Much of the heavy lifting of this optimization is lifted from the Pixman
project, which is distributed under an MIT-style license. As far as possible,
these elements have been relicensed to the zlib license.
Fixes an issue in macOS 10.15 where the displayed content would move up after entering, exiting and re-entering exclusive fullscreen when certain display modes were used (bug #4822).
Bug #3949 is also related to this change.
Use eglGetProcAddress for everything on EGL >= 1.5. Try SDL_LoadFunction first
for EGL <= 1.4 in case it's a core symbol, and as a fallback if
eglGetProcAddress fails. Finally, for EGL <= 1.4, fallback to
eglGetProcAddress to catch extensions not exported from the shared library.
(Maybe) Fixes Bugzilla #4794.
Sylvain
Seems to be a regression in this commit: https://hg.libsdl.org/SDL/rev/7fdbffd47c0e
SDL_CalculatePitch() was using format->BytesPerPixel, now it uses SDL_BYTESPERPIXEL().
The underlying issue is that "surface->format->BytesPerPixel" is *not* always the same as SDL_BYTESPERPIXEL(format);
BytesPerPixel defined as format->BytesPerPixel = (bpp + 7) / 8;
vs
#define SDL_BYTESPERPIXEL(format) ... (format & 0xff)
Because of SDL_pixels.h format definitions, one is giving a BytesPP 1, the other 0.
"This patch does the following:
* Instead of SDL_FillRects calling SDL_FillRect in a loop the opposite
happens -- SDL_FillRect (a specific case) calls SDL_FillRects (a general case)
with a count of 1
* The switch/case block is moved out of the loop -- it modifies the color
once and stores the fill routine in a pointer which is then used throughout
the loop"
Fixes Bugzilla #4674.
The 10 ms delay effectively caps input polling at 100 Hz and rendering
at 100 FPS if applications use these functions in their event loop. The
delay may also lead to dropped frames even at 60 FPS due if they are
unlucky enough to hit the delay and rendering takes longer than 6 ms.
fix building with Mesa 19.2
With Mesa 19.2 building fails with:
/include/GLES/gl.h:63:25: error: conflicting types for 'GLsizeiptr'
The same type is defined in include/SDL_opengl.h for OpenGL and the two
headers should not be included at the same time.
This was just never noticed because the same header guard '__gl_h_' was
used. This was changed in Mesa. The result is this error.
Fix this the same way GLES2 already handles this: Don't include the GLES
header when the OpenGL header was already included.
(https://hg.libsdl.org/SDL/rev/6a3670d6108d)
The X11 target sets mouse->last_x and last_y in EnterNotify and then calls
SDL_SendMouseMotion(), which throws away the new position because it matches
the mouse->last_x and last_y we just set, meaning that if the pointer is
in the window when it created, SDL_GetMouseState() will report a position of
0,0 until a MotionNotify event (the pointer moves) arrives and corrects the
mouse state.
Mostly fixes Bugzilla #1612.
The SDL_USE_LIBDBUS define is set inside SDL_debug.h, therefore the
circular dependency made it impossible for this feature to be enabled.
Instead, guard SDL_dbus.h based on the autoconf variable HAVE_DBUS_DBUS_H
Additionally, fix one of the rtkit comments. CAP_SYS_NICE isn't required
to achieve high priority. But there is some scheduler config that rtkit
needs the app to setup.
The Offscreen video driver is intended to be used for headless rendering
as well as allows for multiple GPUs to be used for headless rendering
Currently only supports EGL (OpenGL / ES) or Framebuffers
Adds a hint to specifiy which EGL device to use: SDL_HINT_EGL_DEVICE
Adds testoffscreen.c which can be used to test the backend out
Disabled by default for now
Background:
Chengdu Haiguang IC Design Co., Ltd (Hygon) is a Joint Venture
between AMD and Haiguang Information Technology Co.,Ltd., aims at
providing high performance x86 processor for China server market.
Its first generation processor codename is Dhyana, which
originates from AMD technology and shares most of the
architecture with AMD's family 17h, but with different CPU Vendor
ID("HygonGenuine")/Family series number(Family 18h).
Related Hygon kernel patch can be found on:
http://lkml.kernel.org/r/5ce86123a7b9dad925ac583d88d2f921040e859b.1538583282.git.puwen@hygon.cn
Best regards.
The LE transformation for vec_perm has an implicit assumption that the
permutation is being used to reorder vector elements (in this case 4-byte
integer word elements), not to reorder bytes within those elements. Although
this is legal behavior, it is not anticipated by the transformation performed
by the compilers.
This causes pygame-1.9.1 test failure on PPC64LE because blitted pixmaps are
corrupted there due to how SDL uses vec_perm().
From RedHat / Fedora: https://bugzilla.redhat.com/show_bug.cgi?id=1392465
Original patch was provided by: Menanteau Guy <menantea@linux.vnet.ibm.com>
If KMSDRM_drmModeSetCrtc is called when the swap interval is
set to 0, the driver behaves as though vertical sync is engaged by
limiting framerate to the refresh rate, but performance is much worse
than with vertical sync enabled.
Resolve this issue by ensuring that the Crtc is only set up once,
and KMSDRM_drmModePageFlip is called, albeit without any followup
queueing or waiting for flips.
Daniel Drake
A long time ago, it was possible to play neverball on Linux using the accelerometer found in HP laptops.
The kernel exposes the accelerometer as a joystick (/dev/input/jsX) as well as an evdev device (/dev/input/eventX). I guess it worked fine when SDL was using the js interface, but then stopped working here: http://hg.libsdl.org/SDL/rev/fdaeea9e7567
Looking at current code which uses udev to discover joysticks, it looks for the udev tag ID_INPUT_JOYSTICK.
However udev's internal input_id logic specifically tags accelerometers as ID_INPUT_ACCELEROMETER and nothing else.
This looks like a good fit for SDL_HINT_ACCELEROMETER_AS_JOYSTICK.
Ozkan Sezer
As for the issue: This bmp reports bpp=0, therefore SDL_CalculatePitch()
returns pitch==0, which is then fed to SDL_malloc() (which is malloc())
and malloc(0) returns _something_ which is not NULL but not someting
that we expect.. Then testsprite.c:LoadSprite() accesses the pixels
as *(Uint8*)pixels which valrind reports as:
==15533== Invalid read of size 1
==15533== at 0x8048C08: LoadSprite (testsprite.c:45)
==15533== by 0x80492FC: main (testsprite.c:224)
==15533== Address 0x449e588 is 0 bytes after a block of size 0 alloc'd
==15533== at 0x40072B2: malloc (vg_replace_malloc.c:270)
==15533== by 0x4045719: SDL_CreateRGBSurface (SDL_surface.c:126)
==15533== by 0x40403C1: SDL_LoadBMP_RW (SDL_bmp.c:237)
==15533== by 0x8048BB2: LoadSprite (testsprite.c:36)
==15533== by 0x80492FC: main (testsprite.c:224)
Besides, valrind also reports this:
==15533== Conditional jump or move depends on uninitialised value(s)
==15533== at 0x40403F3: SDL_LoadBMP_RW (SDL_bmp.c:247)
==15533== by 0x8048BB2: LoadSprite (testsprite.c:36)
==15533== by 0x80492FC: main (testsprite.c:224)
Easy/quick solution would be early-rejecting a bmp with 0 bpp from SDL_bmp.c:SDL_LoadBMP_RW()
Caleb Cornett
SDL_ShowMessageBox on UIKit doesn't do anything special with buttons that are marked with the flag SDL_MESSAGEBOX_BUTTON_RETURNKEY_DEFAULT. According to Apple's documentation on UIAlertController, a button can respond to a return key if it's marked as the preferredAction of the controller. SDL doesn't set a preferredAction currently, so I've attached a patch to fix that.
M Stoeckl
To reproduce:
1. Run any SDL-based program with a Wayland compositor, orphaning it so that it doesn't have an immediate parent process. (For example, from a terminal, running `supertux2 & disown`.) The program should use the wayland backend, i.e. by setting environment variable SDL_VIDEODRIVER=wayland.
2. Kill the compositor process.
Results:
- The SDL program will keep running.
Expected results:
- The SDL program should close. (What close should mean here, I'm not sure - is injecting an SDL_Quit the appropriate action when a video driver disconnects?)
Build data:
2019-06-22, hg tip (12901:bf8d9d29cbf1), Linux, can reproduce with sway, weston, and other Wayland oompositors.
This is currently supported on Linux and macOS. iOS and Android are not
supported at all, Windows support could be added with some changes to the libusb
backend. The Visual Studio and Xcode projects do not use this feature.
Based on Valve Software's hid.cpp, written in collaboration with Andrew Eikum.
Galadrim
As I have seen, SDL implements its own command line parser for Windows in SDL_windows_main.c. Unfortunately, it doesn't seem to allow command line arguments with trailing backslashes if quoting is required.
Usually, when you write an application that gets command line arguments passed as argc and argv, the parsing is done by parse_cmdline. The Windows API also provides the function CommandLineToArgvW, so an application can parse itself if only the command line string is provided. Both functions behave almost identically according to their documentation. If the argument "\\" (including the quotes) is passed, they both turn it into a single backslash.
The SDL command line parser on the other hand doesn't recognize the second quote character as the closing character in this example and therefore includes it in the parsed argument. The parser does not count the number of backslashes preceding a quote. It always treats a quote as escaped if a backslash is in front of it. Therefore, it should be impossible to quote and escape an argument correctly, if it has a trailing backslash and contains characters that require quoting.
Of course, each application is allowed to implement its own parsing rules, so SDL is free to do so. But the problem I see is that there are arguments, that are impossible to be passed to the parser correctly, as I described above. Is there a reason, why SDL does not simply use CommandLineToArgvW instead of implementing its own parser?
Here are some links that show that correct argument parsing, as it is usually done in Windows, is quite complicated:
https://docs.microsoft.com/en-us/windows/desktop/api/shellapi/nf-shellapi-commandlinetoargvwhttp://www.windowsinspired.com/how-a-windows-programs-splits-its-command-line-into-individual-arguments/
Caleb Cornett
Unlike iOS and macOS, tvOS does not have any persistent local storage. In fact, the ApplicationSupport directory pointed to by the existing Cocoa GetPrefPath() throws an error whenever any attempt is made to access it. To get any local storage on an Apple TV, our only option is to use a temporary cache directory.
This patch changes the tvOS PrefPath to this cache directory and also logs a critical warning that this if developers want their save data to persist across game sessions, they must use some form of iCloud storage.
superfury
I notice that, somehow, when locking the mouse into place(using SDL_SetRelativeMouseMode), somehow at least the movement information gets through to both mouse movement and touch movement events?
My app handles both, so when moving a touched finger accross the app(using RDP from an Android device) I see the mouse moving inside the app when it shouldn't(meaning that the touch movement is ignored properly by the app(press-location dependant) but the mouse movement is still performed due to the mouse movement events)?
alexrice999
I have a knock off wired xbox 360 controller called afterglow for xbox 360 controller. Despite there being a few afterglow controllers in the controller mapping the guid of my controller seems to map to Generic Xbox pad. This binding is as follows:
```
"030000006f0e00001304000000010000,Generic X-Box pad,a:b0,b:b1,back:b6,dpdown:h0.4,dpleft:h0.8,dpright:h0.2,dpup:h0.1,guide:b8,leftshoulder:b4,leftstick:a0,lefttrigger:a2,leftx:a0,lefty:a1,rightshoulder:b5,rightstick:a3,righttrigger:a5,rightx:a3,righty:a4,start:b7,x:b2,y:b3,",
```
When running openmw I have a strange issue that the joysticks work for up and down movements but not for side to side. I managed to track this down to the side to side events being classified as joystick events instead of gamepad events.
I believe this is due to both "leftstick" and "leftx" being bound to "a0" which seems odd to me. If I change openmw's mappings to remove these the controller works as expected. I was hoping someone who knows a lot more than me (as I have only been exploring this today trying to fix my controller) would know what is happening
With multitouch, register to receive XI_Motion (which desctivates MotionNotify),
so that we can distinguish real mouse motions from synthetic one.
(bug 4690)
Windows generates fake raw mouse events for touchscreens for compatibility
with legacy apps that predate touch support in Windows. We already handle
touch events explicitly, so drop the synthetic events to avoid duplicates.
This time, we make anything we think is a MacBook trackpad report its touches
as SDL_MOUSE_TOUCHID, even though they're not _actually_ synthesized events,
and let all mouse input--even if the OS synthesized it from a multitouch
trackpad on our behalf--look like physical input. This is backwards from
reality, but produces the results most apps will expect.
Note that if you have a real touch device that doesn't appear to be the
trackpad, it'll produce real touch events with unique device ids, so it's
not a total loss here, but also note that the way we decide if it was the
trackpad is an imperfect heuristic; it happens to work out right now, but
it's not impossible that a real touchscreen could come to the Mac at some
point and (incorrectly?) call it a "mouse" input, etc.
But for now, good enough.
Fixes Bugzilla #4690.
Anthony Pesch
The previous code first configured the period size using snd_pcm_hw_par-
ams_set_period_size_near. Then, it further narrowed the configuration
space by calling snd_pcm_hw_params_set_buffer_size_near using a buffer
size of 2 times the _requested_ period size in order to try and get a
configuration with only 2 periods. If the configured period size was
larger than the requested size, the second call could inadvertently
narrow the configuration space to contain only a single period.
Rather than fixing the call to snd_pcm_hw_params_set_buffer_size_near
to use a size of 2 times the configured period size, the code has been
changed to use snd_pcm_hw_params_set_periods_min in order to more
clearly explain the intent.
Cameron Gutman
The bugfix in https://hg.libsdl.org/SDL/rev/ca9417a52f18 caused SDL_SetRelativeMouseMode() to begin failing on Vivante (Steam Link). Even though Vivante doesn't have a SetRelativeMouseMode() or WarpMouse() function, it's in relative mode already (because it uses evdev) so the function was actually working as intended.
I think providing a no-op SetRelativeMouseMode() function for Vivante is a reasonable fix. Since it's already getting relative events through evdev, it really is a no-op to "enter relative mode".
In fact, this is probably the right thing to do for all backends that use evdev (vivante, raspberry, and kmsdrm). Raspberry and kmsdrm both have WarpMouse() implementations so SDL_SetRelativeMouseMode() isn't failing there, but it still seems to make sense not to have to do the fake warping if they're evdev-based anyway.
"Starting with changeset 12433, the mouse cursor is not displayed on the
Raspberry Pi platform, due to a bug in the handling of the new
"global_cursor" in RPI_ShowCursor(). Currently, if cursor == global_cursor,
the function immediately returns 0. The function should not return here.
Instead, if cursor == global_cursor, it shouldn't try to hide the current
cursor and update global_cursor = cursor. However, it *should* still continue
through the rest of the function."
Fixes Bugzilla #4699.
The real problem is that SDL_atomic.c was built in thumb mode instead of ARM mode, which is required to use the mcr instruction on ARM platforms. Added a compiler error to catch this case so we don't generate code that does infinite recursion.
I also added a potentially better way to handle things on Linux ARM platforms, based on comments in the Chromium headers, which we can try out after 2.0.10 ships.
daniel.c.sinclair
Hi, this patch breaks dpad/hat input on my PS4 controller. The attached patch restores functionality. Calling SDL_PrivateJoystickHat() at the end of BSD_JoystickUpdate was setting the hat state to zero on every kind of input, instead of just the HUG_DPAD events.
Using IOKit for this pops up a warning at startup on macOS 10.15 ("Catalina"),
asking the user to authorize the app to listen to all keyboard input in the
system, which is unacceptable.
I _think_ we were using IOKit under incorrect presumptions here; the Stack
Overflow link mentioned in it was complaining about not being able to use
flagsChanged to differentiate between left and right mod keys, but that's not
an issue for capslock.
It's also possible this code was trying to deal with capslock changing when
the window didn't have focus, but we handle this elsewhere now, if we didn't
at the time.
GetWindowText() wants you to tell it the size of the buffer--including the
terminating NULL char--but we weren't counting that last char, losing the
last char of the string in the process. This was only seen with the special
case of SDL_CreateWindowFrom() to use an existing native window, not
the usual SDL_CreateWindow() codepath.
Fixes Bugzilla #4696.
It causes the HIDAPI devices to always be opened on enumeration, which causes crashes in the Windows drivers when multiple applications are reading and writing at the same time. We can revisit this after 2.0.10 release.
The Nintendo USB GameCube Adapter has two USB connectors. Black for data
and grey for additional power for rumble. The Wii U and other software
require both cables to use rumble. The rumble is weaker without the
second USB cable. Other than that I don't know if there is any negative
side affects from using rumble with only one cable.
Jan Martin Mikkelsen
Patch to scan /dev/dri based on names rather than file type
Loading KMS/DRM on FreeBSD fails because the "available" code in the driver checks for character device nodes under /dev/dri and the /dev/dri/card* files are symlinks rather than device nodes nodes on FreeBSD. The symlink points to /dev/drm/0.
The attached patch counts /dev/dri/card* entries rather than directory entries which are character devices.
Sylvain
I think what happening with the software renderer is:
* you're somehow in background (so texture creation is not possible)
* it resizes and wants to push a SDL_WINDOWEVENT_SIZE_CHANGED
It call:
https://hg.libsdl.org/SDL/file/a010811d40dd/src/render/SDL_render.c#l683
* GetOutputSize
* SW_GetOutputSize
* SW_ActivateRenderer
* SDL_GetWindowSurface
* SDL_CreateWindowFramebuffer which is mapped to SDL_CreateWindowTexture
and it ends up re-creating the surface/a texture, while being in background
Manuel Sabogal
There is an issue on the latest commit of the mercurial repo when SDL_THREADS_DISABLED is set:
src/core/linux/SDL_threadprio.c:79:28: error: unknown type name 'Sint64'; did you mean 'int'
It was originally added to work around an input event problem in the code of a specific app which mixed SDL and native UIViews, but that app solved its problems in a better manner since then.
Also implemented SDL_JoystickGetDevicePlayerIndex() on iOS and tvOS, and added support for reading the new menu button state available in iOS and tvOS 13.
(technically, this function never returns an error at this point, but since
it _does_ have an "uhoh, is this corrupt data?" comment that it ignores, we
should probably make sure we handle error cases in the future. :) )
Matteo Beniamino
Pressing a trigger button on a Steam Controller causes a segmentation fault both with stable version and latest mercurial head on Linux. I'm using the recent hid_steam kernel module with lizard_mode disabled (that is no keyboard/mouse emulation). I suspect this is what's happening: the driver exposes two hats. The two hats have indices 0 and 2. Inside linux/SDL_sysjoystick.c two hats are allocated in allocate_hatdata for joystick->hwdata->hats. In HandleHat function the hat parameter (that can be 2) is directly used as the index of the array that only has two elements, causing an out of bounds access. SDL is not expecting to have "holes" between hats indices.
The index 2 is calculated in HandleInputEvents() as (ABS_HAT2X - ABS_HAT0X) / 2 where ABS_HAT2X is the value associated to the hat inside the hid_steam module.
Jan Martin Mikkelsen
The attached patch adds support for single-touch evdev devices.
These devices report ABS_X, ABS_Y and BTN_TOUCH events. This patch sets them up as MT devices with a single slot and handles the appropriate messages.
This was to deal with broken vsync support in macOS 10.14, which we assumed
would remain broken indefinitely, but a later 10.14 released fixed it.
This is a loss of late-swap support, but there are several subtle problems
in our CVDiplayLink code that are also evaporating, to be fair.
Fixes Bugzilla #4575.
(Backed out changeset 8760fed23001)
Dzmitry Malyshau
Current code, search paths, and error messages are written to only consider MoltenVK on macOS as a Vulkan Portability implementation. It's not the only implementation available to the users. gfx-portability [1] has been shown to run a number of titles well, including Dota2, Dolphin Emulator, and vkQuake3, often out-performing MoltenVK in frame rate and stability (see Dolphin benchmark [2]).
There is no reason for SDL to be that specific, it's not using any MVK-specific functions other than the WSI initialization ("VK_MVK_macos_surface"). gfx-portability exposes this extension as well, and a more generic WSI extension is in process. It would be good if SDL was written in a more generic way that expect a Vulkan Portability library as opposed to MoltenVK specifically.
[1] https://github.com/gfx-rs/portability
[2] https://gfx-rs.github.io/2019/03/22/dolphin-macos-performance.html
