emulators.flycast.Flycast

Initialize the emulator with no ROM loaded yet.

Classify an archive member for the manifest.

Flycast keeps the savestate loose beside the VMU images, so the subtree-based default cannot split them; the .state suffix can.

Declare what Flycast takes: VMU and flash saves, raw VMU cards, and one resume state.

The state rides the archive and resumes through save.resume_slot, like the one the broker saves on exit. There is still no mid-session save or load, so supports_states stays False.

Place one declared member in the flat data dir.

Resolve a ROM path to a single playable disc image or file.

A pattern that cannot be walked (an unreadable subdirectory, a broken mount) costs only what that pattern would have contributed: the candidates already found still rank, since reporting a title unbootable over one bad directory is worse than booting the best of what could be read.

Run xdotool, returning its stdout, or None if it failed.

The window id belonging to this launch's own process, or None.

Matched by title then confirmed by pid rather than trusting the first title hit, since Alt+F4 acts on whatever id comes back and a stray match would close the wrong thing instead of just failing a hotkey.

Activate the emulator's own window and send it Alt+F4.

The keystroke goes through XTEST (no key --window) on purpose: Alt+F4 is a labwc keybind, and the compositor only acts on input that reaches the seat, so a synthetic event delivered straight to the client window would never fire its Close action and flycast would never see SDL_QUIT. XTEST lands on whatever holds focus, so focus is read back and the request is abandoned when a different window owns it, rather than closing something the player did not ask to close.

Stop any running instance and launch Flycast for the given ROM.

Request a graceful exit via Alt+F4, then escalate if it does not take.

Activates the emulator's own window and sends Alt+F4 to trigger Flycast's only clean-shutdown path (SDL_QUIT). Falls back to the base class's SIGTERM/SIGKILL escalation if no window can be found, activated or focused, or if the process has not exited within term_timeout.

Drop every session-owned file left in DATA_DIR before a restore.

Resume states, their owner markers, VMU images, arcade nvmem and eeprom all sit loose in one flat directory, none of them namespaced by title or by player, and this runs before the rom is even known (activate resolves the rom, clears the slot, then launches). A leftover would either outrank its incoming archive member by mtime and get skipped by extract_save_archive's newer-file guard, or, with no member to replace it at all, be handed to the next player and swept into their exit archive as their own progress. Anything still here belongs to a session that already exited, so dropping it all is the same trade-off DuckStation's clear_working_slot makes.

CONTAINER_FILES and subdirectories (flycast's controller mappings) are left: they are container setup, not one session's data. So are states set aside under UNTRUSTED_SUFFIX, which can be the only copy of that progress and which no resume can pick up anyway.

Rename a resume state that a force-killed exit may have torn, marker and all.

Size and mtime are all the broker has to judge a state by, and that is enough to refuse to resume from one but not enough to destroy the only copy of a player's progress, so the file is moved to a UNTRUSTED_SUFFIX sidecar instead of unlinked. A previous sidecar for the same rom is replaced, which bounds the leftovers at one per title. The owner marker travels with it: left behind it would claim the next state flycast writes under the same basename.

Stop the emulator and report whether this exit actually saved state.

A state the exit may have torn is not reported and not resumed from, but it is kept: _set_aside_untrusted_state renames it rather than deleting it. That quarantine does not depend on slot, since AutoSaveState is on for every launch and the caller dumps DATA_DIR by mtime whether or not a slot was asked for.