5.4 KiB
Private syntax text producers — Windows evidence
Tested source: 97a8703adc75467a7f8e75f5c6cb00ca542b707c, local branch
codex/analysis-text-producers-956atf-20260925, based on published
8d52dab97c92cc31baae9f20d8c324b55818aa5d. Source and this packet are local-only
at this checkpoint. Gitea accepts reads/comments but rejected Git publication
with No space left on device; remote evidence remains at b2ef3f54.
This is a private prerequisite, not integrated application or release acceptance.
The contract reserves retained syntax strings and named decode/scanner/consumed-UUID temporaries before creation against actual carried storage. Raw JSON payload P is reserved before decode, then refunded to decoded D. This deliberately permits conservative escaped-input refusal only on the new fixture route. Successful syntax has no outstanding named temporary; resolution independently reconciles retained text and transfers carried storage once. Legacy and carried-phase behavior remain.
The archive is 606,451 bytes, 66 members, SHA256
252ea94e7c7f500a2afb3d0fa47c43a051f39d9c844328337289eea0c463becd.
manifest.json pins every raw member. It contains initial/final
receipts, logs, declared inputs, captured launchers, predecessor logs, source
patches, contract/design, compiler prologue observations and nine exact review
mirrors. The executable is identified by receipts but is not included. Historical
absolute builder paths identify the original environment, not peer dependencies.
Actual results and reproduction
Windows x86_64 gnullvm, Rust/Cargo 1.98.1, LLVM-MinGW 20260908; one Cargo job,
CARGO_INCREMENTAL=0. Linux/WSL builds and checks remain deferred. From the
identified source checkout with its locked dependencies and toolchain:
cargo test --locked --offline -p oa-core --lib text_producer -- --test-threads=1
cargo test --locked --offline -p oa-core --lib -- --test-threads=1
cargo test --locked --offline -p oa-core --lib -- --list
cargo clippy --locked --offline -p oa-core --all-targets -- -D warnings
cargo fmt --all -- --check
Exact executed argument arrays, environment and raw return codes are retained in
gates/*.result.json and their captured launcher inside the archive; those
records govern any differences from these readable commands.
| Gate | Observed result |
|---|---|
| Six new focused tests | Passed; 269 prior names filtered |
| Full core library | Failed: 259 passed / 16 failed, 275 unique names |
| Prior outcomes | All 269 unchanged; all ten existing stack controls pass |
| Strict core all-target Clippy | Passed |
| Workspace format | Failed, identical inherited eight blocks in five files |
All 556 declared inputs match Git blobs and raw disk across final gates. The
focused/full/list executable is 42,614,784 bytes, SHA256
41da1a4d788e2bd7392e69a724b7de259752d9f813ae6b42eadf580ec682a54a.
This is recorded-input/artifact concordance, not independent compiler provenance.
Fifteen inherited failure blocks match after numeric thread-PID normalization. One named block lacks exactly the old Rust backtrace-hint line; all its remaining bytes match. The checker note preserves the original strict comparison refusal and exact, narrowly pinned amendment. The cause involving a successful caught panic is an inference. Do not summarize this as 16 identical failure bodies or a passing full suite.
Initial 6b0ef99eca6f82a93e2cfc9eb16185c4f12403a7 already passed six focused
tests and completed the same full outcomes, but strict Clippy rejected
unnecessary_unwrap. Final source uses direct if let; tests and expectations
are unchanged. Initial receipts and patches remain archived. Its executable was
overwritten by the final build; the review does not claim current readback of it.
Independent checks and limits
- Source review, test review and finite literal oracle cover actual-owner caps, Unicode/escaped strings, scanner timing, all three consumed UUID paths, nested/projected copies, multi-file accumulation and once-only transfer. The original E offset error remains explicitly superseded: token34..72, ordinary diagnostic34..35; the test was already correct.
- Runtime review, readback JSON and reviewed checker bind raw logs, baseline archive, names, inputs and final artifact. The original checker preserves the stricter comparison that refused the single log delta.
- Stack observations record selected prologue increases totaling1,136 bytes relative to8d52dab. The repaired binding/prefix reservations remain2,520/11,720 bytes; these observations are not total headroom. New text cases use the ordinary harness stack, not a new128KiB guarantee. Older setup failures and predecessor repetitions keep their original scopes.
The named logical String inventory excludes decoder internals, numeric scratch under its existing policy, model/validator allocations and allocator overhead. Entry/metadata producer timing, remaining traversal/duplicate-name scratch, actual emitted-output verification, complete desired-Project equality and an owning accepted-result transition remain open. No public source8, save/history, GUI behavior or release change is established by this packet.