Repository navigation
Conversation
This comment has been minimized.
This comment has been minimized.
61a1d77 to
3c643b1
Compare
3c643b1 to
13bdc62
Compare
|
Codex renamed the internal helper to Validation on the final stack: 112 focused build-tools tests and typecheck pass. CLI-only serve-sim owner handoff remains unchanged. — Codex |
13bdc62 to
b72f76e
Compare
|
Codex made the prepared-session test suite explicitly emulate Darwin and restore the original platform afterward. Linux CI was correctly skipping the production Darwin-only stop path, so the suite's stop assertions were testing the wrong platform. All 17 tests now pass with a Linux bootstrap; runtime behavior is unchanged. — Codex |
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## gwdp/prepared-simulator-adoption-20261022 #4724 +/- ##
=============================================================================
+ Coverage 75.63% 75.70% +0.08%
=============================================================================
Files 1070 1072 +2
Lines 52967 53225 +258
Branches 11542 11613 +71
=============================================================================
+ Hits 40054 40290 +236
- Misses 11923 11944 +21
- Partials 990 991 +1 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
4660651 to
b951b79
Compare
b951b79 to
7239ed8
Compare
7239ed8 to
3318b8c
Compare
|
Codex contribution: build-tools now passes the raw selector to the pinned worker adoption CLI and validates the returned device before native/auth/version/artifact/event preflight. CLI adoption/replacement or invalid-result failure awaits positive terminal cleanup and then propagates; it cannot resolve the requested device through a separate caller cold-start path. Unavailable initial candidates and post-result native preflight misses retain safe fallback. Failed stop propagates. The existing preflight function block and private serve-sim CLI ownership handoff are unchanged. Local validation: 74 worker and 144 build-tools tests, both typechecks, bundle build, lint/format and isolated stubbed-native bundle smoke. This does not prove native replacement/cleanup or hosted behavior. The candidate is not all-service readiness; 420/425 seconds are provisional caller bounds, with active cleanup still potentially unbounded. — Codex |
3318b8c to
fc81018
Compare
fc81018 to
5b142c6
Compare
5b142c6 to
594641f
Compare
594641f to
b92f43b
Compare
b92f43b to
2f345b4
Compare
|
Share the host logging wrapper and launch environment. Remove the unused supported option, repeated Zod checks and duplicate adapter cleanup wrapper. Keep one memoized terminal stop; the function caller owns remaining startup failure cleanup, and screenshot finalization still waits for writer teardown. Validate endpoint ancestry against the current-boot controller PID and native start time; missing/old identity has no pathname fallback. Keep separate native/auth/version/device and artifact/event baseline checks. Reuse layer3's shared early-stop helper. 241 focused build-tools tests, typecheck and lint/format pass, including existing cold-session functions. No new native or hosted acceptance. — Codex |
2f345b4 to
613b96d
Compare
613b96d to
2116f99
Compare
|
Renamed endpoint validation, made environment fallback reasons specific, and separated serve-sim device-registration and URL warnings. Removed pre-adoption artifact IDs and event offsets under the fresh-VM-per-job assumption; native identity, version and authenticated preflight checks remain. Collection now uses the normal artifact/event path without a pre-adoption exclusion boundary. 193 focused tests, build-tools typecheck, lint and format pass. No native/hosted acceptance or performance run. Worker replacement and positive-stop-before-fallback behavior remain unchanged. — Codex |
2116f99 to
06d2e55
Compare
|
Integrated adoption-owned video recording with the agreed serve-sim CLI contract. When the job recording step is enabled, EAS supplies the output directory and recording ID and verifies the start acknowledgement before app actions. The poller reuses the job/device/server-token reservation instead of starting a second recorder. Shutdown finalizes that exact recording before worker cleanup; collection requires the owned finalization response and expected manifest. 197 focused tests, build-tools typecheck, lint and format pass. Checks use isolated filesystem and stubbed CLI/HTTP/native operations; no EAS native or hosted acceptance run. Existing v1 record-video process handling and fallback policy remain unchanged. Lower three heads and layer 5 logical diff preserved. — Codex |
06d2e55 to
ce28595
Compare
|
Renamed the worker-facing helper to 91 worker and 197 build-tools tests, types, build, lint, format and isolated offline runner smoke pass. Native/hosted acceptance was not rerun for this rename. — Codex |
ce28595 to
547d529
Compare
|
Renamed the worker-facing helper to 245 focused tests, both typechecks, worker build, lint, format and isolated offline bundle smoke pass. No native/hosted acceptance rerun for this rename. — Codex |
547d529 to
06099f0
Compare
5d7bf48 to
fc79bcb
Compare
fc79bcb to
fffa6d9
Compare
|
Capture stdout explicitly for the four native boot/process-identity preflight calls. turtle-spawn inherits stdout when options are omitted, which made a valid warm session fall back before claim. Global spawn defaults, identity checks and cleanup behavior are unchanged. Validation: 106 focused tests with coverage, typecheck, lint and format passed. The regression uses the real turtle-spawn wrapper with disposable Node children and unchanged call options; it does not fabricate captured stdout or execute native Simulator commands. Fresh joined-chain native retest is pending. — Codex |
Edited-by: Codex
fffa6d9 to
a972d92
Compare
|
✅ Thank you for adding a changeset! |
No description provided.