Files
ARCADIA-Studio/harness/experiments/analysis-source-s2b-windows-20260913/archive/evidence/.local/analysis-source8-s2b-full-failure-review-20260913.md
T

6.1 KiB
Raw Blame History

S2b corrected full-suite failure causes

Independent read-only review, 2026-09-13, /root/source8_s2a_review. Corrected source a2a28f8d23a46ba1696b6a8410d4d3b5c1808f49; comparison source b2c970b4673e062e47eadd12906d2316c4b1a7b3. No compiler, test rerun, source change, archive execution or Git write occurred in this review.

Actual a2 full closure: 126 passed / 16 failed / 0 ignored, exit101, ended 2026-09-13T07:03:51.186178+00:00. All142 unique outcomes exactly match the corrected listing; all13 fixed new lower tests pass. The corrected typed selection independently contains34 passes and no failure. List, typed and full record the same test executable identity,36,685,824 bytes, SHA ce9d77ed1e68a21e06836264bf50ba9dddbfa69507e1a0bc2c05a7b55db325a9. Their531-entry manifests are identical and every entry/membership was independently compared with the exact a2 Git input scope and raw blobs. No current mutable executable was reread.

The a2 failed-name set exactly matches the actual b2 full 113/16 set. This review also compared each separate panic block: all16 have the same panic source location and the same complete diagnostic line as the corresponding b2 failure, apart from surrounding thread IDs and run presentation. Every diagnostic rejects the header at line1, column9, byte span8..9 with Unsupported schema version: expected arcadia 3, 4, 5, 6 or 7. There is no new AsKind failure in the corrected full run.

Exact raw bindings

All files are original .local/analysis-source-gates-20260913/ artifacts:

File SHA256
full-a2a28f8.log 9ae08b9a0157ebef587d36b40e5f7da8dd61b13bd5594bbae659f06c75e59044
full-b2c970b.log 0e0fd205258cd7f1b6a898486c35514a359eb02e3b17382b0f2cffb4d443cfb3
full-a2a28f8.result.json 81e171692b5c782bd234888e94468853b47a0e54375106e4cfce831e7b58e6c1
full-a2a28f8.inputs.json 08b6dabbffbd06f7abea717f9aa34cd910c41c4a8c650247006a0f6858561c84

Source-level cause, separately inspected

These are retained public-source writer/migration failures. They must not be described as all beginning with source8 fixtures: several begin with valid source5, and the checked-in source project begins with source3.

model.rs:9 defines SCHEMA_VERSION = 8. The public canonical writer emits that value at language.rs:506, while the public parser's explicit accepted header set remains3–7 at language.rs:1335. The two direct round-trip failures at language_nesting_tests.rs:63 and124 occur at parse(format(project)) and parse(format(demo)): initial parsing/earlier setup has progressed, but the formatted source8 header is rejected.

The remaining four nesting failures and all ten transaction failures reach compile_bundle. That function invokes source_migration::upgrade when migration metadata exists (source_project.rs:298). The upgrade writes the current schema value into the source version token (source_project/source_migration.rs:83) and reparses the edited bundle using compile_once at115. That source8 header is rejected by the same unchanged public parser. The source5 transaction fixture fails during its shared compile_bundle(...).unwrap() at source_files.rs:1560, before the named save/recovery operation starts. The checked-in source3 example is likewise rejected during migration/reparse; this does not establish successful source3 whole-project migration in these tests.

The implicated writer, parser, migration, model, nesting tests, transaction fixture and checked-in entry bytes were individually compared across b2→a2 and are unchanged: language.rs, language_nesting_tests.rs, source_files.rs, source_project.rs, source_project/source_migration.rs, model.rs, and examples/source-project/project.arcadia.

Exact failing names, grouped only for readability:

  • language::nesting_tests::checked_in_example_supports_cross_file_reparent_without_changing_allocations
  • language::nesting_tests::checked_in_nested_example_has_stable_ids_typed_exchanges_and_three_clear_views
  • language::nesting_tests::nested_elements_infer_only_parent_and_keep_typed_fields_and_allocations_explicit
  • language::nesting_tests::nested_fragment_id_stabilization_preserves_comments_and_indexes_real_bytes
  • language::nesting_tests::non_elements_includes_and_declarations_inside_typed_values_are_rejected
  • language::nesting_tests::normative_nested_ownership_example_compiles_as_a_source_fragment
  • source_files::transaction_tests::interrupted_publication_without_write_marker_never_rolls_back_external_matching_bytes
  • source_files::transaction_tests::interrupted_save_and_interrupted_recovery_can_both_be_resumed
  • source_files::transaction_tests::malformed_v2_write_marker_refuses_recovery_without_changing_source_bytes
  • source_files::transaction_tests::partial_save_failures_restore_every_original_and_write_entry_last
  • source_files::transaction_tests::recovery_before_any_write_accepts_v2_marker_and_keeps_v1_rollback_compatibility
  • source_files::transaction_tests::recovery_rechecks_external_changes_between_restorations
  • source_files::transaction_tests::recovery_rejects_untrusted_escape_paths_before_touching_files
  • source_files::transaction_tests::rollback_refuses_external_edits_even_in_an_unchanged_dependency
  • source_files::transaction_tests::rollback_removes_only_our_new_file_and_retains_detached_files
  • source_files::transaction_tests::unwritten_journal_never_rolls_back_a_completed_writer_with_identical_after_bytes

Disposition: the corrected private lowerer tests pass, while the full library remains failed for the same separately verified public8 writer/migration boundary. The sixteen tests do not validate their later nesting/save/recovery assertions. Do not waive them as full-suite passes or change public dispatch during this bounded correction. The original472 typed33/1 failure and its source-oracle correction review remain separate historical evidence.

This note does not claim token/strict completion; those gates were still pending when the review began and were not inspected here. The historical b2 production-language result remains 7 passed / 9 failed at b2, as preserved in the S2a checkpoint; no a2 legacy integration rerun or relabeling is claimed.