- Restart bits: a group waits only when the edit touches it (next differs
from the live value) and the result differs from boot. Dynamic
resolution moves the live internal resolution off boot, so an in-game
texture filter change relaunched the game. test_menu_video_snapshot
covers both cases.
- Return to the frontend from a local game opens the launcher as at
boot; only a lobby match start (the runtime's pending-launch exit, or
the provider's lobby view asking once its match started) keeps the
launch and resumes the room.
- Default provider: with no title choice, the UI library's provider is
used when it ships one (recomp_ui_psx_menu.h), so recomp-ui stays the
default; PsxMenuHost.default_selection tells it to keep the title's
default presentation.
- Debug: menu_host reports provider, open, pause, capture and snapshot.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Register netplay_rb_verify_lost_post_test (#636). Each seat is built from
the production POST receive loop, Verify retransmit/wait/timeout block,
quorum publish and progress helpers cut from psx_netplay_rb.c, with
recomp-net's quorum, a controlled clock and a lossy/delaying POST
transport. Covers a Live seat past the target with its POST dropped,
genuine slow replay up to and at the target, first observation, multiple
owing seats, a disconnected or never-present seat, late and duplicate
POSTs, a later epoch and a rematch; each timeout must land within
RB_VERIFY_TIMEOUT_MS of the last in-episode progress, and the following
realignment episode must confirm one digest on every seat and advance the
confirmed frontier. The lost-POST case is also built against the pre-fix
wait (rb_owing_seats_progressed) and must hang there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review follow-up on #661 (#636):
- test_depth24_present_hold_vblank.py reads sources as UTF-8 (cp1252 on
native Windows Python raised UnicodeDecodeError).
- The present skip is now guest state: set at guest VBlank, read (not
consumed) by the present and packed into the GPU snap's hold word (bit 8;
older snaps derive it from the count). The executable test showed the
host-side latch made the present of a restored tick differ from live on
the last held VBlank (hold 0, skip 1), exposing stale trailing FMV
columns after a rollback load.
- Depth24 scanout staging and trailing-column blanking move from main.cpp
into gpu_depth24_stage_scanout (gpu.c), so the test renders through the
same code every CPU depth24 present runs.
- New depth24_present_hold_exec_test: drives real gpu.c through a 15-bit to
24-bit FMV, GP1(07h) retargets and mode transitions; compares hold, skip,
upload span and snapshot bytes across windowed/headless/deferred presents,
rollback replay from every tick and disk loads, and checks staged frames
never show stale movie pixels while the new movie stays visible.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
restore_edge_irq_replay_test links the real scheduler (traps.c) and IRQ
delivery (interrupts.c) against a deterministic harness guest. Each case
runs live, captures state at a block edge with the edge's resume PC, then
restores from inside guest execution through interrupts_resync_after_restore
+ psx_scheduler_resume_at and requires every delivery (cycle, EPC, CAUSE,
SR, I_STAT, interrupted GPRs, TCB save) and every later block to match live.
Covers compiled-game, BIOS ROM, CPS-continuation and interpreted resumes,
no pending IRQ, I_MASK / SR.IEc / SR.IM2 gating, same-edge dedup against the
game dispatch-entry check, and null-PC recovery (RESUME_CURRENT without
resume_at) not inheriting the one-shot check. On master the ROM,
continuation and interpreted resumes take the IRQ one block late.
The source-guard test now reads traps.c as UTF-8.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Verify retransmits POST and aborts the episode (realign) after
RB_VERIFY_TIMEOUT_MS, but re-arms that timer whenever an owing seat's
tip moves, on the theory that it is still replaying. A seat that already
finished the episode and runs Live past the target also moves its tip,
so a lost POST from it kept the others in Verify until the liveness
timeout ended the match (3 seats across a Mac and a PC: two seats waited
for "peer POST" at tick 12099 while the third was Live at 12114; #636).
Only progress up to the episode target now re-arms the wait; past it the
timeout is honoured and the episode realigns.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GP1(07h) arms s_d24_present_hold to skip a few Swaps after a depth24
display-band retarget. The hold is in the GPU snap, but it counted down
only at host present: a headless run (and any path that skips the present)
left it stuck, and a rollback snap load zeroed it. In R4's 4-seat rollback
self-test the replayed saves then differed from the live saves of the same
ticks in the GPU section (hold 3 live, 0 replay; #636).
The count now moves at guest VBlank (gpu_vblank_tick), which also latches
whether the following present skips its Swap; the present only consumes
that latch. A rollback snap load keeps the restored count. A disk
savestate load still clears it so the restored VRAM shows at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every state restore (savestate, rewind, selfcheck, rollback load) resumes
through psx_scheduler_resume_at at a block edge where a poll inside
psx_check_interrupts took the snapshot. The live run then evaluated IRQ
delivery at that same edge. The resumed run went straight into the block,
so an IRQ pending at the snapshot point was taken one block later, at a
different PC: a rollback replay delivered the tick-boundary VBlank 10-27
cycles after the live run did (live in BIOS ROM at 0xBFC07644, replay at
0xBFC0766C), and the replayed save differed from the live save of the same
tick in EPC and the TCB in RAM (#636).
resume_at now arms a one-shot check that the scheduler runs on the
RESUME_CURRENT path before psx_dispatch. It is the dispatch-entry check
that compiled game entries already run (deduped by cycle and PC), so a
compiled entry is not checked twice; BIOS ROM and continuation resumes get
the check they lacked. The recovery path that sets RESUME_CURRENT directly
does not arm it.
R4 4-seat local rollback self-test (PSX_RB_DET_SELFTEST=120): the boot
replay-vs-live EPC differences are gone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Matthew Stanley