fix: main's source checks pass again (15 red since #2574, #2617 and others), and the three bugs behind them #2656

Closed

notquiteog wants to merge 2 commits from notquiteog/fix/main-source-checks into main

notquiteog commented

2026-11-12 04:30:54 +00:00

(Migrated from github.com)

main's Game-free source regressions check is red: 15 source checks fail at de866017, from PRs that were merged without a rebase onto each other. This makes them pass again, and fixes the three real bugs that were behind four of them.

What was wrong

Real bugs (the checks were right):

RX_VK_CAPTURE_KEEP_OLD=0 and RX_VK_CAP_BLOCKS=0 turned the switch on (vk_memory.cpp, #2437). Both are on/off switches read as set-means-on. They read through env_on now, as line 1054 of the same file already did. (test_env_on)
Mouse camera: the view snapped back after first-person aim or Z-targeting (host/mouse_native.c, #2617). The mouse's kept angles were reset when the game takes the stick, but not when the camera leaves states 3 and 4, so on return the old angles were written over where the game had left the camera. (test_mouse_native: yaw 4 where -1 is expected)
DLSS could be widened into Ultra Performance below a third (dlss_checks.h, #2574). (w + 2) * 3 >= ow fixed 853x480 at 2560x1440 (853 × 3 = 2559) but also let a size one or two pixels under the third through, which mode_for then widens a range for. It is w * 3 + 2 >= ow now: a third rounded down, which is NGX's own Ultra Performance size, and no less. (test_dlss_checks: "below a third")

Stale translations: three English texts changed in vk_menu.cpp after the 13-language table was made (#1930's ocarina sentence in the gamepad hint, #2362's RADV note in the Mesh shaders hint, #2435's keyboard face in the Button prompts hint). Their 39 entries in menu_i18n.h are keyed and translated again. They are machine-made like the rest of the file and not reviewed by speakers. (test_menu_i18n)

Fixtures that cut source by its text and missed what the source gained:

Check	What it lacked
`test_wayland_libdecor_hint`, `test_display_startup_failure`	`window_title()` (#2116)
`test_texel_buffer_ownership`	`dead_import_bytes` (#1906)
`test_audio_biquad`, `test_audio_aux`	`env_on` in `reverb_frame`
`test_mods_size_table`	a stub for `hos_xinput_pads` (#2435)
`test_readback_part`	its marker matched #2458's forward declaration of `wait_recorded`
`test_hdr_encode`, `test_framegen_preset`, `test_rumble_menu`	`kEngine`'s `retry` and `temporal_reset_failure` (#2574)
`test_linux_docker_packages`	its parser took the workflow's `--no-update` for a PPA

How it was tested

python -I -B tests/run_all.py on this branch (Linux, the emsdk's clang, glslang 16.6.0): 468 passed, 0 failed, 22 skipped (Windows-only, or needing dotnet / PIL / a Vulkan device). On de866017 without it: 15 failed.
The Vulkan back end builds from the branch (Linux, clang, with the NGX SDK).
In the game, RTX 5060 Ti 16 GB, driver 615.71.09, Linux (COSMIC, Wayland), Kokiri Forest, 2048x1152 output: DLSS at Ultra Performance (683x384), Performance, Balanced, Quality, 75%, 83%, 92% and Auto, 54 to 60 fps, each feature made once (feature 2 made ... 683x384 (683x384 to 683x384) -> 2048x1152, mode 3), no device lost.

Not tested

The mouse camera fix and the two capture switches are proven by their source checks, not by a run with a real mouse or with the switches set.
Windows (the fixtures are OS-neutral; CI will say).
The three re-translated menu texts were not read by a speaker of any of the 13 languages.

🤖 Generated with Claude Code

`main`'s *Game-free source regressions* check is red: 15 source checks fail at de866017, from PRs that were merged without a rebase onto each other. This makes them pass again, and fixes the three real bugs that were behind four of them. ## What was wrong **Real bugs (the checks were right):** - **`RX_VK_CAPTURE_KEEP_OLD=0` and `RX_VK_CAP_BLOCKS=0` turned the switch on** (`vk_memory.cpp`, #2437). Both are on/off switches read as set-means-on. They read through `env_on` now, as line 1054 of the same file already did. (`test_env_on`) - **Mouse camera: the view snapped back after first-person aim or Z-targeting** (`host/mouse_native.c`, #2617). The mouse's kept angles were reset when the game takes the stick, but not when the camera leaves states 3 and 4, so on return the old angles were written over where the game had left the camera. (`test_mouse_native`: yaw 4 where -1 is expected) - **DLSS could be widened into Ultra Performance below a third** (`dlss_checks.h`, #2574). `(w + 2) * 3 >= ow` fixed 853x480 at 2560x1440 (853 × 3 = 2559) but also let a size one or two pixels under the third through, which `mode_for` then widens a range for. It is `w * 3 + 2 >= ow` now: a third rounded down, which is NGX's own Ultra Performance size, and no less. (`test_dlss_checks`: "below a third") **Stale translations:** three English texts changed in `vk_menu.cpp` after the 13-language table was made (#1930's ocarina sentence in the gamepad hint, #2362's RADV note in the Mesh shaders hint, #2435's keyboard face in the Button prompts hint). Their 39 entries in `menu_i18n.h` are keyed and translated again. They are machine-made like the rest of the file and not reviewed by speakers. (`test_menu_i18n`) **Fixtures that cut source by its text and missed what the source gained:** | Check | What it lacked | |---|---| | `test_wayland_libdecor_hint`, `test_display_startup_failure` | `window_title()` (#2116) | | `test_texel_buffer_ownership` | `dead_import_bytes` (#1906) | | `test_audio_biquad`, `test_audio_aux` | `env_on` in `reverb_frame` | | `test_mods_size_table` | a stub for `hos_xinput_pads` (#2435) | | `test_readback_part` | its marker matched #2458's forward declaration of `wait_recorded` | | `test_hdr_encode`, `test_framegen_preset`, `test_rumble_menu` | `kEngine`'s `retry` and `temporal_reset_failure` (#2574) | | `test_linux_docker_packages` | its parser took the workflow's `--no-update` for a PPA | ## How it was tested - `python -I -B tests/run_all.py` on this branch (Linux, the emsdk's clang, glslang 16.6.0): **468 passed, 0 failed, 22 skipped** (Windows-only, or needing dotnet / PIL / a Vulkan device). On de866017 without it: 15 failed. - The Vulkan back end builds from the branch (Linux, clang, with the NGX SDK). - In the game, RTX 5060 Ti 16 GB, driver 615.71.09, Linux (COSMIC, Wayland), Kokiri Forest, 2048x1152 output: DLSS at Ultra Performance (683x384), Performance, Balanced, Quality, 75%, 83%, 92% and Auto, 54 to 60 fps, each feature made once (`feature 2 made ... 683x384 (683x384 to 683x384) -> 2048x1152, mode 3`), no device lost. ## Not tested - The mouse camera fix and the two capture switches are proven by their source checks, not by a run with a real mouse or with the switches set. - Windows (the fixtures are OS-neutral; CI will say). - The three re-translated menu texts were not read by a speaker of any of the 13 languages. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

github-sync referenced this pull request from a commit

2026-11-12 06:19:02 +00:00

fix(tests): run_all.py prints a failed check's quotes into a Windows pipe instead of stopping (#2666)

~~github-sync referenced this pull request 2026-11-12 06:24:03 +00:00~~

feat(framegen): AMD FSR 4 and AMD FSR 3.1 as Frame generation by choices #2665

github-sync commented

2026-11-12 14:43:49 +00:00

Owner

@Joey-Magnus reviewed on GitHub · 2026-11-12 14:33 UTC · original

Current main already has every part of this, through other PRs. On main at 51da0464, tests/run_all.py has only test_lift_cache failing here (a local issue), and every check this PR fixes passes. The PR now conflicts in nine test files.

The capture switches read through env_on: #2651 (1101a8e2), gpu/vk/vk_memory.cpp:937/960/995.
The mouse camera's kept angles after leaving states 3 and 4: #2653 (0ec31ff6).
The stale translations: #2654 (69465621).
The source fixtures (window_title, dead_import_bytes, kEngine's retry, the readback marker, the mods stubs): #2648 (9776c36d). The docker PPA parser: #2652 (f82caa6b).

The one piece main did differently is scene_size_fits in gpu/vk/dlss_checks.h. Main kept (w + 2) * 3 >= ow, and #2648 made tests/test_dlss_checks.cpp:29-33 require that one and two pixels under a third fit. So w * 3 + 2 would now fail main's own check. If you think sizes two pixels under should be refused (because mode_for widens them), that's worth its own small PR arguing against those test cases. Otherwise I'd suggest closing this one as superseded.

> [@Joey-Magnus](https://github.com/Joey-Magnus) reviewed on GitHub · 2026-11-12 14:33 UTC · [original](https://github.com/NotRexed/Hookshot/pull/2656#pullrequestreview-5876226972) Current main already has every part of this, through other PRs. On main at 51da0464, tests/run_all.py has only test_lift_cache failing here (a local issue), and every check this PR fixes passes. The PR now conflicts in nine test files. - The capture switches read through `env_on`: #2651 (1101a8e2), gpu/vk/vk_memory.cpp:937/960/995. - The mouse camera's kept angles after leaving states 3 and 4: #2653 (0ec31ff6). - The stale translations: #2654 (69465621). - The source fixtures (window_title, dead_import_bytes, kEngine's retry, the readback marker, the mods stubs): #2648 (9776c36d). The docker PPA parser: #2652 (f82caa6b). The one piece main did differently is `scene_size_fits` in gpu/vk/dlss_checks.h. Main kept `(w + 2) * 3 >= ow`, and #2648 made tests/test_dlss_checks.cpp:29-33 require that one and two pixels under a third fit. So `w * 3 + 2` would now fail main's own check. If you think sizes two pixels under should be refused (because `mode_for` widens them), that's worth its own small PR arguing against those test cases. Otherwise I'd suggest closing this one as superseded.

github-sync closed this pull request

2026-11-12 15:51:25 +00:00

github-sync commented

2026-11-12 15:51:26 +00:00

Owner

@notquiteog commented on GitHub · 2026-11-12 15:15 UTC · original

Closing: main has all of this now through #2648, #2651, #2652, #2653, #2654 and #2661, as the review says. Thanks for checking it against main. The one thing that differs is scene_size_fits: main keeps the two-pixel tolerance with tests that say so, which is a deliberate choice and fine by me.

> [@notquiteog](https://github.com/notquiteog) commented on GitHub · 2026-11-12 15:15 UTC · [original](https://github.com/NotRexed/Hookshot/pull/2656#issuecomment-6827533029) Closing: main has all of this now through #2648, #2651, #2652, #2653, #2654 and #2661, as the review says. Thanks for checking it against main. The one thing that differs is `scene_size_fits`: main keeps the two-pixel tolerance with tests that say so, which is a deliberate choice and fine by me.

github-sync referenced this pull request from a commit

2026-11-12 16:54:32 +00:00

fix(vk): DLSS comes back after a feature that was not made, and frame generation takes the multiplier its sets were made for (#2658)

Commenting is not possible because the repository is archived.

No reviewers