Repository navigation
Known Issues
Status reviewed 2026-10-30 against merged master 34000cf, including browser, MPEG preset and MOV sample-size follow-ups through PR #42. All nine reports below remain open. Consult the linked issue for newer reports; this page is a dated evidence summary.
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.
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 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 is supplied. PR #41 changed MPEG-1/2 converter presets from GOP 50 to 15 to shorten prediction chains. Existing converted files are unchanged: re-convert the same source and compare both encodes on the affected setup. This is a mitigation candidate; the specific reconstruction-drift explanation and a complete color fix are not established by host or packaging tests.
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. PR #42 additionally rejects incomplete variable sample-size tables, with production-reader truncation controls. This does not establish faster opening. 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.
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 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.
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.
PR #34 merged an optional R2 JPEG cover grid for #26 into master at 3eab0ee. Same-directory, same-basename JPEGs supply covers through a six-thumbnail cache; missing or unsupported artwork uses normal icons. See setup, limits and controls. Lookup, cache, navigation, packet and lifecycle behavior have host controls. Merged follow-ups through PR #40 also cover long browser/HOST/M3U paths, navigation allocation failures and failed device entry. Master includes PR #35's playback fixes; console rendering, device removal, repeated navigation/playback transitions and resource-stability acceptance remain pending. #26 remains open.
| 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.