Currently, Qt's User Interface Compiler (UIC) emits a warning about the
tabstop assignment for horizontalLayout_sdrPaperWhite, which is a
QLayout rather than a QWidget. Remove the tabstop assignment to fix the
warning.
Notable changes:
* Add Jansson to Windows deps
* Update zlib to 1.2.13
* Update libpng to 1.6.38
* Update opus to bce1f39235
* Update SVT-AV1 to 1.3.0
* Update aom to 3.5.0
* Update x264 to r3101
* Update libsrt to 1.5.1
* Update AMF to 1.4.26
* Update FFmpeg to 5.1.2
* Update LuaJIT to 2.1 7a0cf5fd4c
* Windows: Update cmocka to cd95ee9f77
* Windows: Update curl to 7.86.0
* Qt: Update Qt5 to Qt 5.15.7 for Windows
* Qt: Update Qt6 to Qt 6.4.1 for Windows
* Qt: Update Qt5 to Qt 5.15.7 for macOS
* Qt: Update Qt6 to Qt 6.4.1 for macOS
In updated versions of FFmpeg (5.1.x) and SVT-AV1 (1.2.0/1.3.0), the
rate contol mode selection mechanism has changed, causing CBR to throw
an "Invalid Parameter" error and the encoder will fail to start. This
appears to be caused by us specifying rc_max_rate, which will also fail
on FFmpeg CLI. Adjust how we manage rate control methods, appropriately
version-gated, to work with both our current dependency versions and the
versions that we'd like to use.
To avoid the user having lookahead enabled with no way to disable it,
change the Simple Output Mode NVENC Preset from P6 to P5.
Follow up to 8806738a04.
In dfccb26bb2, in my haste to fix Windows
CI, I accidentally included a change to disable Windows scripting which
I had been using locally. Reverse that change to re-enable scripting in
Windows builds.
The backtick double-quote pattern does not work in PowerShell 7.3.x.
There are still some other possible PowerShell 7.3.x issues in the
packaging steps, but let's fix this first to get regular builds working
in PowerShell 7.3.x again.
Erroring out of NVENC init early if the Max B-frames setting was higher
than the encoder's capability causes an encoder failure on NVIDIA Pascal
(10-series) and earlier GPUs due to an unfortunate interaction between
Simple Output Mode, HEVC, and our default B-frames setting of 2. Since
we already check the Max B-frames capability of the encoder, cap at that
value instead of erroring out.
Fixes#7698.
The old NVENC preset "Default", labeled in OBS as "Performance", maps to
the new P3 preset when not using multipass. Since we previously only
enabled multipass/two-pass on "Max Quality", we can map this to P3 in
all cases.
This commit also modifies UI.
The old preset "mq" was "Max Quality", which used the old HQ preset with
two-pass enabled. The old preset "hq" was "Quality", which used the old
HQ preset without two-pass. Of the two, "mq" would have been considered
"slower" or "higher quality" than "hq", so let's swap these entries.
The AMF docs specify that the maximum value for TARGET_BITRATE and
PEAK_BITRATE is 100,000,000 bit/s. Trying to set values above this
maximum value results in AMF choosing from its defaults instead of
capping at the maximum value. Let's cap at 100,000 Kbps in the UI to
prevent users from running in to this.
Fixes GitHub Issue obsproject/obs-studio#7423.
OBS Studio 28 no longer supports Windows 7, 8, or 8.1, and Qt 6 will
fail to run on those versions. Users on those systems being offered the
update will end up in a state where OBS will not run if they install the
update. Let's prevent the updater executable from running on those
versions to avoid that scenario.
The x() and y() values of coordinates for events inside a scrollable
QWidget are relative to viewport's scrolled origin, the coordinates of
the upper left corner of the visible space, not the widget's true
origin. Since we do not allow horizontal scrolling, the value of x() is
okay. However, the value of y() needs to be adjusted by an offset of the
top()/y() value for the first widget in the SceneTree. When not
scrolled, this offset will be 0. When scrolled down, this offset will be
a negative value.
The UI layout math used to determine if scrollbars should be displayed
in the Scenes Dock was off by one pixel. This caused the scrollbar to
disappear when performing actions while scrolled to the bottom, such as
dragging items or resizing the dock.
Calling resizeEvent for SceneTree after a dropEvent has occurred
prevents a dropped item from being displayed in the incorrect location
while in Grid Mode. There might be a better way to fix the incorrectly
displayed location of a dropped item, but we already do this in
SetGridMode and rowsInserted, so this is probably okay.
On CI, do not fetch tags in packaging scripts. For some reason, the
checkout action seems to locally update any new git tags on the runner:
t [tag update] (commit-hash) -> tag-name
This causes future calls to fetch git tags to fail on CI with:
! [rejected] tag-name -> tag-name (would clobber existing tag)
To avoid this, we can simply not fetch tags a second time on CI.
Additionally, fix the Windows Installer job.
In .github/workflows/main.yml, for the linux_build job, the variable
BUILD_FOR_DISTRIBUTION is set to the string "true" or "false" on CI.
Later, in CI/linux/03_package_obs.sh, we perform a boolean check on
this variable. However, "false" will evaluate as true, because it is a
non-null string. This was causing CI Linux packages to always build as
if BUILD_FOR_DISTRIBUTION was enabled, which caused the git commit hash
to be omitted from package filenames.
Since we know the expected values, let's just test directly if the
variable equals "true" to get the expected behavior.
Notable changes:
* Use VS2022
* Update zlib to 1.2.12
* Update opus to ab04fbb1b7
* Update libvpx to 1.12.0
* Update SVT-AV1 to 1.1.0
* Update aom to 3.4.0
* Update x264 to r3095
* Update mbedTLS to 3.2.1
* Update libsrt to 1.5.0
* Update librist to 809390b3b7
* Update AMF to 1.4.24
* Update FFmpeg to 5.0.1
* Update FreeType to 2.12.1
* Update LuaJIT to 2.1 3065c910ad
* Update SpeexDSP to 1.2.1
* macOS: Update Jansson to 2.14
* Windows: Update libogg to 1.3.5
* Windows: Update libvorbis to 84c023699c
* Windows: Update cmocka to bbf5ff7dd5
* Windows: Update curl to 7.84.0
* Windows: Update VulkanSDK to 1.3.216.0
* Windows: Update nasm to 2.15.05
* Windows: Update Python to 3.8.10
* Qt: Update Qt5 to Qt 5.15.5 for Windows
* Qt: Update Qt6 to Qt 6.3.1 for Windows
* Qt: Update Qt5 to Qt 5.15.5 for macOS
* Qt: Update Qt6 to Qt 6.3.1 on macOS
* Qt: Backport Windows native scale factor fix
Switch to Visual Studio 2022 and clang-format 13.
The Ubuntu 22.04 image currently has clang-format 12.0.1, 13.0.1, and
14.0.0. VS2022 17.2 ships clang-format 13.0.1, so let's use that for now
until VS2022 17.3 is available with clang-format 14.
This commit also makes the necessary changes to allow the clang-format
check to pass.
This commit also updates the obs-browser submodules with the necessary
changes.
If the path to clang-format has spaces, as might occur when testing
locally with a version not covered in this script, it will fail. Enclose
the call to clang-format in quotes to prevent this failure.
When trying to update to mbedTLS 3.2.1, I ran into failures with RTMPS
output. After consulting an mbedTLS implementation example, I determined
that we were not setting up the SSL/TLS context config in the correct
order, causing the connection to fail. Performing the setup in the
recommended order fixes that.
Notable changes:
* Windows: Add Qt 5 and 6 builds
* Windows: Remove nasm native build
* Windows: Update srt from 1.4.2 to 1.4.4
* Windows: Update SWIG from 3.0.12 to 4.1.0-git
* macOS: Update libvpx from 1.10.0 to 1.11.0
* macOS: Update srt from 1.4.1 to 1.4.4
* macOS: Update SWIG from 3.0.12 to 4.1.0-git
* macOS: Update x264 from r3059 to r3060
Removing the first call to `video_output_stop` prevents the `data_mutex`
field in `struct video_output` from being destroyed while still in use.
The `render_preview_source` function will call `video_output_lock_frame`
upon that mutex and encounter a NULL pointer exception.
Co-authored-by: Paul Hindt <paulh@aja.com>
When OBS has been connected to a YouTube account, the Manage Broadcast
dialog becomes available. Opening and closing this dialog leaks about
2-3 MB of memory each time. This happens because the dialog is never
deleted, and a new object is created every time the dialog opens.
If we set Qt::WA_DeleteOnClose on the dialog, then the dialog->exec()
call would call delete this. However, if the dialog->Valid() call
returned false (this should currently be impossible), then exec is never
called, so it would not clean up.
To fix this, make the dialog stack-allocated instead of using new and
delete.
Detected by clang-analyzer.
The Windows build script would respect the user-specified build
directory for the configure step, but not for the build step. Fix that
by applying the same logic in Configure-OBS to Build-OBS.
Ryan Foster
