Codex
e2ecfea69b
Preserve current core, source, REST, desktop, recovery, documentation and validation work, including AGENTS.md and the development summary. This is a work-in-progress checkpoint: the new compact palette has outstanding compilation errors and native acceptance; earlier passing gates retain their documented scope.
7.1 KiB
Source7 language, API and integration checks
Current integration on codex/release1-continuation after checkpoint4b6b5fc. Fresh source7 Windows core/server/desktop and strict checks pass as recorded below; native7 and current Linux acceptance remain separately scoped and incomplete.
The normative reST adds caller-owned ContextualLink endpoints and three distinct purposes, instance-specific analysis, typed dependency cleanup, explicit operation allocation selection and Call-derived Reply behavior, omission rationale on SA→OA, and frozen old5/6 recovery requirements. OpenAPI adds closed records and five typed semantic leaves, with the same200omission capacity on ordinary and semantic preview/Apply.
crates/oa-server/tests/semantic_schema.py --check --validate passes 72structural fixtures,204component schemas,14families and90scalar leaves. It uses the task-local .local/source6-schema-tools Python dependencies. Missing jsonschema in the bundled runtime was resolved by using that existing dependency directory; no fixture was skipped. This is structural-schema evidence, not runtime proof of path or port semantics.
Both docs/language.rst and docs/api.rst rendered with Docutils0.22.2 and halt_level=2, report_level=2, using the html5 writer. Initial short heading underlines were corrected; both strict renders pass and checked-in HTML was refreshed. The new contextual source example is a declaration fragment requiring its enclosing fixture, not a standalone compilable project. Complete normative examples remain subject to the core language tests.
Independent state-contract review found and root addressed:
- Category/view fields are rejected by versions1–5; version6 accepts them. The API wording now names that boundary explicitly.
- New authored path entries consume the existing caller's ordered-reference budget: REST50000, desktop500000, core explicit ParseLimits. Cache state never alters counting. Path depth64 and analysis50000nodes/100000edges are separate limits. No desktop capacity reduction was introduced.
- A selected operation without actual channel/lifeline/direction proof must be refused; core was corrected rather than weakening the ratified proof requirement. Explicit allocationNone remains editable.
- Sole allocation deletion remains subject to independent carried-item contracts, including a separate Reply channel. Surviving same-item allocations may admit a channel but never replace the deleted selected identity automatically. This clarification is ratified in the source7 decision.
- HTTP old-version rejection must isolate the new fields so one rejection cannot mask another. The new transport test sends ContextualLink-only and operation-null-only schema6 candidates in addition to the combined case.
Actual HTTP tests cover a modular source import, changing only a caller reference path and a parameterless operation allocation in one transaction, querying ContextualLink, exact full-model comparison, unchanged authored CRLF/Unicode note, source recompilation parity, invalid definition-as-path/item-as-allocation rejection, old-version rejection and exact model/source/revision/ETag/disk preservation after restart. Separate ordinary/semantic tests enforce200/201omission limits on Preview and modified prepared Apply envelopes.
cargo test --locked -p oa-server --test http source7_ -- --test-threads=1 compiled in2m56s and the new source7 test passed in0.35s (.local/source7-http-contract-tests1.log). The complete HTTP target then passed 51/51 in17.15s (.local/source7-http-full-tests1.log), including both omission-limit tests and all prior transport cases. One-job Windows GNU/LLVM environment was used. This accepts that earlier compiled core/server checkpoint. The fresh complete core/server runs below supersede it and include the later bounded contextual-work correction; the historical log remains unchanged.
The current teaching entry, model JSON and transition-before JSON have only their version tokens changed6→7, plus constructor operation:None defaults and README labeling. All four teaching tests, including exact generator equality, passed in the core checkpoint. All pinned source6 native executables, copied source baselines and historical fixtures are preserved. Recovery fixture JSON receives scoped -text attributes so pinned SHA/FNV bytes are not changed by Git line-ending conversion.
The offline guide now explains the implemented contextual endpoint tree, instance navigation, Focus canvas, operation selection/derived Reply and omission rationale controls. That text is included in the fresh source7 desktop build; remaining native walkthroughs must verify the actual compiled UI.
Fresh Windows integration gates — 2026-09-07
The current one-job GNU/LLVM runs supersede the earlier scoped compile/test checkpoints. Counts below are independently read from each log's test result: summaries; affected reruns are not added to full-suite totals.
| Gate | Observed result | Evidence |
|---|---|---|
| Core, all targets | 433 passed, 1 ignored, zero failed across 44 summaries; exit0,138.72s | .local/source7-core-all-targets.log; .local/source7-current-gates-status.json |
| Server, all targets | 55 passed, zero failed/ignored across 4 summaries; exit0,64.74s | .local/source7-server-all-targets.log; current-gates status |
| Full desktop | 561 passed, zero failed/ignored; tests263.39s | .local/source7-desktop-full-tests2.log |
| Strict workspace/all-target Clippy | Passed; finished in2m50s | .local/source7-workspace-clippy2.log |
| Final format check | Root observed cargo fmt --all -- --check exit0; empty log is expected |
.local/source7-format-final.log; root's execution observation |
| Affected semantic rerun after equivalent lint edits | 3 passed, zero failed/ignored; tests3.46s; whole invocation exit0 in57.12s | .local/source7-semantic-after-clippy.log; current-gates status |
| Native binary build | Exit0,124.47s | Current-gates status entry native-build |
The first full desktop run exposed a real scroll-state regression: a conditional navigation control changed the widget structure above the canvas. A stable navigation Column fixed it; the unchanged regression passed in the complete561-test rerun. The later three-test semantic rerun covers contextual Incident composition, reverse operation omission after recovery, and the operation virtual-None Candidate. It does not claim a second complete desktop run. The ignored core test remains ignored in this run and is not counted as executed evidence.
Root reports native7 PID11060 passed the Reset case's complete model/source Save, Undo and Redo oracles. Reopening explicitly displayed allocation30 on the derived Reply. This accepts that bounded case only; other native7 authoring, recovery and engineering walkthroughs remain open.
Root launched a separate bounded90minute Linux gate with launcher5000 and a455-file source snapshot created2026-09-07T07:02:43Z, SHA256 7104946c89b2be7a1a030fdff47cb4405bacf05357d68cf35d181e860f55c279. Its build/test gates have not started at this documentation checkpoint. Earlier Linux core/MSRV and server/HTTP results retain their older snapshot scope; no current Linux application or release-package acceptance is claimed.