Merge plan after v0.4.0: PRs tested by us, not yet in the release

[itaskaev2]

@valentderah, the merge plan for the release after v0.4.0. It replaces #965, which I am closing: everything from it that was merged is in v0.4.0.

From #965, now in v0.4.0

#917 (sync_ahead, #676 by ProSep), #916 (dynamic resolution, #573 by ProSep), #1040 (ipavlin98), #1060, #1061, #1066 and #1006 (Luxury213). Thank you for all of them.

Merged after v0.4.0

#1112 (fixes #1106, DLSS on 1.2.1), #913 (NVIDIA Reflex, #638 by STS78721), #1071, #864 (tests/run_all.py, #675 by zzs181), #1087 (grass_range) and #1092 (the mouse wheel), #1118 (the game window's own mouse cursor image back, as pointer_image, off by default), #1142 (fixes #1122, the setup names the program that holds totk.exe) and #1002 (1.1.0 patches by shenjackyuanjie); on 2026-11-14 #1161 (with #1184), #1162 and #1163 (ipavlin98) and #1165 (shadow as a known totk.ini key: the false "unknown key 'SHADOW'" warning is gone); on 2026-11-15 #1200 (contact shadows leave faces turned away from the sun alone). Thank you.

Also in dev since v0.4.0 (04ccb9c is the head as of 2026-11-14 14:16):

• other authors' PRs: Unpack a release archive again when its pin changes #1075, Pick the oldest way of a target history set by index from the set's start, so a new target no longer overwrites the chunked replay's state (#1104) #1128 and Record a new pool-bound image's first transition and clear inside the chunk that makes it #1164 (ipavlin98), Serve a mod's file named for another game version under its own name instead of leaving it out #1101 (xellow78-lgtm), The left mouse button presses A on the Purah Pad's map (map_click, off by default, game 1.4.3) #1096, In dialogue the left mouse button presses B and the right one A (dialogue_click, off by default, game 1.4.3) #1097 and PC prompts: text prompts show Y as its keyboard key (T by default), and the setup draws them from totk.ini's bindings #1098 (lwj10case), Ask fontconfig for the menu's CJK font instead of hard-coding five paths #1115 (jcleng), Add optional mouse wheel and keyboard gameplay tweaks for Ultrahand, weapons, quick menus, and map #1136 and Add separate Music, SFX, and Voice volume sliders (#722) #1146 (RxEmam), Fix incorrect HDR tonemapping to match SDR presentation #552 (Zenrap0518), Lift again only the chunks whose inputs changed #996 (Wesley-CR, the lift cache), Linux AppImage: ignore user site-packages so installer.gui imports #990 (jay23606) and docs: escape table-splitting pipes and unlink the missing docs/milestones dir #1108 (viwe-monai); Fix incorrect HDR tonemapping to match SDR presentation #552, Add separate Music, SFX, and Voice volume sliders (#722) #1146 and docs: escape table-splitting pipes and unlink the missing docs/milestones dir #1108 came in through the maintainer's merges and their PRs still show open;
• the maintainer's own: the Random crash: access violation reading 0xffffffffffffffff in f_7001bd0138 (main+0x1bd0138) #1005 crash guards (optref_guard, then the ability patch set, ability_guard), the V-register locals in the lifter (6732297, 4-6 % less guest time in a town), ov_ability.c compiled per install, the shaped RomFS folders written only for the player's screen shapes, freeing every DLSS size slot on a preset change, the replay comparing and copying its per-draw state inline (about 5 % less render-thread time, Are further performance improvements planned? #381) and the recorder woken on every publish (FPS degrades over play‑session with bound:cpu, GPU idle, no memory leak (0.4.0 vs stable 0.3.11) #1174).

Tested by us, not in dev yet

2026-11-15, 12:46: dev is at 434533a (version 0.4.1, #1207). Every PR of ours below merges into it without conflicts (GitHub: mergeable, clean). #914 (FSR 3.1) is not going into dev (maintainer, 2026-11-15): no frame generation in the FSR 3.1.5 libraries, ~2 ms of bridge CPU per frame, and the plan is AMD FidelityFX SDK 2.4 (#876, #1089). It stays open as a reference; we no longer carry it onto dev. #1188 is rebased onto c26c3a0 and out of draft.

2026-11-15: all of them are rebased again, onto dev 94eb3d4, and show as mergeable. The conflicts:

• runtime/cfg.c's name list in SMAA on the present (#624 by ipavlin98, on dev) #915, AMD FSR 3.1 as the other DLSS / DLAA engine, on dev (#228 by Lucas-blip-png) #914, Run The Second Quest on 1.4.3: MSFix as host hooks, IPS data writes, mods without their code module (from ericdallo's fork) #790 and Achievements tab: statistics from the save and achievements on them #732;
• the menu next to the new audio device row in Achievements tab: statistics from the save and achievements on them #732;
• in Run The Second Quest on 1.4.3: MSFix as host hooks, IPS data writes, mods without their code module (from ericdallo's fork) #790, the patch table keyword. dev's function-entry hook lines (6a23628) took the name Run The Second Quest on 1.4.3: MSFix as host hooks, IPS data writes, mods without their code module (from ericdallo's fork) #790 used for its host hooks, so Run The Second Quest on 1.4.3: MSFix as host hooks, IPS data writes, mods without their code module (from ericdallo's fork) #790's line is now host_hook. dev's hook lines are unchanged (details).

Each was built on 1.4.3 and passed --selftest. Each then ran in the world under the Vulkan validation layer with its feature on: 0 validation errors, 0 shader failures.

• SMAA on the present (#624 by ipavlin98, on dev) #915: vk: SMAA on.
• AMD FSR 3.1 as the other DLSS / DLAA engine, on dev (#228 by Lucas-blip-png) #914: FSR 3.1 anti-aliased the frames, no device loss.
• Achievements tab: statistics from the save and achievements on them #732: achievements on.
• Opt-in quick save and quick load with no menu on the screen (F6 / F7 by default, rebindable) #1188: quicksave on (reworked on 2026-11-15, see its row below).
• Run The Second Quest on 1.4.3: MSFix as host hooks, IPS data writes, mods without their code module (from ericdallo's fork) #790:
  • plugins_test, lift_cache_test and exefs_patches_test pass;
  • --probe-plugins and --probe-mods pass;
  • with The Second Quest and the mods on a TSQ save: MSFix ... runs, 7 hooks on, 5718 frames.

Every PR of ours below shows as mergeable. All of them were rebased onto dev (42b36b0) on 2026-11-14. Two had conflicts: #914 in runtime/vk_dlaa.c, next to #1162's scene_shaped (both kept), and #790 in recomp/gen.py, next to the V-register locals (6732297). For #790, recomp/vlocals.py now also stores and reloads the V locals around a host hook (rt_hook), as regs.ESCAPE already does for the X registers: the MSFix hooks run host code in the middle of a lifted function. Each was built from a fresh lift of dev (#790 from its own lifts, with and without The Second Quest's ExeFS patches), --selftest passed, and each ran in the world on 1.4.3 (RTX 4080 SUPER) under the Vulkan validation layer with its feature on (SMAA, FSR 3.1, achievements): 0 validation errors, 0 shader failures, and no layer message beyond the one warning dev itself shows (a fragment output without an attachment). #790 with The Second Quest, its Russian texts, TSQ Free Warp and a 4K texture pack, on a TSQ save: the MSFix hooks run, 59.7 fps, into the world. Under the validation layer that run crashed inside the layer itself while loading (VkLayer_khronos_validation.dll, 11 GB committed with the 4K pack), so it was run without the layer. #865's compile gate passes locally on the rebased branch (124 files, 0 failed). Since then dev moved to 04ccb9c (#996's lift cache, the inline replay state), and #790 and #732 were rebased onto it the same day (the setting-name list in runtime/cfg.c; for #790 also recomp/patches.py and docs/building.md). #790's hooks already reach the lift cache's key (a hook is a ('hook', index) word of its site, a fn: argument a function start in module_key); tests/lift_cache_test.py now has a case for it, and lift_cache_test, plugins_test and exefs_patches_test pass. Both built, selftested and ran in the world again: #732 with achievements on under the validation layer, 0 errors; #790 on a TSQ save with the mods, the MSFix hooks on, 59.7 fps. Under the validation layer one of two #790 runs without mods stopped inside the layer itself (VkLayer_khronos_validation.dll, the same place as the run with the mods above); the rerun and two dev runs went through, so I count it as the layer's, not proven. #1188 now has rebindable keys, F6 and F7 by default as in Half-Life. Each stands alone, and every new setting is off by default.

Ready

PR	What	Tested
#1219 = #439 by Specht2010	Contact shadows darken only pixels the game's own lighting puts in the sun (the half-resolution lighting pass's sun visibility); caves and shade keep their ambient light. Carried onto dev over the scan-planned march and on top of #1200	dev 6151ac4 vs the branch: canyon, Depths, a Brightbloom cave and Pondside Cave under the validation layer, 0 errors, 0 shader failures; the lighting pass found in every scene; GPU 5.79 vs 5.74 ms. In a player's daily build since 2026-11-15. Screenshots in the PR.
#915 = #624 by ipavlin98	antialiasing = smaa on the present	Iris Xe: works, costs 14 %; 0 validation errors. Rebased onto dev on 2026-11-12 and run again: vk: SMAA on, 0 validation errors.
#865 = #656 by jkoehler11	The runtime compiles on both hosts in CI; the AppImage builds in a pinned container	Both workflows pass on the PR (Windows 3 min, Linux 9 min).

Our own features, waiting for your review

PR	What	State
#790	The Second Quest on 1.4.3: MSFix as host hooks, IPS data writes, mods without their code module	Runs on six saves and in daily play with The Second Quest and the HDPacks 4K pack on dev; plugins_test.py passes. The plugin part waits for the step 2 design, #845.
#732	Achievements tab	Your eight points are done.
#497 (draft, main)	Setup and in-game menu in the game's languages	Synced with v0.4.0. I can move it onto dev.
#367 (main)	A working-loop skill for coding agents	Unchanged.
#1188	Opt-in quick save and quick load with no menu on the screen (quicksave, off by default; keys quicksave_key / quickload_key, F6 / F7 by default as in Half-Life). Reworked on 2026-11-15, out of draft: F6 asks the game for its own autosave (as its OneShotAutoSaveMgrRequestAutoSave node does, the wait between autosaves skipped) and keeps a copy of the slot; F7 holds the picture on "Loading the quick save" while the game's load list runs underneath and its slot reads open the quick save's files. A toast says what happened	Rebased onto dev c26c3a0, vanilla lift: --selftest, the probes, settings_doc_check, mouse_direct_check and check_content pass. Lookout Landing: F6, walk away, F7 brings Link back (save 0.5 s, load 4.9 s), 0 validation errors; on a copy of a long play F6 writes the oldest autosave slot, never the manual one. Real keys in a player's daily build since 2026-11-15. Screenshots and the log in the PR.

Other authors' PRs we tested

PR	Result	State
#1139 #1005 guard for a reused freed holder (cdonatom)	The five new words disassemble as commented; the lifted f_7001bd0138 checks x21 >> 32 in [1, 127]; three saves on 1.4.3 at 60 fps with optref on, 0 validation errors, no crash (comment). The 3.5 h crash itself is not reproducible here.	closed by its author on 2026-11-14: dev's ability guard (e683938, c89b0f3) covers all four of the author's crash targets (comment)
#1148 read-only control socket for companion maps (phatnguyen-v1)	Code read and approved (review): listens on 127.0.0.1 only, closes a connection that starts like HTTP, read-only commands, off by default; not built here	mergeable (rebased by its author on 2026-11-13)
#995 Deck replay_async (deadlynch)	The profile turns it on, 40 fps held; not measured on a real Deck	closed
#996 lift cache (Wesley-CR)	61 s to 4 s on a second lift; output byte-equal to --no-cache	merged on 2026-11-14
#972 in-game update (m3rcury02)	13 update tests pass; install-and-restart not run	conflicts with dev
#1054 usnap_fast (t04dJ14n9)	No gain on a Ryzen 7 5800X3D (median 125.9 fps off, 120.1 on); off by default	conflicts with dev
#975 view formats (snailium)	Never triggers in TotK; a compatibility-class check would help before other games rely on it	mergeable (base main)
#1024 motion sensor recovery (RafaelSoaresP, draft)	The SDL virtual-pad test passes; needs a physical pad	draft
#1050 launcher (paulosnp, base main)	Builds and works with dev's installer; downloads are checked by size only, not SHA-256	closed

Not ready

The rest of #836's "not ready" list stays as it was: #487, #690, #647, #741, #669, #657, #748.

Suggested order

1. Compile the runtime on both hosts in CI, and build the AppImage in a pinned container (#656 by jkoehler11, on dev) #865: CI on both hosts, so the rest is checked on every PR.
2. Darken only sunlit pixels with contact shadows (carries #439 onto dev) #1219: contact shadows (off by default) stop flattening caves and shade; it carries Darken only sunlit pixels with contact shadows #439 onto dev, so Darken only sunlit pixels with contact shadows #439 can be closed with it.
3. SMAA on the present (#624 by ipavlin98, on dev) #915: an optional setting, off by default.
4. Add a read-only local control socket for companion apps (TOTK_CONTROL) #1148 (approved, off by default); Update to a new release from the settings menu #972 after its author's sync.
5. Run The Second Quest on 1.4.3: MSFix as host hooks, IPS data writes, mods without their code module (from ericdallo's fork) #790, Achievements tab: statistics from the save and achievements on them #732, Opt-in quick save and quick load with no menu on the screen (F6 / F7 by default, rebindable) #1188: after your review; all three are ready to merge.
6. Setup and in-game menu in the game's languages, with a Languages table in the README #497 (draft, main): I can move it onto dev when you want it.

[itaskaev2]

@valentderah the plan above is up to date with dev 434533a (0.4.1). All our pull requests to dev are ready to merge: #865, #915, #914, #790, #732 and #1188. Each merges into the current dev without conflicts; how each was built and tested is in the plan and in its PR.

Suggested order: #865 first (CI on both hosts), then #915 and #914, then #790, #732 and #1188. #1188 is reworked since you last saw it: quick save and quick load now work with no menu on the screen, and it was also checked on 1.2.1. #497 stays a draft on main; I can move it onto dev whenever you want it.

If something comes up after a merge, I'll rebase the rest the same day.