Repository navigation
Conversation
Bound simulator metadata startup to three confirmed-closed timeout attempts while preserving cancellation, device identity and command cleanup refusal. Retain the original paired trust failure without changing application policy or test outcomes.
Distinguish fixture readiness expiry from the intended Xcode exit, retain bounded helper cleanup and cover delayed and missing acknowledgements. Preserve first full-root failure and uncertainty about its original timing cause.
|
The first protected queue run for source The preserved Electron stderr records an uncaught |
PR #524’s first paired iPhone failure retained a request timeout while certificate-chain trust evaluation had no observed return. Its diagnostic observer exited on an initial five-second simulator metadata timeout before collecting stacks. This repair retries only confirmed-closed startup metadata timeouts, for at most three sequential reads with the same five-second/one-MiB bounds and existing overall observer deadline.
Cancellation, malformed metadata, wrong device, non-timeout errors and unknown child closure still stop startup. Fixed typed attempt receipts retain no raw output or private error strings. App/process/thread/activity/trustd identity guards and the common cleanup refusal remain intact.
The shell unit fixture now distinguishes an acknowledged Xcode failure from missing observer readiness within a shared monotonic seven-second budget; existing eight-second parent and ten-second test bounds and helper lifecycle assertions remain. Controlled delayed/missing acknowledgement cases failed before the fix. The original full-root failure’s exact timing cause remains unproven and its logs are retained.
Validation: 44 focused controls pass with zero failures/skips. Scoped independent review is clear after correcting a short negative-case allowance. Frozen build and typechecks pass, and full units report 2,607 passes with 36 skips and zero failures/cancellations at signed
7f8b05ece464e44a23b427ccbdb49f3a4435d182; the source manifest is unchanged. The first full-root build/typecheck passed but its unit fixture failed; that failure remains recorded separately.This is diagnostic recovery for #109/#281. Trust APIs, pins, request deadlines and pairing outcomes are unchanged. Fresh hosted collection and causal stacks remain required; a later successful pairing alone does not prove a trust-stall repair. Physical-phone, older-Companion, two-network, human, release and activation gates remain open. See
docs/reviews/2026-11-13-ios-trust-bootstrap.md.Hosted qualification now passes in source run
43559871989. Its actual previewe52e8ceda2e2c9f994dc7e69f2fbbb0e9dd69173has parents538d3573and7f8b05e; tree883e131b35dc2e2ae9aaf7f01fc924ec1afb604ematches an independent clean merge projection. Raw logs show 2,871 unit passes with 36 skips, 549 desktop E2E passes with 14 skips, eight site passes and four web passes, with zero failures/cancellations or observed retries. All four artifacts pass GitHub size/digest and ZIP integrity checks. This preview includes newer main changes; its tree is distinct from the bare source head.The independent iPhone artifact retains 65 native passes and 43 ordinary paired passes plus one expected-failure control, with zero actual failures/skips. The diagnostic manifest confirms the device on its first startup query, then stops with
COMMAND_CLEANUP_UNVERIFIEDand no samples. Thus hosted pairing passes and startup confirmation are demonstrated, while causal stack collection and the original trust-stall repair remain unqualified. Protected queue delivery and all human/device/release gates remain separate.