[dbezverkhnii]

A server can refuse a login with the RFC 5530 response code UNAVAILABLE — a temporary failure on its side, not the credentials. Yahoo answers NNN NO [UNAVAILABLE] AUTHENTICATE Server error - Please try again later about nine seconds after the command (30% of Yahoo mailboxes a day in production, 89% of the episodes clear on the next attempt), iCloud answers BAD [UNAVAILABLE] Service temporarily unavailable, Dovecot servers (iiNet, the report in COR-255) NO [UNAVAILABLE] Account is temporarily unavailable.. No login text matched, so mailcore reported ErrorAuthentication and Spark treated the refusal as a wrong password: auth-failed mark, password prompt, forced OAuth refresh, rebuild of every pool and a retry every ~9 s — against a server that, per production data, keeps refusing longer when a client keeps knocking.

• IMAPSession::login reads the RFC 5530 code of the reply libetpan parsed (imap_response_info->rsp_atom, rebuilt on every parsed reply) after MAILIMAP_ERROR_LOGIN or MAILIMAP_ERROR_PROTOCOL (the BAD form) and reports the new ErrorIMAPUnavailable, with mShouldDisconnect so that the next login starts on a fresh connection. The parsed field, not the text: libetpan strips the bracketed code out of imap_response, so a text match would never see it, and the [LIMIT] check reads the raw stream buffer only because libetpan stops at * BYE before parsing the tagged line.
• The check sits after the connection-limit texts: NO [UNAVAILABLE] Maximum number of connections … stays ErrorGmailTooManySimultaneousConnections, the more specific answer with the pool narrowing it calls for.
• The code is appended after ErrorIMAPTooManySimultaneousConnections (50) in the C++, C and Swift enums; like it, no ObjC constant and no localized description. iCloud's BAD [UNAVAILABLE] moves from ErrorConnection to the new code; Spark maps both to its connection error.

Spark's side (mapping, the pause before the next attempt, discovery, health counter) is the core PR for COR-255; it needs this change released and the pin bumped.

The C++ changed, so the Windows prebuilt needs a rebuild for this revision before a tag.

Tests: IMAPLoginTests against LeaseTestTCPEndpoint, each refusal for LOGIN, XOAUTH2, SASL PLAIN and SASL LOGIN. NO [UNAVAILABLE] gave 5 and BAD [UNAVAILABLE] gave 1 before the fix; [UNAVAILABLE] with the limit text, wrong credentials, Gmail's text and a bare BYE keep their codes. The wire shape matches the user logs attached to COR-255 and the Mac/iOS support logs: one tagged line, no * BYE, no continuation.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes IMAP login error classification and appends a new enum value; downstream clients must handle the new code, but scope is limited to login failure mapping.

Overview
Adds ErrorIMAPUnavailable across the C/C++/Swift error enums so IMAP login failures with RFC 5530 [UNAVAILABLE] are not reported as ErrorAuthentication.

IMAPSession::login now reads the parsed response code from libetpan (imap_response_info->rsp_atom) when login returns MAILIMAP_ERROR_LOGIN or MAILIMAP_ERROR_PROTOCOL (e.g. iCloud’s BAD [UNAVAILABLE]). It maps UNAVAILABLE to the new error and sets mShouldDisconnect so the next attempt uses a fresh connection. Connection-limit message checks still run first, so NO [UNAVAILABLE] with “Maximum number of connections” text stays ErrorGmailTooManySimultaneousConnections.

IMAPLoginTests cover LOGIN, XOAUTH2, SASL PLAIN, and SASL LOGIN, including NO / BAD [UNAVAILABLE], precedence over limit text, and unchanged behavior for wrong credentials and other refusals.

^{Reviewed by Cursor Bugbot for commit db12d3c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[dbezverkhnii]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit db12d3c. Configure here.}