mirror of
https://github.com/mattermost/mattermost-mobile.git
synced 2026-11-09 12:40:12 +00:00
7f75e5c88a
Two independent fixes. app/client/rest/base.ts — cloud load balancers close idle keep-alive sockets (~40s), so the next request down a pooled connection dies with NSURLError -1005 "network connection was lost" before reaching the server (response_status=-1, 0 response bytes). Reproduced against the PR cloud servers: the custom-status PUT failed 2/2, and retrying once — which opens a fresh connection — made it pass 4/4. Deliberately scoped: - ONE extra attempt, not four. The native client already attaches a RuntimeRetrier that delegates to `request.request?.retryPolicy ?? session.retryPolicy`, and the session policy from NetworkManager (EXPONENTIAL_RETRY, retryLimit 3) inherits Alamofire's defaultRetryableURLErrorCodes, which includes .networkConnectionLost. Why that does not fire is unresolved, so this is a mitigation, not the root-cause fix — and it must not stack multiple JS attempts on top of a native retryLimit of 3. - Match the message text only, never the raw codes. "-1005" and "-1001" occur as substrings in our own hostnames (mobile-pr-10050…, mobile-pr-10010…), so a code match retried permanent 4xx/5xx errors. There is a regression test for exactly that. - Idempotent methods only; callers that must not retry already pass `noRetry`. detox recent_mentions MM-T4909_3 — the reply and delete steps hand-rolled a longPress on post_header.date_time with no scroll, no retry and no wait for the sheet, and post_options never opened, so replyPostOption was not there. Every other site in this file uses openPostOptionsFor, which scrolls the row in and longPressWithRetry's until post_options.screen appears — and it already opens options for this same post earlier in the same test, so it clears the @mention handler the date_time target was chosen to avoid. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| rest | ||
| websocket | ||