Skip to content

Observe precise startup failure boundaries in development - #7129

Draft
antuz123 wants to merge 3 commits into
codex/r5-review-base-4bef-20261022from
codex/r5-startup-failure-boundaries-20261022
Draft

antuz123 wants to merge 3 commits into
codex/r5-review-base-4bef-20261022from
codex/r5-startup-failure-boundaries-20261022

Conversation

@antuz123

@antuz123 antuz123 commented Oct 22, 2026 •

Copy link
Copy Markdown
Collaborator

The exact1bc iOS diagnostic run failed its first Create-a-passcode assertion and retained the fixed passcode-check error plus Retry in one hierarchy snapshot. That screen does not identify the failing operation: Splash catches both credential reading and navigation; hasPin covers Redux state, service selection, native Keychain read and JSON parsing. This separate package adds precise fixed development observations while preserving the existing failure and Retry behavior.

Head cd9d50a80bdd9035b532604bd10bcfeb21fe484e; tree 635c57ebb215387e583246bb2c6d9c12fcca1dcd; fixed base 4bef0b78581ccaf4acc89053d3c7d30b04fa1338. Preserve the fixed review base. Ancestors a806da91dcc3d18e8f3a1bb94a4bb2f246dd0107 and ed5749d47d8e8f79b596062c9c8760d8e4a1aa8e retain the initial observer and INFO-query changes. Eleven files, one diagnostic capability; changed code/test18843characters is roughly4.7k tokens. Patch-only comparison.

hasPin has an optional observer receiving only typed fixed stage/category facts. Native code and error-name accessors are guarded; arbitrary values stay UNKNOWN. The observer never receives the Error, credential, account/service identifier, app ID, message, stack or native response. Its failure cannot replace the original rejection or legacy false-on-error result. Splash distinguishes read and post-read navigation boundaries, emits one console.info line in development only for a current failed attempt, and preserves its modal, Retry, unmount and navigation behavior. Production supplies no observer and does not inspect errors for these diagnostics.

Pinned RN0.83.9 source forwards a single info string unchanged through nativeLoggingHook/RCTDefaultLogFunction to OS_LOG_TYPE_INFO. The existing own-process10-minute/25-second query now requests --info. It adds no query, SDK option, timeout, warning/error LogBox or native operation. The pure bounded parser accepts only a complete exact marker, optionally inside the exact [com.facebook.react.log:javascript] compact envelope. It regenerates at most32 fixed facts and discloses the cap; arbitrary prefixes/subsystems, suffixes, foreign processes and malformed records supply no facts. Actual INFO persistence, runtime routing and marker visibility remain UNVERIFIED until a later exact-candidate native run.

Validation on this exact frozen source: author3Jest suites/33PASS and60PythonPASS; independent review38Jest/60Python plus15 additional exact-envelope canaries PASS, zero skips. New helper strict TypeScript5.0.4 and ESLint PASS; changed TypeScript Prettier/whitespace PASS; all2407 tracked raw Git blobs/owner executable modes/regular paths and clean-source fences PASS. The temporary dependency link was removed by matching inode. Existing Splash/secure-store scoped lint remains FAIL15errors/7warnings: the a806 comparison to4bef has identical ordered rules/severity/source lines and messages after only the embedded no-shadow declaration line6→11 shift is normalized; those app lint inputs are unchanged in the later Python-only descendants. Initial missing literal dependency setup and two corrected new-helper style findings are retained as preparation failures.

This is an observation package, not a demonstrated startup cause or app correction. The immutable #7116/1bc run, #7123 fixed readout and #7124 original-byte transport retain their exact identities. iOS1bc build/install/pre-UI reported byte comparison passed; startup failed; persistence/workers/products were not run. Android is the preserved existing serialized continuation. No result transfers to this head or another SHA. The new helper source closure must be included in final combined genuine/native retarget checks. No dependency, credential option, provider protocol, release version or signing/distribution input changes.

Boundary evidence and limitations.

Codex local handoff

State: NEEDS_LOCAL
PR: #7129
Base: codex/r5-review-base-4bef-20261022
Head branch: codex/r5-startup-failure-boundaries-20261022
Head commit: cd9d50a
Goal and acceptance criteria: identify the precise failed startup boundary with fixed nonsecret facts before selecting any causal correction, then complete the combined R5 native matrix.
Changes made: optional fixed read observer, development-only current-attempt info marker, bounded existing iOS query/parser and adversarial controls.
Cloud checks: author33Jest/60Python PASS; independent38Jest/60Python/15 extra canaries PASS; new-helper strict types/lint, formatting, whitespace and raw2407-source fences PASS. Existing scoped lint FAIL15/7 with qualified baseline comparison. Native marker visibility NOT VERIFIED.
Required local checks: IOS, ANDROID, HARDWARE
Local checks requested: later exact combined candidate in hosted macOS/iOS and hosted Android after the current native window finishes naturally; record fixed boundary facts and complete feature checks. Actual physical phones/biometrics/camera/NFC/USB/signers remain separately NOT RUN. No personal Mac is required.
Risks / known gaps: no specific credential/navigation/entitlement failure cause or app fix is established. INFO persistence and exact runtime log routing remain unverified. Combined current-protocol and native acceptance need fresh final-source checks; 5.2 identity/completion/integration are not established by this package. No native opt-in, queue, rerun or cancellation is added. No merge, shared promotion, release signing/upload, distribution, deployment or publication authorization.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants