[atomchild411]

Fixes #307.

What was wrong

• Save state restarted the machine. After saving, iris-gui resumed by calling start, which starts the whole machine again: a second monitor server on a port that is already in use (the warning in the issue), and every peripheral started on top of itself.
• The name boxes closed the menu. The Machine menu used egui's default close behaviour, which closes a menu on any click, including the click that focuses the Save or Restore state name box. So the name could not be typed and the snapshot got the default name.
• IMPACT machines lost their display engines on every restart (found on the way). Machine::stop stops the IMPACT display engines and the Ultra64 board, but restart_peripherals never started them again. Everything that resumes through it (machine-start, reset, snapshot loads and rollbacks, and now iris-gui's Save state) left an IMPACT machine without its HQ3, RSS and refresh threads.

The fix

Two commits:

1. machine: restart the IMPACT display and the Ultra64 board with the rest: restart_peripherals now restarts everything stop stopped, except the CPU. Both starts are idempotent.
2. iris-gui: Save state keeps the machine running: the GUI resumes through start_guest, the path the monitor's machine-start uses, and only if the machine was running when the save began. The Machine menu closes only on clicks outside it; its Save, Restore and CPU choices close it themselves.

HAL2 still reopens its audio output after a save, as it does after save + machine-start from the monitor: a save stops the peripherals so their state is consistent.

Testing

On this branch's commits (01eb442, rebased onto 64c7fbf):

• cargo check --workspace --all-targets with default features, hostgl,jitv2,tcache,lightning,rex-jit,gr4-jit and jitv2,developer: no new warnings (the two it prints, in src/net/testnet.rs and the unused crossbeam-queue dependency, are on main already).
• Lib tests: 1254 pass with default features, 1798 with the full set, none fail.
• iris-gui (built with iris/hostgl,iris/jitv2,iris/tcache), an Indy: clicked into the Save state box and typed a name with the Machine menu staying open, saved (Snapshot saved to saves/saved_indy, no monitor-port warning, the machine kept running), then restored it. The side panel's Save state and Restore state buttons send the same commands.

On the same two commits before the rebase onto 64c7fbf (only the base below them changed):

• IP28 Solid IMPACT at the PROM, machine-stop then machine-start, threads listed with sample: before the fix, MC-DMA, both WD33C93A and the SEEQ threads came back but MGRAS-HQ3, MGRAS-RSS and MGRAS-Refresh did not; with it, all of them come back.
• Boot matrix (Indy Newport, Indigo2 IP22 XZ, IP28 Solid IMPACT): all pass.

🤖 Generated with Claude Code