February 2025 - Wine-GitLab

[PATCH v3 0/2] MR7119: ntdll, loader: Add __attribute__((used)) to a few symbols that shouldn't be discarded.
by William Horvath (＠whrvth) March 3, 2025

March 3, 2025

These changes are enough to allow building the entirety of the native (ELF) side with `-flto`. -- v3: loader: Mark thread_ldt, thread_data, wld_start "used". ntdll: Mark call_init_thunk and abort_thread "used". https://gitlab.winehq.org/wine/wine/-/merge_requests/7119

4 5

[PATCH 0/6] MR7436: d3dx9: More changes to D3DXSaveSurfaceToFileInMemory().
by Connor McAdams (＠cmcadams) March 3, 2025

March 3, 2025

This patch series reworks `D3DXSaveSurfaceToFileInMemory()` to use the new `d3dx_save_pixels_to_memory()` function. It also adds support for selecting a replacement pixel format when the image file format being saved to doesn't have direct support for the pixel format being saved. The code for calculating the best replacement format was inspired by the code in `check_texture_requirements()` in `d3dx9_36/texture.c`. This should fix WineHQ bug 51584, but I haven't tested that. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/7436

3 11

[PATCH v6 0/7] MR7422: winebus/dinput: Support arbitrary number of axes in the virtual PID descriptor
by Tomasz Pakuła March 3, 2025

March 3, 2025

Related Wine issue: https://bugs.winehq.org/show_bug.cgi?id=52714 I'm starting work on updating/extending Linux force feedback API and one of my goals is to expose the number and an array that contains the device's ffb-enabled axes. As I'm mainly working with USB PID driver, I already have a POC of obtaining this data. What prompted this is that currently Wine always creates virtual joysticks with two PID ffb axes. This causes issues with [Richard Burns Rally](https://github.com/ValveSoftware/Proton/issues/6702#issuecomment-267…, as this game incorrectly initializes a `CONSTANT_FORCE` effect with all the FFB axes that contain `DIDOI_FFACTUATOR` flag and then trying to update `cAxes` and `rgdwAxes` values which leads to `DIERR_INVALIDPARAM`. Now, this flag is set by wine while enumerating virtual device's axes and while `(axis index < FFB axes)`, flag is applied. This means, that every device which has FFB functionality and at least two axes, will report FFB on `X`, `Y`. While updated API would mean a lot of steering wheels would correctly report only 1 axis, a lot of USB PID wheels still include `X` and `Y` in their `AXES_ENABLE` and `DIRECTION` usages. This, again, would lead to broken FFB (in affected games). On Windows, there's a possibility of overwriting some joystick capabilities with registry entries, which ends up removing `DIDOI_FFACTUATOR` flag from a selected axis. This doesn't work in Wine. This MR allows Wine to add an arbitrary number (up to `PID_AXES_MAX`) of axes to the PID descriptor, based on the value of Haptic Axes from SDL or from Linux FF api in the future (I'm working on it). Additionally, to work around RBR and other games with similar, wrong FFB handling, I introduced a [hint](https://github.com/libsdl-org/SDL/issues/12341) to SDL that allows a dynamic overwrite of the number of haptic axes. For the Linux API, I'll figure out a nice way to override when the number of axes will be exposed. With these changes, I was able to successfully get force feedback on my Moza Racing R9 with all axes usable. With additional traces in the dinput code, I confirmed that with one axis defined in the PID descriptor, the game created a constant force effect with just one axis and `cAxes = 1`. -- v6: https://gitlab.winehq.org/wine/wine/-/merge_requests/7422

2 10

[PATCH v4 0/6] MR7250: ntdll: Properly implement NtFlushProcessWriteBuffers (try 2).
by Jinoh Kang (＠iamahuman) March 3, 2025

March 3, 2025

Supersedes !741. On macOS 10.14+ `thread_get_register_pointer_values` is called on every thread of the process. On Linux 4.14+ `membarrier(MEMBARRIER_CMD_PRIVATE_EXPEDITED, ...)` is used. On x86 Linux <= 4.13 and on other platforms it falls back to calling `NtGetContextThread()` on each thread. The fast path patches from @tmatthies are slightly modified in the following ways: 1. On unsupported platforms, the `try_*()` functions return `FALSE` instead of `0`. 2. `try_exp_membarrier()` is called first, then `try_mach_tgrpvs()`. --- Known applications fixed by this MR: - osu! (rhythm game) song selection menu stuttering - .NET CoreCLR GC - HotSpot JVM (-XX:+UseSystemMemoryBarrier) safepoints -- v4: kernel32/tests: Add a store buffering litmus test involving FlushProcessWriteBuffers. https://gitlab.winehq.org/wine/wine/-/merge_requests/7250

3 8

[PATCH v2 0/1] MR7439: winedmo: Find mp3 stream info if channel data is missing.
by Brendan McGrath (＠redmcg) March 3, 2025

March 3, 2025

This fixes missing mp3 channel data in the FFmpeg demuxer. -- v2: winedmo: Find mp3 stream info if channel data is missing. https://gitlab.winehq.org/wine/wine/-/merge_requests/7439

3 5

[PATCH v9 0/1] MR7288: winegstreamer: Avoid large buffer pushes in wg_transform.
by Yuxuan Shui (＠yshui) March 3, 2025

March 3, 2025

## Context This is a problem discovered when running "Max: The Curse of Brotherhood" under proton. This game writes 4MB a piece into `wg_transform`, which contains about 4 seconds of compressed video. `wg_transform` calls `gst_pad_push` in `transform_ProcessOutput`, which takes \~300ms (i.e. 20 frames @ 60fps) to decode this buffer of video. So the end result is the game hitches every 4 seconds while playing cut scene videos. Proton currently has a special case for this particular game, for which it breaks up the buffer into small chunks to avoid long blocks. This MR adopts that and applies it generally. One concern raised by @redmcg is: > The only issue I can think of is if there are any decoders which don't have a parser; then they might not know how to deal with an arbitrary 4096 byte buffer (the parser is generally responsible for taking arbitrary buffers and producing something the decoder can work with, for example: a full frame). So this MR only enables this strategy when there is a parser element. -- v9: winegstreamer: Avoid large buffer pushes in wg_transform. https://gitlab.winehq.org/wine/wine/-/merge_requests/7288

3 4

[PATCH 0/2] MR7362: winex11.drv: Better detect and switch windows to managed.
by Paul Gofman (＠gofman) March 3, 2025

March 3, 2025

-- https://gitlab.winehq.org/wine/wine/-/merge_requests/7362

4 8

[PATCH v5 0/1] MR7353: winewayland: Implement SetCursorPos via pointer lock.
by Attila Fidan (＠atticf) March 2, 2025

March 2, 2025

There is no ([not yet?](https://gitlab.freedesktop.org/wayland/wayland-protocols/-/merge_requ…) dedicated protocol for requesting pointer warps, but it's possible to request one with `zwp_locked_pointer_v1.set_cursor_position_hint` which may be performed when the pointer is unlocked. The idea is to quickly lock/set position/unlock. This is used by SDL (in some cases: [SDL_MOUSE_EMULATE_WARP_WITH_RELATIVE](https://wiki.libsdl.org/SDL3/SDL_HINT…) and Xwayland. The limitation is/will be that the compositor will ignore requests to warp the pointer outside of surface bounds. What I'd like to have working is the auto cursor placement feature of some applications, where they just want to call SetCursorPos towards specific UI elements inside their single window, and usually they'll be unclipped and have a visible cursor while doing this. This patch also happens to allow mouselook to work in applications which use SetCursorPos for mouselook and aren't getting covered by the heuristics that turn on Wayland relative motion (for some reason the driver thinks the cursor is visible). Seems to affect Half-Life (GoldSrc) on my system. Though force enabling relative motion with an environment variable in a custom build or somehow improving the heuristics will also fix that. ~~A suface/pointer can only have one lock or confinement, and this implementation depends on the pointer lock. So if needed, I temporarily unlock/unconfine in order to do the warp then relock/reconfine. I modified `wayland_pointer_update_constraint` so I don't have to temporarily disable relative motion during this in case it's enabled. I think it's safe/a no-op to attempt pointer warps while using relative motion, as according to the protocol the warp will not generate a relative motion event, and while there will be a wl_pointer motion event, they're ignored while relative motion is being used.~~ -- v5: winewayland: Implement SetCursorPos via pointer lock. https://gitlab.winehq.org/wine/wine/-/merge_requests/7353

4 7

[PATCH 0/10] MR7407: windowscodecs: Implement query strings enumerator.
by Nikolay Sivov (＠nsivov) March 1, 2025

March 1, 2025

Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> -- https://gitlab.winehq.org/wine/wine/-/merge_requests/7407

3 19

[PATCH 0/1] MR7440: dwrite: Use Noto Sans as fallback for General and Supplemental Punctuation.
by Daniel Lehman (＠dlehman25) March 1, 2025

March 1, 2025

-- https://gitlab.winehq.org/wine/wine/-/merge_requests/7440

3 6