Read the compatibility guide and architecture before changing platform code. Keep implemented behavior separate from tested hardware results.
- On this development host, select
PS5_CLANG=/usr/bin/clang; its installed Clang works. Do not install an older compiler merely to satisfy a filename. - The shared development console uses
SOFTWARE_AUDIO=1. Always deploy it withmake deploy PS5_CLANG=/usr/bin/clang SOFTWARE_AUDIO=1so the optional CPU decoder is preserved. Read the build guide before changing that dependency. - Console helpers resolve
PS5_HOSTfrom the environment or ignored.env. Do not hardcode developer addresses, account IDs, credentials or report hosts. - Preserve the configured HTTPS endpoint, signed-in account and preferences. Do not exercise dashboard actions, password changes or uploads against the production server without explicit authorization and a reversible test.
- Fully close the app before replacing its files. The folder and native package share a title ID; use a separate test identity when validating installation.
- Do not infer successful launch from an accepted install request or a format check. Native FPKG needs compatible kernel and A53/PPR support, separately from the directory loader. Do not load new firmware offsets by assumption.
- Image-sized and larger allocations use
bigalloc::allocate. The runtime heap is small; flexible memory has substantially more capacity. - Decoder direct memory is type 12.
0x32and0x33are mapping protections, not allocation types; confusing them can take the console down. - Keep video decode at pipeline depth 3. Depth 1 serializes difficult 4K streams. Read a frame slot only while the decoder is recorded as holding it.
- Background threads use
images::spare_cores(). They inherit the creator's affinity and otherwise contend with the render thread. Never spin workers on short sleeps. - Do not write large files from the render thread on a timer. Capture and trace output are explicitly requested through markers and may stall a frame.
- Failure paths must park rather than return from native
main. Shutdown belongs to the main loop; do not present after releasing VideoOut. - Use POSIX reads for file contents. This runtime's
freadhas returned zero bytes for files that other interfaces read correctly. - Dashboard uploads and whole-file reads must fit the constrained heap. Shrink and re-encode images; stream log tails or use flexible memory for large data.
- SlopFin is sandboxed.
/datais available;/userand USB paths require the explicit picker payloadassets/slopfin-sandbox.bin. FTP visibility does not establish that the application can read a path.
- HDR follows the console setting through
sceVideoOutGetOutputStatusbyte 4. Do not reintroduce a user-facing marker for it. Preserve the console's HDR setting after experiments; the shared console uses registryVIDEOOUT_hdr=0. A successful HDR buffer change does not prove the TV received HDR. - Never scale, mix, reorder or capture-write an IEC 61937 bitstream ring. Re-run
the FFmpeg byte comparison when changing
src/iec61937.hpp. - Autoplay decisions live in
src/autoplay.hppand run from drawing, so an input-owning prompt cannot prevent an episode ending. Playback overlays must participate in the frame-reuse test or they will not appear. - Text/button focus uses shared dark surfaces, cyan outlines and visible-glyph alignment. Apply the same tokens and motion to Settings, Dashboard, login, profile and sidebar. Keep focused labels bright and bounds stable.
- Text entry uses the PlayStation system keyboard. Host preview captures cannot demonstrate its appearance, sound or final placement.
- Library names and IDs come from Jellyfin. Keep long sidebars, empty libraries, missing images, failed requests and large episode result sets navigable.
Run make test, make host and make lint as appropriate to the change.
Native builds must have zero static FSELF errors. Visual changes require a
reviewed capture. Decoder, HDR, audio, memory and timing claims require console
validation; the Linux preview cannot establish them.
Use documented layouts, reference decoders or known input values as comparison authorities. Two copies of the same output can share the same mistake. Inspect test harness behavior before accepting measurements.
- Captures stall their own frame. Do not capture while measuring focus or
playback timing; write
/data/slopfin-inputdirectly instead. - Judge frame timing by its distribution. At depth 3, the trace's
decodefield measures submission, not completed decode; use completion counts and shown FPS. Trace markers can drop or reorder, so absence is not proof. - Read crash reports first and symbolize only against the matching build ELF. Keep frame pointers and line tables. A stale process can remain after a crash.
- Confirm the startup build stamp before judging deployment. The loader may
rewrite
eboot.bin; remote byte equality is not a reliable build check. - Caption captures with their environment and limits. Remove private names, addresses and credentials before publishing logs or screenshots.
Controller, capture, launch, timing and crash helpers are documented in
testing, Linux preview and
diagnostics. The app closes on /data/slopfin-quit.
tools/launch.sh is for developer iteration; a crashed instance may be killed
before relaunch, so use the orderly close path where possible.
Use the existing Allman C++ formatting. Keep comments for constraints and non-obvious decisions; discussion and investigation notes belong in docs or review descriptions. Preserve copyright/SPDX headers and the ProsperoTV GPL-3.0-or-later attribution. Dependencies retain their own licenses. Do not commit credentials, caches, generated SDK archives or proprietary modules. Binary distributions include the applicable notices and corresponding source; optional FFmpeg builds also distribute the matching FFmpeg source.