Unlock the full feature set of the YouTube Chat dock in OBS by removing
custom scripting/CSS logic. Enable the signed-in experience for live
streaming content creators while also sharing login credentials with the
YouTube Control panel dock.
This will allow OBS users to utilize features _already_ supported in the
YouTube Chat plugin, such as
* creating polls
* managing Q&A sessions
* a rich emoji set in the input panel
* emoji fountains
* moderation tools
and many more. These features are available to users who are logged-in
to YouTube Chat and/or the YouTube Control panel.
Use the cookie manager shared by service integration browser docks for
YouTube Control Panel.
This will enable users of the YouTube Chat panel have a better (creator
facing) experience for observed chat message latency, for those users
who sign-in to the YouTube Control Panel or YouTube Live Chat.
NOTE: This commit does not have any migration logic for existing
logged-in users of YTCP, they will need to sign-in again. Based on usage
stats, this is not going to affect a large fraction of OBS users.
BPM initialization occurs with the first call to the
`bpm_inject()` callback function. When Stream Delay is
active, there is a case where the first call might never
happen, specifically if a user stops the stream with the
discard delay option before streaming begins. In such a
case OBS will crash due to an uninitialized mutex being
referenced in `bpm_destroy()`.
Use `pthread_once()` in both the `bpm_inject()` callback
and `bpm_destroy()` to ensure BPM initialization has occurred.
Per the documentation of CFStringGetCString, the buffer provided must be
large enough not just for the string itself, but also for a NUL
terminator. This space for the NUL terminator is currently ignored, and
we just get lucky that CFStringGetMaximumSizeForEncoding often
dramatically overestimates the space required. However, it is possible
to actually hit the maximum with the string itself (for example by using
strings that contain exclusively Chinese characters such as "我"), in
which case the conversion fails. Adding the extra byte for the NUL
terminator fixes this.
At this point, we can also safely assert that our max_size is larger
than zero, silencing a clang analyzer warning that now is no longer
valid.
obs_scene_sceneitem_from_source is problematic because there can be
multiple items of the same source in a scene, which the function doesn't
account for. In such a case, it would return the first item it finds,
which often might not be what a developer expects.
It was originally added for the undo/redo-operation of "Add New Source"
where the UI guarantees that the item is unique, but for a general case
it's not suitable.
Instead of first creating the scene item and then finding it again using
a method that's generally problematic (and only works here because we
know that there is only one instance of a source), let's just remember
the scene item in AddNew after we create it, similar to the source
itself.
Scene collection names that are not considered "safe" by OBS Studio
get a changed JSON file name with incompatible characters replaced.
The refactored scene collection implementation uses the Load function to
either activate an existing scene collection or create a new one if it
does not exist.
The Load function however overwrote the scene collection name set in
the profile with its own variant based off the "safe" file name, which
created a mismatch with the code that created the collection data
model.
As the Load function is only called by ActivateSceneCollection (which
itself already sets the name and filename for the collection), removal
of this superfluous code in the Load function also fixes the issue.
The variable `input.conversion.allow_clipping` was left uninitialized.
This could result in randomly sending unclamped audio to the output
handlers that did not request the conversion.
Windows and Linux allow the storage of configuration files relative
to the binary location, which is enabled by default on Windows and has
to be explicitly enabled on Linux.
This was originally conflated with the LINUX_PORTABLE build setting
which also allowed the application itself to be run from non-default
locations on a Linux system.
This change reintroduces the functionality behind the
ENABLE_PORTABLE_CONFIG build setting on Linux.
It also adds necessary code to make this setting compatible with the
recently introduced relocatable settings code changes:
When portable mode is enabled, user configuration, scene collections,
and profiles are stored in the config directory created for portable
mode.
To ensure that the find module prefers the variant of LuaJIT shipped
as part of obs-deps, it's necessary to give the name as shipped in
obs-deps the highest precedence.
This ensures that the "find_library" call will discover the less
version-specific variant of the library possibly installed via
Homebrew.
macOS builds should only use dependencies built via the obs-deps
build scripts. Default variants of the same dependencies are not
compatible with our app packaging requirements and thus will create
issues when creating the app bundle.
This is specifically an issue when MbedTLS is installed via Homebrew
which ships a CMake package config by default and is picked up
by our code ever since we switched to prefer CMake packages.
`std::clamp` was introduced at 60a45d3aa3 but it caused a runtime
assertion failure on Windows that checks the low value of `std::clamp`
is not greater than the high value.
The "provisioning" and "notarization" steps that set the
"haveProvisioningProfile" and "haveNotarizationUser" outputs
respectively only run if "haveCodesignIdent" has evaluated to true in
the prior "codesign" step.
This means that if "haveCodesignIdent" is false, the other two outputs
are left unset, evaluating their "value" expressions (and as such the
output of the action) to empty instead of false.
This issue was found in the equivalent action in obs-plugintemplate. As
that action is basically a mirror from the obs-studio action, let's fix
it here too.
Once upon a time there was the idea to enable use of Ccache for
Windows builds on CI as well, but to achieve this the project would
need to switch to clang-cl.exe as its compiler and use an additional
post-build step to generate PDBs from binaries.
Using a different compiler for continuous integration than for building
release builds defeats the purpose of CI however, so the idea was
dropped.
Makes sure that a scene item that gets removed is in the "not selected"
state, no matter whether it was selected before. This causes the
"item_deselect" signal to be sent if the item is selected while being
removed.
MbedTLS changed a lot of internals with their LTS version 3.6.0, which
are incompatible with the find module currently shipped with
OBS Studio.
The solution requires several changes to be applied at once:
* Rename the generated target name to MbedTLS::mbedtls to match the
name used by MbedTLS' own CMake package
* Update find module to use the updated target name(s)
* Set CMAKE_FIND_PACKAGE_PREFER_CONFIG to TRUE before trying to find
MbedTLS to ensure that CMake package files are used with priority
(Those are shipped only with MbedTLS 3.6.0 in obs-deps).
* A deprecation warning is emitted if the find module is used with
MbedTLS 3.6.0 available
We're currently using peter-evans/create-pull-request v6.0.0 which has a
known issue that affects our services-validator action when attempting
to update an existing PR. This issue was fixed in v6.0.1, but we can
update to v7.0.2.
We check that mids is a nullptr which may not be the case if this
remains uninitialized. Earlier attempt to fix this assumed we faulted on
the response itself.
fixes#11221
