[boingball]

Problem

A user reported this: after connecting once, they reopened MintPrint Settings with the printer switched off. The window seemed stuck in a loop looking for the printer, and they couldn't quit.

Cause:

• main() runs the startup printer-status refresh (perform_query_flow_allocated) before process_window_events() starts.
• Every network wait in the Query path (mp_connect_with_timeout, query_printer_attributes send/receive/parse) drained the window's IDCMP port with GT_GetIMsg/GT_ReplyIMsg and discarded each message. Close-gadget clicks were lost.
• A successful Query saves the endpoint as ip:631. On the next start, port_hint == 631 produced the port list {631, 631}: two attempts each at 8 s, so 4 × 8 s ≈ 32 s of "Trying …:631 (attempt n/3)" lines. On a stack where FIONBIO fails, the code falls back to a blocking connect(), which can take minutes.

Changes

• New mp_pump_gui_during_query() replaces the five drain loops. It still replies to every message, but records IDCMP_CLOSEWINDOW (cancel + quit) and Ctrl-C (cancel only).
  • The connect wait stops with EINTR on cancel. The receive loop stops and query_printer_attributes() returns the new -3. The perform_query_flow() retry loops stop too.
  • process_window_events() exits straight away if a close arrived during the startup refresh or during a Query started from a button.
• A saved port 631 is tried on its own (2 attempts, ≈16 s with the printer off) instead of {631, 631}. Explicit non-631 hints and the no-hint case are unchanged.
• A failed or cancelled Query now reloads the endpoint's capability cache. query_printer_attributes() empties the num_supported_* lists before connecting, so with the printer off they no longer matched the dropdowns still on screen. Ink levels aren't cached, so they stay cleared as before.
• Version bumped 1.3.3 → 1.3.4. That covers MINTPRINT_SETTINGS_VERSION, the $VER date (23.10.2026), a CHANGELOG entry, and the README / Aminet readme "What's new" sections.

Not changed

• The blocking-connect() fallback used when IoctlSocket(FIONBIO) fails still can't be interrupted.
• SSDP/mDNS discovery still discards window messages during its own short, bounded waits.

Testing

• make -k check locally: every host suite passes. test-postscript couldn't run here because Ghostscript isn't installed; CI installs it. These suites don't cover MintPrintSettings.c.
• Not compiled or run for m68k. Neither this environment nor CI has the m68k-amigaos toolchain. It needs a make gui build and a check on real hardware:
  1. Query a printer once so ip:631 is saved.
  2. Switch the printer off and reopen Settings.
  3. Click close during "Trying …". The window should close within about 250 ms.
  4. Reopen and wait it out. It should give up after two attempts with "Scan failed", and the dropdowns should keep their cached choices.

---

Generated by Claude Code