Known Issues & Testing
Status reviewed 2026-10-28 against merged baseline 26c3d2e and the playback fixes in PR #35. All nine reports below remain open. Consult the linked issue for newer reports; this page is a dated evidence summary.
MPEG-2/AC3 opening freeze — #29
The reporter's MPG freezes at container opening on an SCPH-77004/USB setup and worked in 2.9rev4. Probing, cursor restoration, short-read handling, metadata retries and progress exits have fixes with host regressions. Retest the unchanged reporter file on the same console/source using standard and DIAG builds. Record the last label: MPEG probing is E48 probe mpg; include its text because diagnostic numbers can overlap other phases.
High-bitrate USB stutter — #24
USB stutter remained in the reporter's later test; MX4SIO reduced it and 2.9rev4 played the same USB video smoothly. PR #35 adds driver-confirmed USB reads with a 64 KiB synchronous refill bound. Unknown/non-USB mass devices retain 4 KiB reads and 32 KiB refills; the MX4SIO workaround is preserved. Host controls establish routing, byte continuity and bounds; console throughput is unmeasured. Compare the identical file, source and settings with matched DIAG read/refill and queue measurements before calling the stutter resolved.
Green or red video color — #20 / #23
Green color and red fluctuation have newer paired recordings and converter logs. Conversion success does not establish correct console output. The exact failing MPG bytes/hash are still needed.
The MPEG RGB ownership fix removes a reproduced held-buffer overwrite. Longer controls in PR #35 exposed signed chroma-vector rounding, XY interpolation and field-row order defects. Their fixes pass a native C/DMA/MMI and field-output oracle; unchanged 150-frame MPEG-1/2 inputs have substantially lower error in both 16/32-bit output. The field-row fix also removes all channel differences over 32 levels in the interlaced B-picture control. Read the measured decoder report for scope and residual errors. These controls exclude the affected console's full display path. Test the unchanged failing MPG, pause/resume, seek, stop and sequence changes on the affected setup. No blanket BT.601/range transform or GOP-default change is supplied.
MP4 late start and stall — #10
The reported 0:28 start and 0:35 video/audio-loop stall remain unconfirmed. MOV packet selection, offsets, timing tables, co64/extended atoms, allocation ownership and failed sample reads have host-tested fixes. Obtain the exact original and converted files and test both ordinary and fast-start layouts through the full player. Faster opening alone does not demonstrate that the stall is fixed.
Saved skin after restart / Spanish — #27
Skin persistence still needs the reporter's USB save/restart test. The late-configuration repaint and skin bounds/allocation paths were fixed, and the Spanish strings were integrated. Test default PAL/NTSC, a non-default output mode and card/USB launch paths; check Spanish glyphs and label clipping separately. HD gradient behavior is not the same issue.
Folder shuffle — #25
Folder shuffle is implemented, with failed-open and playlist-lifetime regressions covered on the host. Enable Randomize playlist under Player settings → MP3 settings, then use Play all → Video. Check every eligible item appears once, Stop/next behavior, missing/unreadable files and a large folder on console.
Converter tracks and subtitles — #22
Multi-audio, text burn-in, source-SD sizing and the high-quality Xvid/PCM preset are implemented and host-tested. SMS audio switching, SD aspect/display and high-bitrate playback on suitable storage remain acceptance work. Switchable embedded text, sidecar export and bitmap conversion are separate from the implemented burn-in control.
Poster / cover-art grid — #26
Draft PR #34 implements an optional bounded JPEG cover grid for #26. Same-directory, same-basename JPEGs supply covers; missing or unsupported artwork uses normal icons. Host lookup/cache/navigation and lifecycle controls pass. The feature remains outside the merged release pending console rendering, device removal, repeated navigation/playback transitions and resource-stability acceptance.
How to read validation claims
| Evidence | What it establishes |
|---|---|
| Host tests / sanitizers | Specific C/Python behavior with fixtures and controlled interfaces; not PS2 scheduling, bus throughput or display output. |
| PS2 target build / ABI checks | Compilation, linking and checked module imports; not actual device operation. |
| Controlled PCSX2 probes | Only the exercised decoder, buffer, IPU and scheduling paths; not a full console playback result. |
| Real-console comparison | The recorded console, source, media bytes, build and settings. Do not generalize a passing profile to every encode or transport. |
For the detailed reproductions and next-agent work, read the engineering handoff and color-control report. Use the reporting checklist for a new result.