forked from Ports/goldeneye-pc-port
SDL_CONTROLLERDEVICEADDED / REMOVED are dequeued by whichever of the two event pumps polls first; fast3d's gfx_sdl_handle_events had no case for them and dropped them, so inputRescanPads never ran after a replug (the stale handle later aborted in malloc). Both pumps now post a request and inputUpdate() consumes it on the pad-reading thread. Findings D435. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>