Status

Status and where to continue

What development main has, what is open, and how to pick each thread up. Updated 2026-10-22. The plan as a checklist is the roadmap.

Releases

Version	Highlights
0.8.3	Mods and Patches pages fixed; character editing in every language; optional FSR 4.1.1 on Metal; trophy notices drawn by the game; instanced tessellated draws on Metal
0.8.2	High-resolution character lighting fixed after texture collection
0.8.1	GX worker crash fixed; MetalFX option; tessellation shader and recurring shader-cache rebuild fixed
0.8.0	Motion-vector crash fixed; settings shared with the in-game menu; outputs above 1080p; the experimental Metal renderer; FPS counter sizes; memory estimate; trophy notices only over the game
0.7.0	Stretched geometry fixed; local trophies
0.6.0	Profiles and saves; the opening-video crash fixed

Imported bbport compatibility: 0.51, plus parts of 0.6 (below).

On main

Area	State
bbport 0.6	Ported: renderer fixes, the present's wait, frame rate through hooks (no measurable cost; F3 for 30 FPS), Laurence's cinematic fixes, live effects, aspect ratios, keyboard and mouse. Next: online (roadmap)
Hooks in the game's code	bbport's InstallGuestCodeHook interface on bbcpu: a stub int3; replaced instructions; jmp back, every register available to the hook
Metal renderer	The frame draws through Metal behind Advanced → Renderer (65.5/61.7 against 58.0/58.1 FPS on MoltenVK); it runs with the recompiled library. The repeatable native-1080p GX worker crash is fixed. MetalFX is an optional upscaler on either renderer
FSR 4.1.1 on Metal	AMD's neural upscaler in Metal on Apple Silicon (runtime #1, from fsr4-native-macos), offered when the Mac can run it and the player has added AMD's model files from their own DLL. M3 Pro, 1080p Balanced: 4.2–4.8 ms of GPU time a frame, a few FPS below FSR 3.1
High-resolution character lighting	0.8.2 keeps GPU-written images whose contents cannot safely be downloaded. Native 1440p clothing highlights returned on Metal and MoltenVK in copied-save runs
Trophy notifications	The game draws notices in its ImGui overlay after a persisted unlock; the launcher imports and lists unlocked trophies
Save editor	The launcher previews characters and edits all six attributes, Blood Echoes and Insight in a backed-up, staged save set; level and saved HP or stamina follow attribute changes. Max all reaches level 544. Edited saves cannot earn new trophies
Docs	One English set per repository; wikis mirror them

Open threads

After 0.8.2. The repeatable GX worker crash on Metal and MoltenVK was the live dynamic-shadow setting re-enabling shadows in a secondary context without a shadow target. A separate null guest virtual call seen once on Metal has not recurred. A GXRenderThread null read in Cathedral Ward has a different signature and is not confirmed fixed; a nearby save is needed to reproduce it. The FSR 3.1 dark picture on an M5 is also unconfirmed: the reported resolution and upscaler settings rendered normally on this M3 Pro in both 0.8.0 and current main. The reporter's check on 0.8.2 is needed. Check the frame-rate hooks in play (running above 90 FPS, cloth, Laurence's cinematic; needs a save near it).

Metal renderer, next. The render adapter is approved: the renderer moves onto the project's own GPU interface (steps A0–A6, Vulkan backend), then a Metal backend creates every resource in Metal and MoltenVK leaves the macOS build (M1–M4). A0–A6 are done: the renderer runs on the adapter with the Vulkan backend (its own vocabulary, resources, commands, pipelines that hold their shaders, the host passes, frames, the swapchain interface and the upscalers' images). M1 is done too: with BB_GFX_BACKEND=metal every buffer and image is made in Metal and handed to MoltenVK, at the same frame time. M2's first step is done: everything both backends run records through the adapter, at M1's speed (4.04–4.17 µs per draw against 4.05–5.01). With the Metal frame (BB_METAL_FRAME=1) the commands are now encoded into Metal directly (step 2), pipelines and shaders are built only in Metal (step 3), the window is presented and drawn through Metal with the upscaler's own passes (step 4; FSR 3.1's library and DLSS still Vulkan), and every submission is committed in Metal (~433 a second, none through MoltenVK). Step 5's automated runs passed: 36.2 FPS on the Metal frame against 27.6 on MoltenVK (FSR 3.1 Balanced, the test save's place) and a 10-minute session without faults. Next: the maintainer's play session on a copied save closes M2; then M3 (MetalFX without copies, FSR 3.1 ported to Metal). Shaders ahead of time follow (N2). The shader compiler, rewrites, pipeline cache and format/state tables came from Huax-1839/bloodborne-macos, through Wooloomooloo2/bloodborne_mac_port (we took 0b50589 and 78b59f5; its "M2" 3ee456f creates resources in Metal).

Recompiled code speed (R3). The library runs the whole game but takes 20–30% more CPU than the translator. Candidates are listed in bloodborne-recomp's CLAUDE.md. The launcher does not offer it yet (a "CPU: Translator / Recompiled" option is a candidate).

Watch list. A short white flash in the opening video and an enemy's body drawn as colour noise have not recurred in recent play. Neither has a confirmed fix. Older rare assertions are in the runtime's stability notes.

Online. bbport 0.6's module (Seamless Hunt with BBSC co-op, The Hunter's Dream) needs its dependencies in deps/, its hooks on our base, and the maintainer's decision on edited saves.

Keep in step

• Memory: run.sh sizes the game's memory; the launcher (src/lib/data/memory.ts) and the in-game menu (MemoryEstimate in bbport_overlay.cpp) estimate it the same way.
• Settings the menu shows: the runtime's bbport_settings.cpp and the launcher's settings_store.rs (GAME_SYNCED).
• Versions: the app and the launcher.

Test notes

• Test games are background-only apps (SDL_MAC_BACKGROUND_APP=1): they never become the active app, so focus and Cmd+Q behave unlike a launcher game. BB_BENCH_FOREGROUND=1 for a test the maintainer plays in.
• The runtime's Linux and macOS CI run on every push and cancel older runs; wait for a run whose result matters before pushing again.