[Bug] Session history remains loading after a local opening failure — validated patch

[shofel]

Reporting as Slava (shofel), a member of 8Hats. Public contact: si@8hats.ai.

I am sharing this here because the contribution guide currently directs bug reports to Discussions and says external pull requests are not accepted.

Bug

On upstream master d743267388641bc76f17c45ce8b4c231aed1d32c, a local exception while opening Session history can reject open() without publishing an error state. The UI can remain at Loading history… instead of offering an observable error/retry. This includes initial snapshot application faults and history-stream construction faults; marked Remote failures already have separate handling.

An intermittent Firefox report motivated the investigation. The error-state defect was independently reproduced with a deterministic test. I am not claiming that the original intermittent episode has been reconstructed or that this patch fixes every underlying history/serialization fault. No HAR, credentials, conversation contents or saved Session data are included.

Deterministic reproduction

In the existing Session Client test fixture:

1. Make session.eventSource.replace() throw a known local Error while applying the initial history snapshot.
2. Call session.open() and observe rejection with that original error.
3. On the unpatched implementation, the snapshot still reports openState: 'loading' and openError: null.

Expected: publish a retryable error state, retain the original exception/cause for diagnosis, and permit a subsequent open() to succeed after the local fault is removed.

Validated patch

• Signed commit 24511d66ee82477a85e2da9dfc2ffff7d5ff335a
• Current-master branch

The patch is one commit ahead and zero commits behind the checked upstream master. It:

• catches construction and snapshot-publication faults and publishes openState: 'error' with a gateway/internal Remote error;
• preserves the original local rejection and cause, existing Remote failure codes, and stale-generation suppression;
• adds history-opening diagnostics and actual local mux-cancellation diagnostics without request/frame payloads; consumed Host terminal frames remain silent;
• preserves upstream history paging options and uplink cleanup, and updates both package README languages and pairing records.

No persistence format, saved conversations, model configuration, dependency manifest or distribution-only change is included. Upstream already contains the local-engine native-constructor comparison that addresses the independent Firefox JSON-validator problem; that change is not part of this patch.

Checks on this upstream port

• 539 focused tests passed: Session Client 66, Gateway Client 132, reference ownership 17, Sessions service 45, Session transport 244, journal stream 35.
• pnpm run typecheck passed, including the Host build and generated Remote declarations.
• Targeted strict lint passed; git diff --check passed.
• All 43 doc-sync checks passed, none failed or skipped.
• Normal commit/push checks passed. The staged linter reports one unchanged unused-suppression warning in the Gateway implementation.
• Independent read-only review found no lifecycle, stale-generation, cleanup, diagnostic-payload or persistence blockers.

A locally deployed 8Hats backport opened/reloaded existing histories in Chromium and Firefox; the user reported no recurrence during an initial day of use. That is not a completed independently monitored multi-day trial, and those browser observations concern the deployed backport rather than a browser launch of this current-master checkout.

I would welcome maintainer feedback on the recovery behavior and diagnostics. The tested patch is available for review or inclusion through the project's preferred process, and I can submit a PR if external pull requests reopen.