We require libdrm for its header so add the cmake module and header path
to the build. We don't need to link libdrm though so we dont add it to
libraries.
DRM_FORMAT_MOD_INVALID is a token for implicit modifier. When importing
a DMA-BUF with implicit modifier we should call
gs_texture_create_from_dmabuf() without modifier.
g_variant_lookup() obligatorily receives the type of the variant to
lookup. This function is used when retrieving the session handle
from the portal's response, and the variant type passed is "s" (a
string).
However, xdg-desktop-portal had a bug: the documentation explicitly
mentions that the session handle is an object path (of variant type
"o"), but it passed a string (of variant type "s"). This mismatch
was fixed in the xdg-desktop-portal release 1.10 [1], but that broke
the PipeWire capture code, which was passing specifically the "s"
value to the variant lookup.
Fix this by not checking the variant type at all. Object paths ("o")
are simply strings with a few extra checks, and we don't actually need
to perform these checks.
This change probably broke other apps, and that makes me extremely sad :(
[1] https://github.com/flatpak/xdg-desktop-portal/pull/609
The gs_texture_create_from_dmabuf function is able to create textures
from multiplanar DMA-BUF data. However, the only in-tree consumer of
this function - the PipeWire capture - does not use that.
Make PipeWire create textures from multiplanar buffers.
Unlike Xcomposite, the XSHM plugin does not use GLX code, and thus
can be used on when EGL renderer is used. It still is X11-specific
though, and shouldn't be used on Wayland.
Rework the obs_module_load() function of linux-capture to use a
switch statement, and load XSHM both on EGL/X11 and GLX/X11.
Fixes https://github.com/obsproject/obs-studio/issues/5122
The gs_draw_sprite_subregion() function is used when a cropping
rectangle is received from PipeWire. It is usually used by
compositors to implement window screencast - where a large and
mostly empty frame is sent, the window contents are only a small
part of it, and the crop rectangle tells us that.
Recently the wlroots implementation of portals started to use it
to implement cropping, and it exposed a bug in the PipeWire code
in OBS Studio. The gs_draw_sprite_subregion() function takes a pair
of integers representing position (x, y) and a pair of integers
representing size (width, height). The PipeWire code, however,
passes a second pair of positions (x2, y2) instead of sizes, and
it causes overrendering the crop area.
This bug wasn't hit yet because both GNOME and KDE implementations
always send (0, 0) as position, which practically never trigger
this condition.
Pass only width and height to gs_draw_sprite_subregion(), instead
of adding x and y to them.
Fixes https://github.com/obsproject/obs-studio/issues/4982
Intead of creating one pair of GDBusConnection + GDBusProxy objects
for each PipeWire capture, be it window or desktop, use the global
ones managed by portal.c.
Even if g_bus_get_sync() ends up reusing the same object under the
hood, it's still a net gain, since it has to perform some thread
synchronization routines that aren't necessary here. Creating the
proxy object was a worse offender, because despite being asynchronous,
it would still incur in a few socket messages + a cancellable fd +
thread synchronization.
Reuse these objects from portal.c. The biggest code change here is
that create_proxy() and on_proxy_created_cb() were merged into
init_obs_pipewire().
Right now we just assume that every compositor and portal implementation
exposes both window and monitor captures, but that's not true, and in fact
the Desktop portal provides a simple mechanism to check which source types
are available: a D-Bus property called "AvailableSourceTypes".
Read this D-Bus property, and use it to conditionally register the desktop
and the window captures.
Related: https://github.com/obsproject/obs-studio/issues/4815
The cursor bitmap is centered on the hotspot, so not accounting
for it means PipeWire captures were positioning the cursor sprite
slightly off.
Properly account for the hotspot by subtracting it from the cursor
position.
Related: https://github.com/obsproject/obs-studio/issues/4766
The assertion exposed a bug in the KDE implementation of the Desktop
portal. However, it comes with the side effect of exiting OBS Studio
in this case, where we can actually make it work (even if buggy).
De-escalate the assertion to a warning, and then attempt to find the
correct stream to use.
Add a new Linux capture based on PipeWire [1] and the Desktop portal [2].
This new capture starts by asking the Desktop portal for a screencapture session.
There are quite a few D-Bus calls involved in this, but the key points are:
1. A connection to org.freedesktop.portal.ScreenCast is estabilished, and the
available cursor modes are updated.
2. CreateSession() is called. This is the first step of the negotiation.
3. SelectSources() is called. This is when a system dialog pops up asking the
user to either select a monitor (desktop capture) or a window (window capture).
4. Start() is called. This signals the compositor that it can setup a PipeWire
stream, and start sending buffers.
The reply to this fourth call gives OBS Studio the PipeWire fd, and the id of the
PipeWire node where the buffers are being sent to. This allows creating a consumer
PipeWire stream, and receive the buffers.
Metadata cursor is always preferred, but on the lack of it, we ask the stream for
an embedded cursor (i.e. the cursor is drawn at the buffer, and OBS Studio has no
control over it.)
Window capturing is implemented as a crop operation on the buffer. Compositors
can send big buffers, and a crop rectangle, and this is used to paint a subregion
of the buffer in the scene.
The new capture is only loaded when running on EGL, since it depends on EGL to
call gs_texture_create_from_dmabuf().
[1] https://pipewire.org/
[2] https://github.com/flatpak/xdg-desktop-portal/
In preparation for the introduction of the new PipeWire-based capture,
use variables for include_directories() and target_link_libraries(),
and move them to the bottom of the file.
xcompcap was taking locks in the wrong order resulting in deadlocks on
close. If you had an xcompcap properties window open on close it was
nearly 100% deadlock. This ensures locks are taken in the same order as
libobs takes when shutting down.
Track all windows corresponding to sources and ensure that we only
disable XSelectInput events once all sources for a given window have
been removed. Previously we may have stopped listening for events if
multiple sources captured the same window and one was removed.
We also move window redirection into the helper to avoid similar issues.
Previously we only captured by window name and class. This prevented
capture of windows with the same name and class, and caused captures to
switch from one window to another of the same name and class.
Well, linear SRGB for screen capture. The window capture path failed to
copy between SRGB textures for some reason, so just force nonlinear
formats instead.
RandR has two sets of screen geometry information:
1. CRTC. These are the physical scanout engines in the hardware
2. Monitors. These are the logical partitions of the screen.
By default, each CRTC gets mapped to a Monitor. However, some monitors
actually require two CRTCs to drive them due to limitations in the
scanout hardware. Users can also create 'virtual' monitors to support
VNC or other systems.
This patch makes the RandR code prefer the Monitor mechanism to the
older CRTC mechanism. If the server doesn't support a new enough RandR
version, the existing CRTC code is used instead.
The name of the monitor is also provided in place of the arbitrary
number to help users select the desired source.
Signed-off-by: Keith Packard <keithp@keithp.com>
When cropping the left or top of a window capture, OBS would misalign
the drawn cursor, placing it as if the entire window were being
captured. Instead, offset the captured cursor by the same number of
pixels, thus placing the cursor back where it belongs.
Move xlock after initial window listing
- XErrorLock uses static fields so when you use them recursively
they collide and do weird things. This keeps the two xlocks from
colliding.
Build obs textures based on returned pixmap texture types to ensure
compatibility during copies. (prevents glCopyImageSubData failed due to
texture incompatibilities)
Fix alpha check by using pixmap depth instead of checking channel sizes
(wrong on nvidia and maybe amd hardware)
Fix X error checking, now you will get 1 BadAlloc when pixmap allocation
fails instead "glCopyImageSubData failed" every frame. (occurs on Gnome
under special circumstances)
Code submissions have continually suffered from formatting
inconsistencies that constantly have to be addressed. Using
clang-format simplifies this by making code formatting more consistent,
and allows automation of the code formatting so that maintainers can
focus more on the code itself instead of code formatting.
