- run: a crash, hang-up or socket timeout on the final frame_fingerprint /
dispatch_stats queries is a RunError (exit 3, INCOMPLETE), not a
JSONDecodeError traceback with exit 1 (MISMATCH). Seeding, log and dump
I/O errors and a runtime that survives SIGKILL also exit 3. A missing
output directory is a usage error. The dump is written to OUT.tmp and
renamed.
- compare: with different FMV-quiet settings, once a quiet frame occurred
(qc > 0, a different frame set, or no qc column) the runs are not
comparable. Report INCOMPLETE and skip the DIVERGE, missing-frame and
locator lines, which were artefacts of the setting. Mixed settings with no
quiet frame are judged as usual, with a note.
- --launch on Windows keeps backslashes (shlex with no escape character),
so C:\psx\build\game-runtime.exe is not mangled into C:psxbuild...
Self-test: 59 cases (11 new), ctest fp_identity passes. R4 dumps re-judged
and fresh warm/cold runs repeat the IDENTICAL results with 497 tolerated
straddles.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
FAITHFUL_TIMING_PLAN gets the dated session entry and ACCURACY_BURNDOWN's
backend-equivalence item names the tool and the R4 warm/cold result.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Moves the warm/cold A/B tool out of R4 so every title can use it. `run`
launches the runtime from a command template ({build} {port} {headless}
{state_dir}) or from --runtime/--game/--disc, optionally installs a saved
overlay state (--seed), turbos to frame N and saves the frame_fingerprint
ring plus the dispatch-miss count. Runs default to PSX_DEBUG_FMV_QUIET=0 and
PSX_OVERLAY_AUTOCOMPILE_OFF=1, and the dump records what they used.
`compare` judges on the guest-fact columns (cyc mmio mc sp sc wc ws qc) and
reports wr/pc only as the frame where the runs part. It tolerates the
one-write VBlank straddle from docs/TCP_COMMANDS.md (ws/wc off for one frame,
equal again at the next) and, when FMV-quiet was on, the wc+qc shift with ws
judged against the carried offset. Every tolerated frame is counted and
listed; --strict turns them off. A straddle on the last frame, a judge column
an older runtime does not report, or different FMV-quiet settings make the
result INCOMPLETE. Exit codes are unchanged from the R4 tool: IDENTICAL 0,
MISMATCH 1, usage 2, INCOMPLETE 3, MISSES 4.
The self-test (ctest fp_identity) covers column judging, both tolerances,
truncated and mismatched frame sets, dumps without ws/qc, and drives `run`
against a fake debug server.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
frame_fingerprint's wr column is an ordered hash over main-RAM writes.
DMA and device writes (MDEC-out, CD, SPU, GPU->RAM) are recorded in host
device-service order. Native shards flush cycles at every store barrier,
while batched interpreted and static code services devices after the
block. So identical guest writes gave a different wr forever after.
- ws: an order-independent sum of a splitmix64 term per (addr, val) over
main RAM. Together with wc it is the main-RAM judge column. wr and pc
stay as ordered locators.
- qc: fingerprint-eligible writes that FMV-quiet kept out. If a straddled
write falls into a quiet frame in one run only, the cumulative columns
never agree again, and qc is where that shows.
- The accumulator moves to runtime/include/frame_fingerprint.h so the
semantics are unit-tested (frame_fingerprint_test).
- The JSON buffer is sized from the real maximum entry length (284 B),
which the old 224 B per entry could overflow.
docs/TCP_COMMANDS.md documents the column roles, the one-frame snapshot
straddle and the FMV-quiet caveat (set PSX_DEBUG_FMV_QUIET=0 for A/B
identity runs). Existing columns keep their meaning. Nothing outside
debug-tools builds changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The emitter writes g_debug_last_store_pc before every guest store. The
overlay preamble gave each DLL a private copy on the theory that only debug
tooling reads it. The runtime actually reads it in every build:
- memory.c psx_write_word_raw keys its RAM 0x0-0xF store filters on it;
- the fingerprint pc/mmio/sp columns use it, and so do wtrace,
record_frame, the MMIO/GP1 rings, parity provenance and the freeze
heartbeat.
So after a native overlay store the host still named an older store, and
those consumers saw an instruction the interpreter would not have.
OverlayCallbacks gains last_store_pc, which the loader always points at
the runtime's copy. In PSX_OVERLAY_DLL_BUILD, cpu_state.h maps the
emitted name onto that pointer, the same pattern as the bail-state
pointers. Generated code is unchanged. A NULL host pointer keeps the old
private copy.
The ABI goes to 24, so the loader rejects and regenerates older shards.
The preamble change already moves the codegen-hash cache path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two failures hit while migrating a title on Windows 11:
- `audit` crashed with UnicodeEncodeError on the "->" mark (U+2192) when the
console code page can't encode it (cp1250 here). main() now reconfigures
stdout/stderr with errors="replace", so reports degrade those characters
instead of aborting.
- record_framework_pins failed. subprocess.run(["bash", ...]) lets
CreateProcess search System32 first, which starts WSL's bash; it can't see
D:/ paths, and a CRLF checkout of record_pins.sh breaks any bash anyway.
It now resolves bash via shutil.which, passes POSIX-style paths, and falls
back to the existing git rev-parse snapshot when the script can't run.
Most of this was done by Claude Opus 5.5; I reviewed and tested it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The plain present path copied the VRAM image into the swapchain with
vkCmdBlitImage, alpha included. VRAM alpha is the PSX mask/STP bit, 0 on
most pixels, and AMD's compositor honours it even with an OPAQUE
compositeAlpha: parts of the window went see-through (8x8 black/white
blocks of whatever sat behind it). Reported in #411.
vk_renderer_present_vram/_wide now draw the displayed rect with a
fullscreen triangle whose fragment shader writes alpha = 1 (new
present.vert/.frag). The pass is self-contained and created lazily; if
any of it fails to build, presents fall back to the old blit. Its render
pass ends in TRANSFER_DST_OPTIMAL so finish_present (OSD + present
barrier) is unchanged. Wide surfaces gain SAMPLED usage for it.
Checked on an RX 7900 XT: intro/logo screens render correctly, no new
validation-layer messages.
Most of this was written by Claude Opus 5.5; I reviewed and tested it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Winsock sockets are inheritable by default. Any child process the runtime
spawns (e.g. the overlay autocompile's `cmd.exe /C python compile_overlays.py`)
inherits the debug server's listening socket; if the game exits while the
child is still running, the child keeps 127.0.0.1:<debug_port> bound and the
next launch cannot bind it (clients get connection refused / talk to a dead
listener).
Observed: after closing the game mid-autocompile, `netstat -ano` showed :4370
LISTENING under the exited game's PID, held by the surviving python child;
killing the compile freed it. Fix: SetHandleInformation(HANDLE_FLAG_INHERIT, 0)
right after socket(). No behavior change on non-Windows.
This change was made with AI assistance (Claude Opus 5.5, Anthropic) and
reviewed/tested by the submitter.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Found with the Khronos validation layer on AMD RX 7900 XT (Windows, mingw),
where D24S8 is unavailable and choose_ds_format() picks the combined
D32_SFLOAT_S8_UINT:
- VUID-VkImageMemoryBarrier-image-03320: every view/barrier/clear on the
canonical and wide depth-stencil images named only the STENCIL aspect.
Without separateDepthStencilLayouts, a combined depth/stencil format must
be transitioned with both aspects. Added DS_ASPECT and used it for the
image views, barriers and vkCmdClearDepthStencilImage ranges.
(vkCmdClearAttachments keeps the stencil-only mask - that is legal.)
- VUID-vkCmdClearDepthStencilImage-pRanges-02659: the canonical DS image is
cleared with vkCmdClearDepthStencilImage but lacked
VK_IMAGE_USAGE_TRANSFER_DST_BIT (the wide DS images already had it).
- VUID-vkCmdCopyImageToBuffer-dstBuffer-00191: staging buffers are used for
readbacks (vkCmdCopyImageToBuffer) but were created with TRANSFER_SRC only.
Validation messages over the same 20 s boot (SmackDown 2, Vulkan):
53 -> 3 (the remaining VUID-vkQueueSubmit-pSignalSemaphores-00067 swapchain
semaphore reuse is untouched). The fix also cleared the downstream
VkImageMemoryBarrier-oldLayout-01213 and vkCmdDraw-None-09600 reports.
This change was made with AI assistance (Claude Opus 5.5, Anthropic) and
reviewed/tested by the submitter.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New projects describe themselves to community lists (recomp.board et al.)
via .recomp.json at the repo root: game, system PS1, type recomp,
toolchain psxrecomp, status exploring, and original region + serial.
write_recomp_json.py takes the serial from disc_probe.json and derives
region from the Sony serial prefix, falling back to --region (which
defaults to USA and would mislabel PAL discs). It only fills missing
fields, so re-running it never clobbers a maintainer's edits.
Both gitignore templates ignored /*.json at the root, which would have
silently dropped the file from the initial commit; un-ignore it.
Closes#406
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Shokunin