The BIOS exception handler ran interpreted for the life of every process,
1.22 billion instructions and 44.6% of all interpreted work in Breath of
Fire III, even though it is compiled and even though the stub that
unblessed it sat at exactly the address bios/SCPH1001.toml already
declared as an install slot. Two mechanisms existed and did not compose:
the emitter hooked the declared slot PC, while the runtime's bless
verifier memcmp'd the whole body including that slot, so the body was
MISMATCH forever and native code never reached the hook.
Three parts, none of which pays off alone:
1. Publish the declared ranges to the runtime. The emitter emits a
PsxKernelPatchRange table next to the kernel body table and routes it
through the backend descriptor; the bless verifier compares a body in
segments that skip them. CLEAN now means "every byte outside every
declared range still matches ROM", which is the claim the native code
actually depends on.
2. Generalise the slot model to what the patches are. The old model
assumed a 4-word lui/addiu/jalr/nop stub written over ROM zeros and
fired on `word != 0`. Measured against both images, only one of five
observed shapes fits: patchers also plant a `jr` over real ROM
instructions, rewrite an 11-word prologue and insert an `mfc0 Cause`,
and NOP out a driver routine. A slot is now a range with a length and
an explicit `resume` (jalr / fallthrough / none), detected by
comparing every word against its ROM-baked value.
3. Hand back from the interpreter at the range end. Kernel page 0 is
permanently dirty, because the handler saves registers there on every
exception, so straight-line interpretation would run the whole
function after the patch. The emitter registers the resume PC as a
continuation key and dirty_ram_interp surfaces there.
A PC inside a declared range must never be a native dispatch key: the
compiled bytes there are not what executes. Seven pre-existing
jal-return continuations violated that and wedged the boot at frame 0 in
an infinite dispatch loop (OpenBIOS RAM 0x357C). The emitter now drops
continuation keys inside a range so dispatch misses through to the
interpreter, and refuses to emit if a range covers a function entry,
which would silently remove a callable function instead.
Both profiles declare their measured ranges, and SCPH1001.toml finally
pins [program.image] sha256 to v2.2 (CRC32 37157331) so a v2.0 dump
refuses to generate rather than producing phantom discovery gaps.
Measured on Breath of Fire III (SLPS-00990, Psy-Q SDK 3.70), headless,
matched frames, PSX_KERNEL_PATCH_RANGES=0 for the OFF side:
OpenBIOS SCPH-1001 v2.2
bless mismatch 21 -> 0 74 -> 0
kernel bodies -87.7%/frame -97.8%/frame
all interp insns -42.6%/frame -94.2%/frame
The residual kernel-body work is exactly declared-words x entries on
every range, so nothing leaks past the hook. The A0/B0/C0 vector
trampolines are unchanged (+0.2%, noise): the profile excludes them by
design and they are a separate job. Card read traces are byte-identical
across the A/B on both images, all 32 entries, which exercises the two
ranges that replace the kernel's card handler.
Rule 18 holds throughout: the guest's patched instructions still execute
as written on the dirty-RAM interpreter. Only the code around them
changes backend. Nothing is synthesised.
Tests: runtime/tests/test_kernel_patch_ranges.c (new, the verifier's
accept/reject behaviour and the hand-back predicate) and the install-slot
range invariants in recompiler/tests/bios_address_model_test.cpp.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Focused port of the fork's fix/setup-host-bios-deadlock feature (10 commits) onto current upstream master, per the 2026-09-10 decision to port rather than rebase (the fork line and upstream share no merge base).
Content: the codegen host accepts any recompiled BIOS stem instead of only OpenBIOS/SCPH1001 (psx_bios_known_images.h); the runtime checks the setup host before validating a chosen BIOS and the first-run loop breaker reports a cause-specific reason; the CLI decodes subprocess output as UTF-8; --setup-selfcheck verifies a headless setup; the rbengine preflight names the recursive init. Two port conflicts resolved: the new CTest registrations in recompiler/CMakeLists.txt are kept (the names exist only in this change), and runtime.cmake takes the new recomp-ui guidance text. Cubic P3 finding fixed: the sibling PSX_NETPLAY FATAL_ERROR now also names --recursive.
Evidence: python syntax checks pass for psxrecomp_cli.py and the new tests; runtime/tests/test_codegen_host_bios_stems.py runs or skips only on a missing RECOMP_UI_ROOT. No full runtime build or retail route is claimed here. Developed with AI assistance (Cline, 2026-09-10).
Matched Ape Escape retail SCPH1001 LLE builds identify 0661ce77 as the visible boot regression: parent 155e269b reaches game code, while 0661ce77 stalls near BIOS frame 375. Revert its four dependent follow-ups too; retain PR #326 readback bounds and PR #328 Vulkan aspect changes.
The row-by-row raster dispatcher amplifies an existing malformed DMA list into massive GL work. Keep the broader CPU/DMA arbitration experiment separate for compatibility testing. Document attribution, validation, and the temporarily restored interlace limitation.
Validation: full Ape target reaches title/menu and frame 11669 on retail LLE; OpenBIOS control reaches frame 3701. Eight focused tests pass; retained GL readback fixture passes 71 checks at both 1x and 4x after linking local SDL with iconv. Beads: beads-iisy; DMA follow-up: beads-eio.3.122.
psxrecomp's chat was already complete -- client, callbacks, and a host-relayed
LAN path snesrecomp does not have -- but nothing tested it. The SNES side got
this coverage when its chat was written; this brings psx level, against the
same properties, so the two cannot drift apart silently.
Covers what the ring actually promises: room order and both ends, local vs
peer vs system identity (decided by player id, not by who called send, because
the server echoes our own line back and that echo is the copy we keep),
monotonic seq, out-of-range reads refused, wrap-around dropping the OLDEST --
a chat that discards what was just said is worse than no chat -- empty lines
ignored, seq surviving a clear so a UI tracking "newest seen" does not skip
the first line of a new room, and an over-long line clipped to the field
rather than past it.
Links the REAL recomp_net and WS helpers rather than a wall of stubs: nothing
here reaches a socket, so fakes would buy nothing and could drift from what
the runtime links. Only psx_host_mono_ms is supplied, because it lives in the
runtime's own main TU.
Verified to discriminate: making the ring refuse new lines at capacity instead
of evicting the oldest fails two assertions.
ctest with -DPSX_NETPLAY=ON: 67/67 pass (2 disabled).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DYwgPRbwbXz8FhpVzDM7tA
starvation_watchdog_check() sampled the clock before loading the heartbeat,
but the heartbeat is also stamped by the debug-server IO thread on every
send() chunk. A stamp landing between the two reads made last > now, the
uint64 subtraction wrapped past any threshold, and a healthy runtime exited
with code 2 (BoF3 2026-09-02: 402 us of real staleness tripped the 4 s
watchdog within ~2 min whenever the area poller was attached).
- Load the heartbeat first, decide through the new pure helper
starvation_watchdog_stale(), which refuses a negative gap by construction
(same shape as the frame_pacing_sleep_ms fix).
- Make s_last_heartbeat_us _Atomic with relaxed store/load.
- Tag the exit origin: "starvation_watchdog" on the trip, console_close /
console_ctrl_break / console_ctrl_c in the Win32 console handler, and
signal_sigint / signal_sigterm in the soft-exit handler. These all used
to reach psx_atexit_handler as atexit/unknown, indistinguishable from an
SDL window close.
- Add starvation_watchdog_test (ctest) pinning the decision function.
Verified on BoF3 build-dbg: 10 min soak with the poller polling at 1 s and
the watchdog enabled at its 4 s default, no dump, process alive; forced
trip with PSX_STARVATION_TIMEOUT_US=1000 fires and the run report reads
exit_origin "starvation_watchdog".
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
host_keymap_load zeroed an action's binds when its line was present and then
apply_defaults() put the built-in key straight back, so "Turbo = None" (or
an empty value) silently re-armed Tab. Track explicit_unbound per action:
set when the line is present but yields no binds, and honoured by
apply_defaults through want_default(). A missing line still receives the
default, so existing config.ini files are unaffected.
Pairs with the launcher's Backspace-to-clear (recomp-ui
feat/additional-ui-functionality), which writes "None" for that reason.
README documents the rule; test_rewind_toggle_combo.py guards it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Turbo (Tab / [hotkeys] fast_forward_pad) is hold-to-run only. Add a
press-to-latch twin that feeds the same manual fast-forward block:
- host_keymap: HOST_KEYMAP_TURBO_TOGGLE, [KeyMap] TurboToggle, default F9.
- main.cpp: g_manual_turbo_latched flipped by the key edge or by a fourth
host shortcut, PSX_ASSIST_BIND_FAST_FORWARD_TOGGLE ("Fast-forward
toggle", [hotkeys] fast_forward_toggle_pad, unbound by default), polled
next to Rewind / Save states through hotkey_pad_binding_down(). OSD shows
"Fast forward: Nx (locked)" / "Fast forward: off". Threaded through every
settings hop (settings.toml -> globals -> launcher seed -> LauncherSettings
-> back), mirroring fast_forward_pad.
- config_loader: [hotkeys] fast_forward_toggle_pad read/write.
- README: describe the toggle and state that a single-button / trigger host
shortcut is an implicit chord with Select (hotkey_pad_binding_down), which
the launcher now labels as "select + ..." (recomp-ui companion change).
- test_rewind_toggle_combo.py guards the new sites.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The repository root carried seven markdown files that were not onboarding
material, plus two that had been superseded by the code they described. This
sorts them: reader-facing docs to docs/, working notes to docs/internal/, and
the two superseded ones merged into the docs that replaced them.
Moved to docs/: WIDESCREEN.md, ENHANCEMENTS.md, TCP_COMMANDS.md
Moved to docs/internal/: PRINCIPLES.md, ISSUES.md, DEBUG.md, GTE_LM_FIX.md,
and from docs/: bios-hardening.md, overlay-status.md,
overlay-plan.md, overlay-recompilation-design.md,
overlay-discovery.md, TOMBA2_EXC_CONSULT_RESPONSE.md,
CASE_A_AOT_GAP.md, audit_inventory.md
Moved to docs/internal/upstream/: UPSTREAM_2026-08-20.md
The root keeps only what belongs there: README.md, CONTRIBUTING.md, LICENSE,
CLAUDE.md, THIRD_PARTY_ATTRIBUTION.md, RELEASE_NOTES.md, the dotfiles and
psxrecomp_cli.py.
MERGES
SPLITGEN_MIGRATION.md folded into docs/BUILDING.md as "Split generated C". Its
migration instructions were stale in a way that mattered: it told every game
repo to hand-edit its CMakeLists and add a file(GLOB ...) over
<serial>_full*.c, but runtime/runtime.cmake now does that itself. GEN_FULL_GLOB
is a multi-value argument of psxrecomp_add_game_runtime(); the loop at
runtime.cmake:1985 makes each pattern absolute against CMAKE_CURRENT_SOURCE_DIR,
expands it with file(GLOB), and concatenates the hits into
GAME_GENERATED_FULL_C. So a title names the pattern and gets however many shards
a regeneration produced. The rewritten section describes that, the two-step
fallback when the globs match nothing (GEN_FULL_FALLBACK, else <serial>_full.c
derived from the dispatch marker by substituting _dispatch.c), and the one
caveat the code actually has: that file(GLOB) is plain, not CONFIGURE_DEPENDS,
so a regeneration which changes the shard COUNT needs a re-configure to be seen.
psxrecomp-toml-readme.md folded into docs/GAME_PROJECT_SETUP.md as "EXE-only
autofill (psxrecomp-toml)", placed after "Disc autofill (probe_disc.py)" and
before "Project Studio" so the two autofill paths sit together. Its documented
flags were checked against the usage text in recompiler/src/main_toml.cpp:28-42
and against the argument parser at :150-165 -- all seven options match with no
drift, so the flag list is reproduced as-is. Added a pointer that probe_disc.py
is preferable when a cue exists, since that tool does the same EXE-header work
plus disc identity and netplay gates.
FUNCTIONAL FIXES
tools/gen_tcp_commands.py wrote to REPO/"TCP_COMMANDS.md". After the move that
path does not exist, so the generator would have created a new stray root file
and --check would have failed against it forever. Retargeted to
REPO/"docs"/"TCP_COMMANDS.md", along with the seven docstring and message
strings that name the path.
Running it surfaced two further problems, both fixed here:
* The doc was genuinely stale -- pre-existing drift, not introduced by this
commit (the moved file is byte-identical to origin/master, and no
debug-server command table is touched here). Four commands registered since
the last regeneration were missing from the index -- cdrom_timing_dump,
savestate_status, ws_ui_groups, xprobe_watch -- and scanline had gained a
backend. Regenerated: 307 -> 311 commands, and the change is confined to the
AUTOGEN block (the hand-written head and tail are byte-identical to master).
tools/gen_tcp_commands.py --check now exits 0.
* DOC.write_text() has no newline argument, so Python translated every newline
to os.linesep. On Windows that rewrote all 585 lines of the file to CRLF and
turned a four-line index update into a whole-file diff -- measured: 589 CRLF
/ 0 LF after a run, against 0 CRLF / 585 LF in the repo. Now written through
an explicit newline="" handle, verified to leave the file at 0 CRLF.
runtime/include/debug_server.h:9 and runtime/src/debug_server.c:8 cited "TCP.md
and DEBUG.md". TCP.md has never existed in this repo under that name; both now
name docs/TCP_COMMANDS.md and docs/internal/DEBUG.md.
LINK AND REFERENCE FIXES
Every inbound reference to a moved file was located by grepping each basename
across the whole tracked tree, not just the markdown. Markdown hrefs in
README.md, CONTRIBUTING.md, docs/ARCHITECTURE.md, docs/COMPILING_OVERLAYS.md,
docs/AOT_OVERLAY_PLAN.md, docs/RELOCATION_MANIFEST_FORMAT.md,
docs/config_schema.md, docs/OVERLAY_XREF.md, docs/TCP_COMMANDS.md and several
docs/internal/ files now point at the new locations; where two files became
siblings under docs/internal/ the path was shortened rather than lengthened
(bios-hardening.md's reference to overlay-discovery.md, for instance).
docs/COMPILING_OVERLAYS.md's "in this folder" phrasing was wrong once two of the
three files it listed moved, so it now names internal/ explicitly.
24 source-comment mentions of ENHANCEMENTS.md sections G1.x across 19 files
(recompiler, runtime, cmake) were retargeted to docs/ENHANCEMENTS.md with the
section numbers untouched. docs/SHADOW_ENHANCEMENTS.md is a DIFFERENT file and
was deliberately excluded.
20 rationale strings citing docs/CASE_A_AOT_GAP.md across the four
recompiler/seeds/*.json files were updated. Done as a byte-level substring
replacement rather than a JSON round-trip: two of those files are stored with
CRLF endings and two with LF, and re-serializing them reformatted 5,700 lines.
The byte-level version changes exactly 20 lines and each file still parses.
References that look similar but point elsewhere were deliberately left alone,
and are worth naming so a future pass does not "fix" them:
* "recomp-template PRINCIPLES.md" / "NES recomp template PRINCIPLES.md" in
CLAUDE.md:43, docs/SHADOW_ENHANCEMENTS.md:28,
docs/internal/HLE_SCHEDULER_CARVEOUT_PLAN.md:5,
runtime/src/debug_server.c:4934 and tools/parity_diff.py:8 are a different
repository's file. Only the five that mean this repo's PRINCIPLES.md were
updated.
* Every remaining bare ISSUES.md mention is a GAME repo's tracker
(MegaManX6Recomp 2/7/8/9, TombaRecomp 6), not this repo's.
* docs/disc-speed.md:4,54,56 and docs/internal/bios-hardening.md:5 name
overlay-discovery as a BRANCH, not a file, so no path applies.
INDEXES
README.md's "New here?" table gains rows for docs/WIDESCREEN.md,
docs/ENHANCEMENTS.md and docs/TCP_COMMANDS.md. docs/internal/README.md gains
sections indexing all fourteen documents that moved in, plus an "Upstream
contributor reviews" section describing upstream/ and the new sweep file, plus a
"Removed" bullet for CYCLE_TIMING_ARCH.md (deleted in the previous commit)
pointing at FAITHFUL_TIMING_PLAN.md, CYCLE_MODEL_BEETLE.md and
PRECISE_IRQ_SLICE.md -- following the SLJIT_PERSIST_CACHE.md precedent already
in that file.
docs/ENHANCEMENTS.md:7 claimed framework bugs live in "game repos' ISSUES.md
until a framework tracker exists". A framework tracker does exist; it now says
so and names the Beads epic.
VERIFICATION
A link checker was written to scan every tracked *.md for relative links,
resolving each against the tree being scanned. Baseline at origin/master: 129
files, 137 relative links, 3 broken. After both commits: 127 files, 159 relative
links, 3 broken -- the SAME three, all pre-existing and untouched here:
docs/DATA_SHARDS.md -> INSTANT_LOADS_PARKED.md (never existed)
docs/DATA_SHARDS.md -> INSTANT_LOADS_SPIKE2.md (never existed)
docs/NETPLAY.md -> ../lib/recomp-net/README.md (inside a submodule)
Zero new broken links. The checker's own coverage was checked against a raw grep
for markdown link openers: the 3-link difference is C++ lambdas inside code
blocks, correctly not treated as links.
Also verified: py_compile over all 113 tracked tools/*.py plus psxrecomp_cli.py;
all four seed JSONs still parse; the test-registration guard still reports 123
test files all registered; recompiler/tests/test_coverage_vault.py still passes;
and no file in this commit had its line endings flipped (checked mechanically
across all 50 modified files, after two files were caught and redone).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Removes 204 tools/ scripts (203 .py/.sh plus tools/_missing_funcs.json) that
no tracked file references. Verified mechanically, not by eye: for each of the
204 stems, a word-boundary grep across every tracked text file in the repo,
excluding files that are themselves being deleted, returns zero hits. So
nothing kept imports, execs, or documents any of them. Most are the
underscore-prefixed one-shot probes the .gitignore convention already treats
as scratch (tools/_*.py is ignored; these predate that rule and were force-
added), plus a long tail of session-specific dump/decode/poll scripts with
hardcoded local paths and addresses.
Also removed at the root:
* card2.mcd.bak.before_beetle_copy -- a stray backup of a memcard.
* .mcp.json -- local MCP server wiring, not repo configuration.
* SCPH1001.toml -- a root duplicate of bios/SCPH1001.toml, still on the
pre-refactor [[recompiler.bios_vectors]] schema while the bios/ copy has
moved to [recompiler.address_model]. Every reference in the tree already
names bios/SCPH1001.toml, so the root copy was a stale trap: editing it
would have had no effect on any build.
* docs/CYCLE_TIMING_ARCH.md -- superseded by
docs/internal/FAITHFUL_TIMING_PLAN.md, CYCLE_MODEL_BEETLE.md and
PRECISE_IRQ_SLICE.md; nothing linked to it.
Memcard fixtures move out of the repo root into
runtime/tests/fixtures/memcards/ with a README recording what each of the
four images actually contains (directory entries read out of the images, not
guessed). Nothing in the build read them from a fixed path -- the four
hardcoded consumers (tools/check_card.py, inspect_card.py, peek_cards.py,
verify_card_buffer.py) are all in the deletion above.
.gitignore is corrected to match: runtime memcards a session creates stay
ignored (*.mcr anywhere, /*.mcd at the root, which is where the runtime drops
card1.mcd/card2.mcd when no path is configured), while the tracked fixtures
are re-included. The old !dummy.*.mcr negation existed only to keep the two
root oracle cards tracked and is replaced by
!runtime/tests/fixtures/memcards/*. Verified with git check-ignore --no-index
on each moved file, on the pre-existing runtime/tests/test_card_synth.mcd and
runtime/tests/test_dir/card1.mcd fixtures (all not-ignored), and on
hypothetical root card1.mcd / card2.mcd / saves/card1.mcr (all ignored).
beetle_libretro.cpp's warn-only memcard sanity check keeps its behaviour; its
WARNING now names where populated fixtures can be copied from. The cards are
opened relative to the process's working directory because the libretro
GET_SAVE_DIRECTORY environment callback answers "." (same file, line 419), so
the message says to copy them into the CWD rather than implying a repo path.
BIOS profiles SCPH101.toml and SCPH5552.toml move from the root into bios/,
joining OpenBIOS.toml and SCPH1001.toml so that bios/*.toml is the whole set.
Behaviour-preserving: config paths inside a profile resolve against
find_project_root() (config_loader.cpp:821), which walks up from the profile's
directory for .gitignore/.git/CMakeLists.txt -- bios/ has none of those, so it
lands on the same repo root as before, and bios/SCPH1001.toml already relied
on exactly that. tools/build_cli.py's staging map is updated to the new
sources; it already copied them to framework/bios/, so the packaged layout is
unchanged. recompiler/src/main_cli.cpp's bios/-first lookup keeps its root
fallback (harmless, and it is what makes older layouts still resolve).
Fixes a real latent break while in bios/SCPH5552.toml: seeds pointed at
"seeds/phase2_ghidra_seeds_SCPH5552.json", but there is no seeds/ directory at
the repo root -- the file lives at
recompiler/seeds/phase2_ghidra_seeds_SCPH5552.json, which is what SCPH101.toml
already named. Porting these two profiles to the address_model schema is
deliberately NOT done here; it is a separate tracked follow-up.
Test files that no build system knew about:
* tools/test_coverage_vault.py PASSES (8/8). Moved to
recompiler/tests/test_coverage_vault.py and registered as the ctest
coverage_vault, which puts it inside runtime/check_test_registration.cmake's
glob so it can never silently rot again. Its ROOT is adjusted from
parents[1] to parents[2] for the new depth, matching every other python
test in that directory. Re-run from the new location and from an unrelated
CWD: 8/8. The registration guard was run standalone before and after, and
with the add_test() deliberately broken as a negative control -- it does
name recompiler/tests/test_coverage_vault.py as an orphan when
unregistered, so the registration is load-bearing rather than decorative.
* tools/test_compile_overlays_additive.py FAILS, 10 of 14 tests erroring
against an API that has moved on, so it is removed rather than wired up.
It calls four module attributes that no longer exist in
tools/compile_overlays.py -- load_additive_captures,
validated_cached_bundle_ids, cached_bundle_pairs, compile_and_publish_dll
(zero occurrences each) -- and two signatures have changed under it:
parse_overlay_func_ids() no longer takes require_stored_crc, and
load_region_coverage() now takes 2-3 positional arguments where the test
passes 5. Reviving it means rewriting it against the current additive
capture API, which is a change to overlay-cache test coverage and needs its
own commit and its own evidence, not a rename.
Verification: py_compile over all 113 remaining tracked tools/*.py plus
psxrecomp_cli.py passes; bash -n over every remaining tracked tools/*.sh
passes; importing sdk_progress, pad_delivery and cursor_writers from tools/
succeeds.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Matthew Stanley