Skip to content

generated_sample_fields_5_0_6.md

Latest commit

 

History

History
96 lines (78 loc) · 5.44 KB

generated_sample_fields_5_0_6.md

File metadata and controls

96 lines (78 loc) · 5.44 KB
Raw

Generated sample scalar, temporal and reference defaults

Status: partial G12/P6 qualification. Runtime: fe82a18982415f1746df9a2cf7c42f5e9abdc6d3; producer: 45e1572c4956ca963410d8ea436ead3c8857087a.

Models and unchanged source

Three explicit models retain complete original output and provenance hashes:

  • sample_fields.xml includes a quoted JSON literal. The frozen generated crate cannot compile because sample_data.rs emits that literal without escaping its quotes. verify_sample_fields_source_failure.py --check preserves the actual compiler rejection; no generated source is edited to make it compile.
  • sample_fields_runtime.xml uses ordinary optional jsonMe(). It compiles, but the original sample setter receives Text("jsonMe() 1"), and typed checking fails. This runtime failure is compared directly, including the exact error and empty persisted rows.
  • sample_fields_temporal.xml retains the other fields and supplies successful scalar/temporal execution. This separate configuration is not evidence of a successful full JSON-literal original crate.

The successful model covers mandatory/optional date, timestamp and formatted clock text; true/false/optional booleans; Decimal, double and optional float; optional integer; ordinary text; numeric defaults zero, signed plus, negative parse failure and maximum u64 bound. A second model has two required references to generated records, exercising reference publication, fanout and per-row selection of unused IDs. Negative literal -9 is inferred as String; the distinct integer(sampleValue=-9) field tests numeric parse fallback.

Corrections to the generated adapters

The template's numeric property observes inherited isNumber metadata. The implementation workspace's old direct isNumber() method excludes float. C++ sample generation now reads the attribute as well, so optional float receives the same I64 source operand and typed value as the frozen template.

JSON sample input stays Text in the mutation ledger. The previous C++ recipe wrapped it in Json, silently making the source's rejected input acceptable. The corrected thin adapter leaves native JSON storage unchanged and marks the field loaded through the existing indexed state, matching the source setter's failed conversion/fallback. Public typed JSON update APIs remain available for explicit valid values. Model-aware Assist explains the sample failure, and the common guide explains first-error stop and retention of earlier commits.

Controlled clock and paired evidence

A test-only Linux preload library controls CLOCK_REALTIME in each probe process. Other clock IDs delegate to the system syscall. No TEAQL runtime target links this fixture, and production code continues to use the actual process wall clock. Every observation independently checks the observed wall time; comparing two uncontrolled runs or normalizing timestamps is not allowed.

Four clocks cover the 2025 year boundary, 2024 leap day, the 2000 boundary and 123 milliseconds after the Unix epoch. Seeded past dates at that last clock produce negative timestamps and pre-epoch calendar dates. A negative current system clock is outside this corpus: chrono::Utc::now rejects that process-clock stimulus, rather than supplying a comparable ordinary sample observation.

All 85 unchanged original/C++ observations match across 89 invocations and 1,980 newly persisted rows. Coverage includes seeds zero/seven, 1094/1095 modulo boundaries and maximum-u64-minus-19, Tiny/Small fanout, present/missing roots, repeated generation, full nullable/scalar storage and ordered dual-reference rows. Errors, integer/real kinds, decimal storage, calendar strings, milliseconds, IDs, dependency skips and reports are compared without normalization. Both probes use the same explicit BOOLEAN/NUMERIC/DATE schema declarations; this is not complete automatic bootstrap qualification.

The native module integration runs actual generated sample creation and typed readback with a null default PMR resource. It explicitly copies a generated model into Global, actually deletes UserContext, and verifies retained dates and nested optional clock-text allocation. No Context lifetime is extended implicitly.

Remaining qualification

Complete JSON-literal source generation remains rejected and explicitly recorded. Further constant/candidate/inherited/custom-type configurations, full source trace and sample progress log behavior, every Assist/extension responsibility and the P7 integration matrix remain open. These observations do not close G12 or P6.

Completed gates

  • Live original/C++ replay: 85/85 observations, 89 invocations, 1,980 persisted rows.
  • Native CTest: 2/2 (new field replay and existing generated sample execution).
  • ASan/UBSan: 1/1 field replay, with leak scanning disabled.
  • Module generated sample/readback/Global retention: 1/1.
  • Java CppGenerationTest: 27/27, including numeric attribute and JSON Text adapters.
  • C++ reproduction: 18 sets / 216 files; unchanged original reproduction: three complete new fixtures / 78 files.
  • Original JSON-literal compiler rejection reproduces with --locked and its recorded diagnostic. Pinned format/check covers 979 C++ files; source locks, generator ledger, module catalog/path and whitespace checks pass.

The process clock library is test-only and has no sanitizer runtime dependency; the actual C++ sample runtime/probe remains instrumented in the sanitize profile. MIPS/QEMU and physical-device execution were not run for this change.