These previously mapped SIZEALL to "move", but "move" is not guaranteed
to be a four-pointed arrow: according to the CSS spec, it's actually
intended to be a drag-and-drop cursor, analogous to "alias" and "copy".
Map it to "all-scroll" instead, as in Wayland: while this is *also* not
semantically guaranteed to be a four-pointed arrow, it is at least
*suggested* to make it a four-pointed arrow.
Signed-off-by: Simon McVittie <smcv@collabora.com>
The "left_ptr" name is an X11 thing, and there's no guarantee that
Wayland cursor themes contain it. In particular, GNOME's Adwaita theme
as of version 46.beta only contains the CSS/freedesktop names.
To test, either move one of the known cursors out of the way, or edit
the switch statement above to use a wrong name for one of them.
Signed-off-by: Simon McVittie <smcv@collabora.com>
These originally checked for expected ± EPSILON as logged, but since
commit 880c6939 they check for expected ± max_err, where max_err may
need to be greater than EPSILON for very large expected results like
the ones in exp_regularCases().
Also, EPSILON is so small that the default precision of the %f format
(6 decimal places) would never actually have shown its effect, so log
it in scientific notation instead.
Fixes: 880c6939 "testautomation_math: do relative comparison + more precise correct trigonometric values"
Signed-off-by: Simon McVittie <smcv@collabora.com>
While looking at the other tests in this file, I noticed that instead
of checking for a result in the range of expected ± FLT_EPSILON as I
would have expected, these tests would accept any result strictly less
than expected + FLT_EPSILON, for example a wrong result that is very
large and negative. This is presumably not what was intended, so add
the SDL_fabs() that I assume was meant to be here.
Fixes: 474c8d00 "testautomation: don't do float equality tests"
Signed-off-by: Simon McVittie <smcv@collabora.com>
In the Steam Runtime 1 'scout' environment, when compiling for i386
using the default gcc-4.6, Exp(34.125) matches the desired value to the
precision shown in the log (6 decimal places) but is not an exact match
for the desired value.
Signed-off-by: Simon McVittie <smcv@collabora.com>
On my laptop, the battery is configured to stop charging at around 80%
most of the time, to increase the overall useful lifetime of the battery.
When in that state, upower reports UP_DEVICE_STATE_PENDING_CHARGE
(numeric value 5), which SDL previously mapped to SDL_POWERSTATE_UNKNOWN.
This made the platform_testGetPowerInfo automated test fail, because
it assumes that SDL_POWERSTATE_UNKNOWN means no battery is connected,
and does not expect to see a percentage.
Map UP_DEVICE_STATE_PENDING_CHARGE (5) to SDL_POWERSTATE_CHARGED, which
seems close enough.
Also map UP_DEVICE_STATE_PENDING_DISCHARGE (6) to
SDL_POWERSTATE_ON_BATTERY, which matches how at least GNOME presents it.
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit eebd5d18a2)
It's reasonable to assume that any of them might need a display and an
audio backend. We run them with SDL_VIDEODRIVER and SDL_AUDIODRIVER
set to dummy anyway.
Signed-off-by: Simon McVittie <smcv@collabora.com>
If the error behaviour in one of these cases was wrong, that could have
been hidden by the error indicator remaining set from a previous test.
Signed-off-by: Simon McVittie <smcv@collabora.com>
A surface of width (0x7fff'ffff) / 2 = 0x3fff'ffff is not quite large
enough to make the pitch overflow in the way we wanted to test here:
with a 32-bit format, that makes each row 0xffff'fffc bytes, which
(just) fits in a 32-bit unsigned size_t. Increasing it to 0x4000'0000
pixels per row is enough to trigger the overflow we intended to test.
In SDL 2, this test bug was hidden by the fact that allocating
0xffff'fffc bytes on a 32-bit platform is very likely to fail, and SDL 2
reported both "malloc() failed" and "this amount of memory is too large
for a size_t" with the same error code.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Adding 3 bytes of alignment to 0x7fff'ffff is not enough to make it
overflow a 4-byte unsigned size_t, so this test was not exercising
the intended failure mode. We cannot actually make this overflow
with a signed 32-bit width and an 8-bit format: the maximum width is
not enough to achieve that. However, if we switch to a 24-bit format,
we can make the calculation overflow.
In SDL 2, this test bug was hidden by the fact that allocating
0x7fff'ffff bytes on a 32-bit platform will usually fail, and SDL 2
reported both "malloc() failed" and "this amount of memory is too large
for a size_t" with the same error code.
Signed-off-by: Simon McVittie <smcv@collabora.com>
AUDIO_F32 is an alias for AUDIO_F32LSB, 32-bit floating point with least
significant byte first, resulting in audio_resampleLoss() failing on
big-endian machines. This particular test generates a buffer with
native-endianness float values, so it ought to be using AUDIO_F32SYS,
which is 32-bit floating point in the native endianness.
Already fixed by commit 47bcb078 "Fixed some incorrect SDL_AUDIO_F32 uses"
on the SDL 3 branch.
Resolves: https://github.com/libsdl-org/SDL/issues/8818
Signed-off-by: Simon McVittie <smcv@collabora.com>
Unlike the test images in the previous commit, this one is
4-bytes-per-pixel RGBA32, so the masks used here appear to be correct
for both endiannesses. Converting it to SDL_PIXELFORMAT_RGBA32 just
makes it more concise.
Signed-off-by: Simon McVittie <smcv@collabora.com>
All of these test images are provided as an octal dump of
3-bytes-per-pixel data with red, green and blue bytes in that order,
referred to as RGB24 in SDL's taxonomy. However, the call to
SDL_CreateRGBSurfaceFrom() used masks that would have been appropriate
for RGBA32 data.
On little-endian platforms, the test images loaded as intended anyway,
because SDL does not actually check all four masks before deciding to
use RGB24 (it only looks at the red mask), and the red channel happens to
be in the 0x000000FF position for both RGB24 and RGBA32 on little-endian.
Unfortunately, on big-endian platforms, the required masks are not the
same and the call failed with "Unknown pixel format". As far as I can
tell, this means testautomation_surface has never succeeded on big-endian
platforms, but presumably nobody has tried to run it on such platforms
until now.
In the SDL 3 branch, this was fixed as a side-effect of commit
932f6134 "Remove mask versions of SDL_CreateRGBSurface* #6701 (#6711)",
which I have used as a reference to confirm that RGB24 is correct.
Resolves: https://github.com/libsdl-org/SDL/issues/8817
Signed-off-by: Simon McVittie <smcv@collabora.com>
We can't rely on irrational numbers like pi being represented exactly,
particularly when compiling for i386, where the i387 floating-point
interface carries out calculations in registers that have higher
precision than the actual double-precision variable. The 1980s were a
strange time.
Resolves: https://github.com/libsdl-org/SDL/issues/8311
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 6248472c0c)
SDL_GetGamepadAxis() is a SDL3-specific name: SDL2 calls gamepads
"game controllers".
Fixes: d3c21666 "gamepad: Clarify range and direction of axes"
Signed-off-by: Simon McVittie <smcv@collabora.com>
When a test has been disabled because it's known not to work reliably
or it's a test for unimplemented functionality, we probably don't want
to encourage developers and testers to run it and report its failures
as a bug.
Helps: #8798, #8800
Signed-off-by: Simon McVittie <smcv@collabora.com>
When cross-compiling, pkg.m4 will set the PKG_CONFIG variable
to a suitable pkg-config executable, which might be a
host-architecture-prefixed name like aarch64-linux-gnu-pkg-config. This
ensures that we will use a suitable search path that is suitable for
the host architecture (the machine we are compiling for) rather than
the build architecture (the machine we are compiling on) to find
dependency libraries like SDL. When using pkg-config as a substitute
for sdl2-config, we will similarly need to use the
host-architecture-prefixed pkg-config.
Setting the PKG_CONFIG environment variable is also the canonical way to
request that a different pkg-config implementation be used, if that
becomes necessary for some reason.
Co-authored-by: Helmut Grohne <helmut@subdivi.de>
Bug-Debian: https://bugs.debian.org/1059749
Signed-off-by: Simon McVittie <smcv@collabora.com>
Previously, if acting on a surface with less than 32 bits per pixel,
this code was placing the pixel value from the surface in the first
few bytes of the Uint32 to be decoded, and unrelated data from a
subsequent pixel in the remaining bytes.
Because SDL_GetRGBA takes the bits to be decoded from the
least-significant bits of the given value, ignoring the higher-order
bits if any, this happened to be correct on little-endian platforms,
where the first few bytes store the least-significant bits of an
integer.
However, it was incorrect on big-endian, where the first few bytes are
the most-significant bits of an integer.
The previous implementation also assumed that unaligned access to a
32-bit quantity is possible, which is not the case on all CPUs (but
happens to be true on x86).
These issues were not discovered until now because
SDLTest_CompareSurfaces() is only used in testautomation, which until
recently was not being run routinely at build-time, because it contained
other assumptions that can fail in an autobuilder or CI environment.
Resolves: https://github.com/libsdl-org/SDL/issues/8315
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit d95d2d7051)
Battle for Wesnoth apparently relies on being able to disable rendering
of UI elements by setting the clip rectangle to be empty.
Resolves: https://github.com/libsdl-org/SDL/issues/6896
Fixes: 00f05dcf "render: only enable clipping when the rectangle is valid"
Signed-off-by: Simon McVittie <smcv@collabora.com>
It is possible for retrieving the machine ID to fail, either because
dbus was installed incorrectly (machine ID absent or corrupt), or in
32-bit builds, because stat() on the machine ID fails with EOVERFLOW
if it has an out-of-range timestamp or inode number.
dbus has historically treated this as a faulty installation, raising
a warning which by default causes the process to crash. Unfortunately,
dbus_get_local_machine_id() never had a way to report errors, so it has
no alternative for that (bad) error handling.
In dbus >= 1.12.0, we can use dbus_try_get_local_machine_id() to get
the same information, but with the ability to cope gracefully with
errors. ibus won't work in this situation, but that's better than
crashing.
(cherry picked from commit 91198baed4)
Mitigates: https://github.com/ValveSoftware/steam-for-linux/issues/9605
Signed-off-by: Simon McVittie <smcv@collabora.com>
This heuristic for gamepads without a more specific mapping already
tried two incompatible conventions for handling triggers: the Linux
Gamepad Specification uses hat switch 2 for the triggers (for whatever
reason), but the de facto standard set by the drivers for older Xbox
and Playstation controllers represents each trigger as the Z-axis of
the nearest analog stick.
Android documentation encourages Bluetooth gamepad manufacturers to use
a third incompatible convention where the left and right triggers are
represented as the brake and gas pedals of a driving simulator
controller. The Android convention also changes the representation of
the right stick: instead of using X and Y rotation as a second pair
of axes, Android uses Z position as a second horizontal axis, and
Z rotation as a second vertical axis.
Try to cope gracefully with all of these. This will hopefully resolve
the issue described in #5406 (when using unpatched kernels).
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit cf1dc66e2c)
The bitfield `mapped` has two different sets of meanings, depending
whether we're setting up the triggers or the d-pad. Represent them
as symbolic constants rather than opaque integers.
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit c4d49fadd4)
AM_PATH_SDL2 doesn't add much beyond PKG_CHECK_MODULES, and having a
special m4 macro for every library that you might depend on scales
poorly.
The macro does add special support for macOS frameworks, but that feature
was broken for around 6 years without anyone noticing (#6141), and is
likely to be only rarely useful according to comments on #6141.
Resolves: #6140
Signed-off-by: Simon McVittie <smcv@collabora.com>
Library-specific foo-config scripts duplicate very similar logic across
various different projects, and tend to break cross-compiling, multilib
(gcc -m32), Debian/Ubuntu multiarch and so on by only being able to have
one sdl2-config at a time as the first one in the PATH.
The direct replacement is pkg-config(1) or a compatible reimplementation
like pkgconf(1), which relies on each library installing declarative
metadata, like SDL's sdl2.pc (available since at least 2.0.0) and
centralizes the logic into the pkg-config/pkgconf tool.
Most uses of `sdl2-config --foo` can be replaced by something similar
to `${PKG_CONFIG:-pkg-config} --foo sdl2`. Instead of adding a custom
sdl2-config to the PATH or using its --prefix or --exec-prefix options,
users of a custom installation prefix can use any of pkg-config's
non-SDL-specific ways to influence the result, for example setting
PKG_CONFIG_PATH, PKG_CONFIG_SYSROOT_DIR or PKG_CONFIG_LIBDIR environment
variables, or setting the PKG_CONFIG environment variable to point to
a wrapper script.
For Autotools specifically, the replacement for AM_PATH_SDL2 (which
will be officially deprecated in a subsequent commit) is
PKG_CHECK_MODULES.
CMake has its own semi-declarative mechanism for dependency discovery,
"config packages", and the SDL build already installs a config
package. There's a good example of using a config package to discover
SDL in `cmake/test/`.
Meson natively supports pkg-config, and already uses it in preference
to sdl2-config.
Other build systems can run pkg-config instead of sdl2-config,
preferably checking the PKG_CONFIG environment variable first.
https://github.com/ioquake/ioq3 is a good example of a project doing
this correctly.
Helps: #6140, #3516
Signed-off-by: Simon McVittie <smcv@collabora.com>
Downstream distributors can use this to mark a version with their
preferred version information, like a Linux distribution package version
or the Steam revision it was built to be bundled into, or just to mark
it with the vendor it was built by or the environment it's intended to
be used in.
For instance, in Debian I'd use this by configuring with:
--enable-vendor-info="${DEB_VENDOR} ${DEB_VERSION}"
to get a SDL_REVISION like:
release-2.24.1-0-ga1d1946dc (Debian 2.24.1+dfsg-2)
which gives a Debian user enough information to track down the patches
and build-time configuration that were used for package revision 2.
In Autotools and CMake, this is a configure-time option like any other,
and will go into both SDL_REVISION (via SDL_revision.h) and
SDL_GetRevision().
In other build systems (MSVC, Xcode, etc.), defining the
SDL_VENDOR_INFO macro will get it into the output of SDL_GetRevision(),
although not SDL_REVISION.
Resolves: https://github.com/libsdl-org/SDL/issues/6418
Signed-off-by: Simon McVittie <smcv@collabora.com>
Instead of using a URL and git sha1, this uses `git describe` to
describe the version relative to the nearest previous git tag, which
gives a better indication of whether this is a release, a prerelease,
a slightly patched prerelease, or a long way after the last release
during active development.
This serves two purposes: it makes those APIs more informative, and it
also puts this information into the binary in a form that is easy to
screen-scrape using strings(1). For instance, if the bundled version of
SDL in a game has this, we can see at a glance what version it is.
It's also shorter than using the web address of the origin git
repository and the full git commit sha1.
Also write the computed version into a file ./VERSION in `make dist`
tarballs, so that when we build from a tarball on a system that doesn't
have git available, we still get the version details.
For the Perforce code path in showrev.sh, output the version number
followed by the Perforce revision, in a format reminiscent of
`git describe` (with p instead of g to indicate Perforce).
For the code path with no VCS available at all, put a suffix on the
version number to indicate that this is just a guess (we can't know
whether this SDL version is actually a git snapshot or has been
patched locally or similar).
Resolves: https://github.com/libsdl-org/SDL/issues/6418
Signed-off-by: Simon McVittie <smcv@collabora.com>
If this assertion fails on some platform (unlikely), we will need a
third implementation for SwapLongLE().
Signed-off-by: Simon McVittie <smcv@collabora.com>
find_lib() uses sort -V, which is a GNU extension. Users of non-GNU
operating systems should either install GNU coreutils (assumed to
provide a gsort executable), or use the CMake build system.
Resolves: https://github.com/libsdl-org/SDL/issues/6106
Signed-off-by: Simon McVittie <smcv@collabora.com>
If the width is sufficiently ludicrous, then the calculated pitch or
the image size could conceivably be a signed integer overflow, which
is undefined behaviour. Calculate in the unsigned size_t domain, with
overflow checks.
Signed-off-by: Simon McVittie <smcv@collabora.com>
When SDL is built with Wayland support on Linux, and Wayland libraries
are linked as dependencies instead of being loaded with dlopen(), its
dependencies will include libraries whose names contain a dash, like
`-lwayland-client` and `-ldecor-0`. Don't replace such libraries with
`-lwayland` and `-ldecor`: those don't exist and linking them will fail.
Signed-off-by: Simon McVittie <smcv@collabora.com>
On filesystems with large inode numbers, such as overlayfs, attempting
to stat() a file on a 32-bit system using legacy syscalls can fail
with EOVERFLOW. If we opt-in to more modern "large file support"
syscalls, then source code references to functions like stat() are
transparently replaced with ABIs that support large file sizes and
inode numbers, such as stat64().
This cannot safely be done globally by Linux distributions, because
some libraries expose types like `off_t` or `struct stat` in their
ABI, meaning that enabling large file support would be an incompatible
change that would cause crashes. However, SDL appears to be careful to
avoid these types in header files, so it should be OK to enable this.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Since #5602, SDL is intended to have the same ABI across the whole
major-version 2 cycle, so we should not check that the minor version
matches the one that was used to compile an application.
There are two checks that could make sense here.
The first check is that the major version matches the expected major
version. This is usually unnecessary and is not usually done (if we're
calling into the wrong library we'll likely crash anyway), but since we
have the information, we might as well continue to use it.
The second check is whether the version provided by the caller is
equal to or greater than a threshold version at which additional fields
were added to the struct. If it is, we should populate those fields;
if it is not, then we cannot. This is only useful on platforms where
additional fields have genuinely been added during the lifetime of
SDL 2, like Windows and DirectFB (but not X11).
This commit changes the first check to be consistent about only looking
at the minor version, while leaving the second check using SDL_VERSIONNUM
(which will be removed or widened in SDL 3, but it's fine for now).
Resolves: https://github.com/libsdl-org/SDL/issues/5711
Fixes: cd7c2f1 "Switch versioning scheme to be the same as GLib and Flatpak"
Signed-off-by: Simon McVittie <smcv@collabora.com>
I had assumed that only Linux users would be interested in GNOME-style
installed-tests, but in principle there's no reason why they can't be
used on non-Linux.
Signed-off-by: Simon McVittie <smcv@collabora.com>
See editorconfig.org for details of the format, which is understood by
multiple text editors, either directly or via plugins.
This is not comprehensive, but should cover most of the SDL codebase.
Please extend as needed.
Signed-off-by: Simon McVittie <smcv@collabora.com>
If the size to be allocated is very large and untrusted, then adding
the padding etc. might be enough to cause unsigned overflow, after
which a very small amount of memory will be allocated.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This can be used to check whether untrusted sizes would cause overflow
when used to calculate how much memory is needed.
Signed-off-by: Simon McVittie <smcv@collabora.com>
The official source code release isn't much use unless it contains
everything that users and downstream distributions need to do a
new build, so check that it does.
Signed-off-by: Simon McVittie <smcv@collabora.com>
The use of square brackets for a character set collides with the use
of square brackets for m4 quote characters, so use the other quoting
mechanism that Autoconf provides, by escaping `[` as `@<:@` and so on.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This is currently only done for the Linux Autotools build. The CMake
build does not add a significant amount of extra test coverage, and
Github Workflows run in an environment where `cmake` and `sudo cmake`
point to different executables, which makes it awkward to install into
/usr/local from CMake.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This makes it more convenient to compile them alongside SDL, install
them in an optional package and use them as smoke-tests or diagnostic
tools. The default installation directory is taken from GNOME's
installed-tests, which seems as good a convention as any other:
https://wiki.gnome.org/Initiatives/GnomeGoals/InstalledTests
Signed-off-by: Simon McVittie <smcv@collabora.com>
As well as reducing duplication, this lets the tests load their resources
from the SDL_GetBasePath() on platforms that support it, which is useful
if the tests are compiled along with the rest of SDL and installed below
/usr as manual tests, similar to GNOME's installed-tests convention.
Thanks to Ozkan Sezer for the OS/2 build glue.
Co-authored-by: Ozkan Sezer <sezeroz@gmail.com>
Signed-off-by: Simon McVittie <smcv@collabora.com>
In Autotools, these are run by `make -C ${builddir}/test check`.
In CMake, they're run by `make -C ${builddir} test` or
`ninja -C ${builddir} test` or `ctest --test-dir ${builddir}`.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Some CI workers don't seem to understand `cmake -v`, and Windows' shell
doesn't understand `VERBOSE=1 cmake`.
Signed-off-by: Simon McVittie <smcv@collabora.com>
If we're strict about applying something resembling semantic versioning
to the "marketing" version number, then we can mechanically generate
the ABI version from it.
This limits the range of valid micro versions (patchlevels) to 0-99.
Signed-off-by: Simon McVittie <smcv@collabora.com>
For stable releases, this gives us the ability to make bugfix-only point
releases such as 2.24.1 if we want to, and distinguish between them
programmatically. For example, this ability could have been useful after
2.0.16 to fix Xwayland regressions, and after 2.0.18 to fix event loop
regressions.
For development releases, this gives us the ability to make multiple
prereleases during the same feature cycle, and distinguish between them
programmatically. For example, this would have been useful during 2.0.22
development, which went through three prereleases before reaching the
final release.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Now that we've said this will be removed from SDL 3, we're free to use
any encoding that is compatible with existing SDL versions and will still
compare correctly for all SDL 2 version numbers. This allows the SDL 2
minor version to go beyond 1 digit, limited only by the size of
SDL_version.minor (which is 8 bits), making the largest possible version
number 2.255.99.
The patchlevel (micro version) is still limited to 2 digits.
Signed-off-by: Simon McVittie <smcv@collabora.com>
The encoding used in SDL_VERSIONNUM (e.g. 2.0.22 -> 2022) cannot
represent 2-digit minor versions without overflowing from the hundreds
digit into the thousands digit, which produces confusing version
numbers that will compare incorrectly when the major version is increased
to 3.
However, we can sidestep this problem by declaring that SDL_VERSIONNUM
will no longer be present in SDL 3, which means it only needs to be able
to represent SDL 2 version numbers losslessly.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This comparison normally happens at compile-time, not at runtime, so
it doesn't matter if it isn't optimal. This avoids incorrect comparison
if the minor version in SDL_COMPILEDVERSION and SDL_VERSIONNUM has more
than one digit, which would cause it to overflow from the hundreds place
into the thousands place.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This is usually desirable for batch processing: it lets us see exactly
what is happening in the logs.
Signed-off-by: Simon McVittie <smcv@collabora.com>
It looks as though something in the test subproject "leaks" into the
main build system, causing us to try to install ${builddir}/test/sdl2.pc
instead of the correct ${builddir}/sdl2.pc. Moving the tests subproject
further down avoids this.
Resolves: https://github.com/libsdl-org/SDL/issues/5604
Signed-off-by: Simon McVittie <smcv@collabora.com>
At the time I contributed this unit test, SDL had a relatively narrow
definition of what is a keyboard, approximately matching udev
ID_INPUT_KEYBOARD. Now it uses the equivalent of udev ID_INPUT_KEY,
which matches anything with keyboard keys, and not just reasonably
complete alphanumeric keyboards.
Fixes: 040bd7a9 "Fix udev not detecting ID_INPUT_KEY devices when udev is not running"
Signed-off-by: Simon McVittie <smcv@collabora.com>
At the time I contributed this unit test, SDL didn't understand Linux
touchpads, but now it does.
Fixes: 373216ae "Added support for touchpads in the Linux evdev code"
Signed-off-by: Simon McVittie <smcv@collabora.com>
When using shared linking (linking in the normal way with
-lwayland-client) rather than loading Wayland libraries dynamically at
runtime, listing symbols that don't exist in the current version results
in a build failure. We don't actually call wl_proxy_marshal_flags() or
wl_proxy_marshal_array_flags() directly; the reason we need them is
that they're called by the code generated by wayland-scanner >= 1.20.
If we're building against an older Wayland library, then we'll have its
corresponding version of wayland-scanner (mismatched versions are not
supported), so we won't need those two symbols, and can avoid generating
a dependency on them.
Conversely, if we're building against a newer Wayland library, the
generated code will call them unconditionally, so we cannot treat them as
optional and gracefully fall back: that would result in a crash. Instead,
treat them as a mandatory part of the Wayland library, so that if they
are not found at runtime, we can fall back to X11 without crashing.
libwayland 1.18 is in several LTS distributions (Ubuntu 20.04,
Debian 11, RHEL 8) so avoiding a hard dependency on 1.20 is quite
useful.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Resolves: https://github.com/libsdl-org/SDL/issues/5376
If we are already in the desired mode, changing it is a no-op at best,
and harmful at worst: on Xwayland, it sometimes happens that we disable
the crtc and cannot re-enable it.
Resolves: https://github.com/libsdl-org/SDL/issues/4630
Signed-off-by: Simon McVittie <smcv@collabora.com>
These are explicitly written in C code rather than generated at build
time, so they weren't affected by changing how we invoke
wayland-scanner.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This uses pre-recorded evdev capabilities, so that we can check for
regressions without the devices having to be physically present.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This improves SDL's ability to detect joystick hotplug in a container
environment.
We cannot reliably receive events from udev in a container, because they
are delivered as netlink events, which are authenticated by their uid
being 0. However, in a user namespace created by an unprivileged user
(for example bubblewrap, as used by Flatpak and Steam's
pressure-vessel-wrap), the kernel does not allow us to map uid 0, and
the netlink events appear to be from the kernel's overflowuid (typically
65534/nobody), meaning libudev cannot distinguish between genuine uevents
from udevd and an attack by a malicious local user.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Anything with X, Y and Z axes but no buttons is probably an
accelerometer (this is the assumption made in udev).
Signed-off-by: Simon McVittie <smcv@collabora.com>
Previously we only checked for at least one button or key and at least
the X and Y absolute axes, but this has both false positives and false
negatives.
Graphics tablets, trackpads and touchscreens all have buttons and
absolute X and Y axes, but we don't want to detect those as joysticks.
On normal Linux systems ordinary users do not have access to these
device nodes, but members of the 'input' group do.
Conversely, some game controllers only have digital buttons and no
analogue axes (the Nintendo Wiimote is an example), and some have axes
and no buttons (steering wheels or flight simulator rudders might not
have buttons).
Use the more elaborate heuristic factored out from SDL's udev code path
to handle these cases.
In an ideal world we could use exactly the same heuristic as udev's
input_id builtin, but that isn't under a suitable license for inclusion
in SDL, so we have to use a parallel implementation of something
vaguely similar.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This works on capability bitfields that can either come from udev or
from ioctls, so it is equally applicable to both udev and non-udev
input device detection.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Device enumeration via libudev can fail in a container for two reasons:
* the netlink protocol between udevd and libudev is considered private,
so there is no API guarantee that the version of libudev in a container
will understand netlink messages from a dissimilar version of udevd
on the host system;
* the netlink protocol between udevd and libudev relies for security on
being able to check the uid of each message, but in a container with
a user namespace where host uid 0 is not mapped, the libudev client
cannot distinguish between messages from host uid 0 and messages from
a different, malicious user on the host
To make this easier to experiment with, always compile the fallback
code path even if libudev is disabled. libudev remains the default if
enabled at compile time, but the fallback code path can be forced.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Simon McVittie