[BUG] Release shell unit-test compilation exceeds the bounded silence deadline #9
Labels
No labels
alpha-1
bug
ci
docs
duplicate
enhancement
help wanted
invalid
performance
phase-6
question
refactor
security
wontfix
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
erix/exsh#9
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
The bounded release test-compilation regression is resolved in signed Exsh ef26d89c. Shared borrowed test callbacks remove repeated parser and fixture setup compilation while preserving production behavior and all test assertions. All ordinary release selections now pass their original limits. The separate native frame proof remains open in issue 4.
Reproduction
The retained source/dependency graph matches signed Exsh
dbc958bcdaa557a461e9308a31d23d3b8c189296except its later explanatory ROADMAP. Run the following distinct release configurations with the selected Rust 1.97.1 toolchain,CARGO_BUILD_JOBS=2,CARGO_INCREMENTAL=0, warning-denied compilation and the manifest's unchanged release profile:Each original command has a 600-second hard deadline, a 120-second output-inactivity deadline and a 64-MiB output ceiling under a process owner that terminates and reaps its children. The attempts use the original selected Git objects and local source mappings. They are distinct configurations, not reruns of an unchanged failed command. A separate clean-environment reproduction and failure frequency are not established.
Expected and actual behavior
Expected: both required release unit-test configurations compile and execute under the unchanged validation policy, with all test outcomes and compiler warnings visible.
Actual: the default all-targets attempt ends after 132.761407 seconds; the library-only attempt after 120.264245 seconds; the all-feature attempt after 120.047942 seconds. All three are stopped by the existing inactivity watchdog before unit tests execute. The owner records failure status 124, actual child status -15 and no cleanup failure. These are incomplete release tests, not passing test results.
Environment and identity
dbc958bcdaa557a461e9308a31d23d3b8c189296; its original full-hash Cargo dependency selections remain unchanged in this report.opt-level=2, LTO enabled and one codegen unit as declared in Cargo.toml.Evidence and prior checks
Measured host test compilation accepted — 25 September 2026: signed Exsh ef26d89c shares nine borrowed parser adapters across eighteen library test modules and four runtime fixture setup bodies. No allocation, global state or authority is added.
Matching library compiler diagnostics reduce Parser::run specializations from 166 to 2 and command duration from 250.624 to 75.908 seconds; both exact optimized library binaries pass 507 tests. Matching binary diagnostics reduce command duration from 135.869 to 107.280 seconds; both exact optimized binaries pass 480 tests. These are single instrumented host observations, with nested timings kept separate; no full-workspace or native speedup is claimed. Reproduction and interpretation records the unchanged compiler, inputs and diagnostic limits.
All fourteen ordinary development/release selections now compile and pass 988 tests each with the original two Cargo jobs, LTO, 600-second hard/120-second inactivity bounds and 64-MiB output bound. Formatting, fourteen strict host/native build, Clippy, doctest and private-rustdoc configurations and 355 frame-checker controls pass without warnings. Current original CI289, CI290 passes 988 unit tests and 355 checker controls per run; all four complete logs are reviewed with zero warnings. CI still fails the existing incomplete native frame proofs, tracked in issue 4. Its runtime/all observations retain the earlier 256/101 and 287/68 reached/unresolved counts. No unchanged CI run was restarted.
The scoped compiler regression in issue 9 is resolved. Earlier silence failures, the superseded partial candidates and their cleanup receipts remain retained. This test-only refactor adds no runtime feature needing a new VM scenario; the previously accepted native image remains bound to its exact source graph. Full CI closure, remaining authority/lifecycle work, external Rust/LLVM/toolchain rebuilding and both complete EriX builds inside EriX remain required. The phase checklist now has 418/878 completed sub-items, 17/460 accepted parents and 15.45% weighted progress.
The synchronous callback borrow adds no allocation, global state, lifetime extension or authority.
The matching instrumented library comparison reduces Parser::run specializations from 166 to 2 in function passes and 158 to 2 in machine optimization. LLVM function-pass time falls from 115.624 to 24.936 seconds, and fat LTO from 108.817 to 30.635 seconds. Command duration is 250.624 versus 75.908 seconds; both exact optimized binaries pass 507 tests without warnings. Nested timings are not additive. These single host observations do not claim native or full-workspace speedup.
The ordinary default release selection now passes all 988 tests in 120.345021 seconds with the original two Cargo jobs, LTO, 600-second hard bound, 120-second inactivity bound and 64-MiB output bound. It uses no diagnostic compiler switches; neither timeout nor cleanup failure occurs. The earlier seven-module candidate still failed the ordinary default release gate and was deliberately stopped through owned cleanup before the extended change. Its failure is retained. The complete revised feature/profile matrix and signed publication remain pending; issue 9, the separate frame proof issue and the full guest builds remain open. No checklist acceptance is added at this checkpoint.
Exact RESOURCE_EXHAUSTED replies must preserve the original Guarded mediator and Sealed peer, then allow normal configuration. Requests transfer no capabilities; every existing byte-I/O, actual exit, retirement and reuse gate remains required.
The preceding probe fails at call 22 in an independent 111-call transcript. Eight focused controls and seven development selections of 988 tests pass. Fourteen strict host/native build, Clippy and rustdoc configurations and 355 frame controls pass without warnings. Seven release compilations retain their output-silence failure. Four native frame reports are incomplete; two further runs exit 1 without reports near the checker's 60-second bound. Those observations overlapped the host matrix; the outer owner did not time out. Original Exsh CI287/288 passes 988 unit tests and 355 checker controls, then emits all four incomplete reports per run: runtime 256 reached/101 unresolved and all-feature 287/68. All four complete logs are reviewed with zero warnings. None accepts the compiler or frame gates tracked in Exsh issue 9 and issue 4.
A separate built-in timing diagnostic completes in 250.623833 seconds without warnings and its exact executable passes 507 optimized library tests. Rust reports 115.623711 seconds in LLVM optimization and 108.817139 seconds in fat LTO. The complete LLVM trace attributes 76.09% of inclusive function-pass time and 65.88% of machine-function optimization time to Parser specializations, with 166 run specializations before LTO and 158 in the machine-function trace. These classes are nested and not additive. Instrumented library-only timings identify a refactoring candidate; they do not establish an ordinary-build speedup or accept the release gate. The exact installed compiler uses explicitly selected diagnostic switches only; maintained configuration is unchanged.
All 187 Integration host commands, 76 original signatures and complete catalog closure pass without warnings. The full 2,657-page manual and all 45 document tests pass; all 491,619 word bounds and three viewed pages pass with no final warnings. Original Docs CI1095/1096 also passes; all four complete logs are reviewed through the final warning-free TeX pass. Fresh native execution with the original markers and 120/45-second limits remains required. This checkpoint adds no checklist credit. Supervisor loss, partial cleanup, NeverExecuted observation, complete CI review, external toolchain rebuilding and both full EriX builds inside EriX remain open in the phase tracker.
The accepted native realm-capacity, preparation and startup results retain their exact signed image identity; full CI and broader compiler/frame/cleanup obligations remain open.
Compiler diagnostic follow-up — 25 September 2026: a separate all-feature library release compile with inline remarks completes in 166.125412 seconds, and its exact executable passes 507 optimized tests. That first diagnostic emits a configuration warning because debug information is required for source locations. The corrected source-location capture reaches its 64 MiB log allowance; its complete failed receipt is preserved. A separately selected 256 MiB diagnostic allowance then retains 72,655,228 bytes and 140,029 remarks without warnings, completes in 180.030756 seconds, and produces another exact executable passing all 507 library tests.
These are diagnostic configurations with one Cargo job, a library-only no-run build, separate output storage and explicit compiler remarks/debug settings. Both retain the 600-second hard and 120-second output-silence limits. Ordinary validation bounds and the earlier failures are unchanged. The installed LLVM accepts the requested pass-timing option but emits no timing report, so inline counts establish compiler decisions, not dominant elapsed cost or a speedup. No normal all-target release gate is accepted from these observations, and no maintained code changes were made for them.
Realm-capacity witness checkpoint — 25 September 2026: signed Exsh bc5cc8ac requires a third BEGIN to return canonical RESOURCE_EXHAUSTED while both selected realm records are occupied. The reply must be Unconfirmed with zero identity and receiver fields, and both original generations must still read as sealed. Every original byte-I/O, native exit, independent retirement, stale-selector and record-reuse check remains required. The three additional requests use the existing Launchd route and transfer no capability. The independent 99-call transcript rejects the preceding probe at its missing request; seven focused controls cover exact refusal, malformed replies, transport loss and unwanted receipts.
Seven development selections pass 987 tests each. All fourteen host/native build, strict Clippy and private-rustdoc configurations and 355 frame-checker controls pass without warnings. Seven release unit compilations retain the original silence failure, and six current native frame observations return incomplete reports with exit 1. Compiler issue 9 and frame issue 4 stay open. One bounded read-only compiler stack sample finds active LLVM inlining; it does not establish time attribution or a speedup. Original red-selector and Markdown-wrapping failures remain preserved.
Signed Docs e2d4ffc1 records the contract and reconciles resolved grant regressions. All 45 document controls pass; the complete 2,655-page manual builds without final warnings, all 491,488 word bounds pass and three affected/adjacent pages are visually reviewed. Original CI1091/1092 passes, with four complete logs reviewed through the final warning-free TeX pass.
Signed Integration bdd04a62 selects both original sources. All 174 helper and six profiler commands, strict orchestration Clippy and both unit selections, formatting, dependencies and Markdown pass: 187 commands, zero warnings. Both catalog policies, all 76 original signatures and complete transitive closure pass. The fresh structural audit covers 3,194 code files below 1,000 lines and 98 crate roots enforcing missing_docs; semantic authority closure remains separate. Fresh native capacity execution is pending with the original 120/45-second limits. No completion credit is added before that evidence.
Earlier guarded-fixture CI1797/1798 each passes 372/492 scenarios, including admission, preparation, startup and ext4 quota. All 120 failures are classified: 119 obsolete policy refusals tracked in Integration issue 96, and the original counted grant refusal resolved in Loaderd issue 5. Six complete logs total 25,636,311 bytes with zero warning candidates. No unchanged job was rerun; the later fixes still need their own full CI. Remaining supervisor loss, exhaustion/cleanup, external toolchain rebuilding and both complete EriX builds inside EriX remain required by the phase tracker.
Configured-client head observation — 25 September 2026: original Exsh 29ada4eb passes 986 development tests in each of default, all, runtime, admission, preparation, startup and configured-client selections. Every corresponding release unit-test compilation reaches the original 120-second silence limit before execution (owned termination -15, deadline status 124, no cleanup failure). All fourteen host/native build, strict Clippy, doctest and private-rustdoc selections pass without warnings. The seven failed compilations remain original evidence; the timeout has not been extended and no release test execution is claimed. This report stays open.
The three original command receipts and both complete output streams remain retained. Both development configurations pass 976 Rust tests; four strict Clippy/private-rustdoc and freestanding build configurations pass without warnings. All 355 frame-checker tests pass. Four separate actual frame checks remain incomplete with 97/63 unresolved routes; they do not establish or explain this compiler failure.
Tracking
It requires two simultaneously ready realms, independent retirement and both stale-handle refusals.
Original exact-source CI runs 277 and 278 both fail the existing cleanup-frame gate. Their four complete logs contain zero warnings. Runtime profiles reach 256 functions with 101 unresolved observations; all-feature profiles reach 259 with 69 unresolved observations. Entry-local frames are 6,864 and 6,880 bytes against 8,192, fallback is complete at 48/33,792, and workspace mapping is complete; these partial measurements do not establish the full frame proof. See frame regression 4 and release compilation regression 9. Local frame attempts separately retain empty reports after successful disassembly. Guest execution, configuration/seal, mediated client I/O and complete guest rebuilds remain unproven.
Related: shell audit, WIP PR 3, frame proof, and Phase 6 completion.
Profile the actual compiler workload before attributing the delay or changing implementation. Preserve the original failed receipts, warning policy, release optimization and resource deadlines. A changed dependency graph may have its own independent validation; it cannot erase these failures. Acceptance requires executed release tests and separate complete frame/runtime proof. Do not increase deadlines, disable LTO, lower optimization or relabel an incomplete compilation as a passing check.
Original terminal-accounting graph validation — 21 September 2026:
Signed 0d5d5ad9aff082ca75ec79dbc9c29d73f799d8ed adopts the original shared terminal-observation, exact-acknowledgement and final-CPU dependencies. Both development configurations pass 976 Rust tests; all four host/native Clippy, warning-denied host/native build, applicable doctest and private-rustdoc selections pass. All 355 checker controls pass. The shell implementation and release optimization/LTO policy are unchanged.
The new graph independently retains two release unit-test compilation failures:
cargo test --locked --offline --release --all-targetsstops at 126.117857 seconds andcargo test --locked --offline --all-features --release --all-targetsat 120.024146 seconds. Both reach the existing 120-second silence bound within the 600-second hard bound; the child exits on signal 15, the owner reports status 124, and cleanup succeeds. No release unit test executes. These are new original-graph observations in compiler regression 9, with all prior failures retained. The cause needs an actual compiler profile; no optimization, LTO or deadline setting was relaxed.Actual runtime and all-feature PIC artifacts are checked in both deployment profiles using the maintained frame configuration. All four reports return incomplete proof (exit 1), with 97 unresolved routes for runtime and 63 for all features. Workspace mapping passes separately. The original ELF and packaged artifact digests, selected tool/helper identities and failed reports remain retained. Frame regression 4 stays open. Original CI 275/276 is being observed without restart.
The WIP PR awards no complete runtime or self-hosting acceptance. Actual service CPU/profiler VMs, complete authority/source/effect/frame proof, native external Rust/LLVM/runtime rebuilding and both full EriX guest build generations remain required in Phase 6 completion.
Original terminal-graph CI classification — 21 September 2026:
Signed
0d5d5ad9aff082ca75ec79dbc9c29d73f799d8edreaches actual analysis in both original CI 275/276. Each passes 976 unit tests and 355 checker controls before the required frame gate fails. Each run retains four incomplete reports: 101 unresolved runtime observations in both profiles and 68 all-feature observations in both profiles. All four workspace mappings pass separately. These CI artifact observations differ from local 97/63 results and are not interchangeable proof.All four terminal logs are complete and hash-verified (319,902 bytes), with zero warnings. The disassembler intake correction reaches the real proof boundary; tool-selection issue 5 is resolved. Complete frame issue 4 and local release compiler issue 9 remain open. No workflow restart, bound change or conversion of incomplete evidence to a pass occurred.
erikinkinen referenced this issue2026-09-21 20:28:25 +02:00
erikinkinen referenced this issue from erix/integration2026-09-21 20:54:16 +02:00
erikinkinen referenced this issue from erix/loaderd2026-09-25 11:27:01 +02:00
erikinkinen referenced this issue from erix/docs2026-09-25 11:27:08 +02:00