refusal by reason prefix.
  
  Brian Searls
  and
  cursoragent
  committed
  672f325
- Use the builtin starts_with for record-reason prefixes.
  
  Brian Searls
  and
  cursoragent
  committed
  f59380e
- Pin the record assemble refusal to remaining resolve reasons.
  
  Brian Searls
  and
  cursoragent
  committed
  0cb1f05
- Narrow the resolve-* refusal family for record graft.
  
  Brian Searls
  and
  cursoragent
  committed
  a176c6a
- Do not treat the module qualified name as a pending type during graft.
  
  Brian Searls
  and
  cursoragent
  committed
  269ddb0
- Probe empty record braces separately from fielded records.
  
  Brian Searls
  and
  cursoragent
  committed
  3f4e166
- Record field bags are Cardinality so a type-only record is not a graft-spine segment.
  
  Brian Searls
  and
  cursoragent
  committed
  e9ee8c5
- Drop diagnostic assemble-reason probes; keep the owed record controls.
  
  Brian Searls
  and
  cursoragent
  committed
  8a1eaa8
- Keep record field types as type-expr nodes instead of bare atoms.
  
  Brian Searls
  and
  cursoragent
  committed
  3fbcf5e
- Enrol empty-record assemble controls; fielded Int records remain a residual.
  
  Brian Searls
  and
  cursoragent
  committed
  3972dce
- Wrap record Conj fields in Cardinality so a type-only record is not a QN spine.
  
  Brian Searls
  and
  cursoragent
  committed
  7d5848c
- Restore fielded record assemble rows now that the spine wrap holds.
  
  Brian Searls
  and
  cursoragent
  committed
  d88814d
- Stop compensating record rehome in graft; keep record assemble rows as standing RED.
  
  Brian Searls
  and
  cursoragent
  committed
  b0bd50f
- Drop lexeme-prefix machinery filtering and observe flatten-drop claims per-PR.
  
  Brian Searls
  and
  cursoragent
  committed
  0697996
- Restore LCR and XL-0 record absence measurements.
  
  Brian Searls
  and
  cursoragent
  committed
  614c429
- Keep residual record type_decl units when a sibling graft succeeds.
  
  Brian Searls
  and
  cursoragent
  committed
  39a95d4
- Note that skipped record type_decl units are reconstructed as residuals.
  
  Brian Searls
  and
  cursoragent
  committed
  4d73e34
- Consume dag production presence and module-metadata walkers; drop unused graft helpers.
  
  Brian Searls
  and
  cursoragent
  committed
  cc7c00f
- Execute record-graft measurements on per-PR discovery; require Rec to survive a successful assemble.
  
  Brian Searls
  and
  cursoragent
  committed
  ace921b
- Build the dag grammar once per graft and locate productions with one subtree walk.
  
  Brian Searls
  and
  cursoragent
  committed
  58d0cb4
- Read type/data names from the dag sequence spine; stop treating Rejected as not-named.
  
  Brian Searls
  and
  cursoragent
  committed
  45c10aa
- Point alias binding at the shared kw-then-ident spine.
  
  Brian Searls
  and
  cursoragent
  committed
  33e111d
- Skip Named rehome only for the record type_decl arm, not variant field blocks.
  
  Brian Searls
  and
  cursoragent
  committed
  f869a5a
- Unify parse_subtree_locate_production branches as Optional rather than if.
  
  Brian Searls
  and
  cursoragent
  committed
  75f569a
- Treat field_decl_block under a coproduct Disj as Named, not a record residual.
  
  Brian Searls
  and
  cursoragent
  committed
  c196aad
- Detect coproduct field blocks by a Disj anywhere under the type_decl unit.
  
  Brian Searls
  and
  cursoragent
  committed
  850f9bf
- Distinguish record types from fielded coproducts by `=` and Disj, not alias-rhs presence.
  
  Brian Searls
  and
  cursoragent
  committed
  c6dceb0
- Restore the field_decl_block Named skip; alias-rhs narrowing Named-rehomed records.
  
  Brian Searls
  and
  cursoragent
  committed
  b8957a6
- Dissolve the bool production predicate and the declaration-target walker.
  
  Brian Searls
  and
  cursoragent
  committed
  82c1fb8
- Restore find-captured descent when the first matching production has no capture.
  
  Brian Searls
  and
  cursoragent
  committed
  6ec06e9
- Keep where-alias Conj members on flatten; still drop record field_decl_block Conj.
  
  Brian Searls
  and
  cursoragent
  committed
  f6ea532
- Keep Named atom flatten members and drop flatten names already rehomed from decls.
  
  Brian Searls
  and
  cursoragent
  committed
  8a027b4
- Track the record graft skip as a failure-mode row; one production DFS for locate and find.
  
  Brian Searls
  and
  cursoragent
  committed
  75ff1be
- State the field_decl_block graft skip as the whole skipped population.
  
  Brian Searls
  and
  cursoragent
  committed
  2e1877f
- Drop unused Arrow import from namespace_graft.
  
  Brian Searls
  and
  cursoragent
  committed
  dd65f7d
- Partition graft units into Named edges and residuals in one fold.
  
  Brian Searls
  and
  cursoragent
  committed
  60d703b
- Assert XL-0 record names survive residual reconstruct, not tree drop.
  
  Brian Searls
  and
  cursoragent
  committed
  485e328
- Do not flatten through type/data units; treat residual field names as present.
  
  Brian Searls
  and
  cursoragent
  committed
  1990e13
- Account for every top-level unit in the graft partition; locate type RHS.
  
  Brian Searls
  and
  cursoragent
  committed
  4498593
Commits on Sep 19, 2026
- WIP: Declaration grafting: type/data decls never become Named edges — the emi
  Brian Searls
  committed
  0516a6d
- Merge origin/main into session/neat-moth-237 (import-list conflict in body_lowering_fold)
  
  Brian Searls
  and
  claude
  committed
  877a89e
- graft partition: import_decl units dissolve (Admission.imports owns them), never reconstruct as a body shell
  
  Brian Searls
  and
  claude
  committed
  204a801
- Graft body carries a producer marker: a one-member body is no longer readable as a spine segment
  
  Brian Searls
  and
  claude
  committed
  e2c262f
- Merge branch 'prereq-body-marker' into pr-11574-rebase
  Brian Searls
  committed
  1bca32b
- reference_site_collector: annotation at module-item grain (parse phase refused the in-body form)
  
  Brian Searls
  and
  claude
  committed
  37f5f89
- Merge branch 'prereq-body-marker' into pr-11574-rebase
  Brian Searls
  committed
  94a0142
- Rebase on the graft body marker; review 68228 deletions; record-beside-Named is a typed expecting-red
  
  Brian Searls
  and
  claude
  committed
  b17d706
- Merge origin/main (#11694 producer marker, #11619, #11700) into the #11574 rebase
  
  Brian Searls
  and
  claude
  committed
  528fdfd
- Ask the producer, not the shape: provenance flatten filter, conservation loss check, content-targeted data edges; re-route the dissolved walker's two callers; flip six greened probes
  
  Brian Searls
  and
  claude
  committed
  306ff98
- Review 68621: run the retired frontier row's other instruments; flip the two that greened, home the parse-tree twins; corpus sweep of renamed declarations
  
  Brian Searls
  and
  claude
  committed
  1797ea4
- cross_module_reference_resolution: the data-declaration row is found-then-identity-unrepresentable, not unbound (4b(4) re-statement at the measured rung)
  
  Brian Searls
  and
  claude
  committed
  4b8be99
Commits on Sep 20, 2026
- declaration_graft_assemble: one assemble per source, shared warm (ten producers), not one per claim
  
  Brian Searls
  and
  claude
  committed
  1176d0b
- Merge origin/main into the #11574 rebase (both sides added a share-roster ground note)
  
  Brian Searls
  and
  claude
  committed
  1c94e33
- Merge origin/main (39 commits incl. #11683 dotted-callee lowering, #11573 test marker channel) into the #11574 rebase; module_graph frontier rows composed by hand
  
  Brian Searls
  and
  claude
  committed
  2fdc7f7
- share roster note: the combined producer serves six claims (what the code wires), with the ledger reading recorded
  
  Brian Searls
  and
  claude
  committed
  d318f1b
- UnitDissolved carrier: name every consumer of the grafted body checked for an import read, and where each actually reads imports (review 69143, withdrawn)
  
  Brian Searls
  and
  claude
  committed
  cb927dd
- type_decl_field_decl_block row: the enrolled record rows establish a refusal WITH resolve_reason_unbound_symbol, not a bare non-acceptance (side-chat source hold)
  
  Brian Searls
  and
  claude
  committed
  e9b8859

Commits

Commits on Sep 18, 2026

Commits on Sep 19, 2026

Commits on Sep 20, 2026