And build with CMake.

Fatih Uzunoğlu authored 9 months ago and Steve Lhomme committed 9 months ago

Currently, the C++ type `FlickableScrollHandler` conflicts with QML type
with the same name.

Thomas Guillem authored 11 months ago and Steve Lhomme committed 9 months ago

After waiting that the previous media is terminated
(vlc_sem_wait(&imem->done)), nothing can predict when the next media
will be opened, therefore, the read_blocking sem could be in any states
when it was checked with an assert. Ensure that each access use their
own semaphores to fix this issue.

Fixes #28633

Fatih Uzunoğlu authored 9 months ago and Steve Lhomme committed 9 months ago

`QVLCDialog::setWindowTransientParent()` first creates a window if it is not
already created.

Depending on the platform, a `QMenu` might use special attributes for its
underlying window. Some of these special attributes, such as, translucent
background requires them to be set prior to creation.

`QMenu` does such adjustments in polishing stage. Calling `ensurePolished()`
before calling `QVLCDialog::setWindowTransientParent()` satisfies that the
window is going to be what `QMenu` really wants.

Alexandre Janniaux authored 1 year ago and Steve Lhomme committed 9 months ago

The test currently checks that buffering is correctly handled, but it
will also check timing for startup, controls, subdecoders, and drain
behaviour in the future.

Alexandre Janniaux authored 9 months ago and Steve Lhomme committed 9 months ago

Check that we have at least the correct amount of buffering to start up,
instead of checking that we have more than the correct amount, which is
more natural to see than having "buffering: 100%" but the stream stuck
at the given level of buffering.

Alexandre Janniaux authored 9 months ago and Steve Lhomme committed 9 months ago

vlc_clock_t was used in the file but not declared before.

Alexandre Janniaux authored 9 months ago and Steve Lhomme committed 9 months ago

std::as_const is available starting with C++17[^1]. qAsConst is now
deprecated:

    ../../../../modules/gui/qt/medialibrary/medialib.cpp: In lambda function:
    ../../../../modules/gui/qt/medialibrary/medialib.cpp:357:39: warning: ‘constexpr typename std::add_const<_Tp>::type& qAsConst(T&) [with T = QList<QJSValue>; typename std::add_const<_Tp>::type = const QList<QJSValue>]’ is deprecated: Use std::as_const() instead. [-Wdeprecated-declarations]
      357 |         for (const auto& cb : qAsConst(it.value()))
          |                               ~~~~~~~~^~~~~~~~~~~~

[^1]: https://en.cppreference.com/w/cpp/utility/as_const

Right now it's built as an iOS target when it shouldn't. It's a tier 3 target [1]

[1] https://doc.rust-lang.org/stable/rustc/platform-support/apple-visionos.html

It's the same define used in configure.ac.

prebuilt contains the cross-compiled tools.

This should never happen. Only the HOST folder is kept in the archive.
And it can only contain win32/win64 binaries.

See the "package" target.

It was originally kept in ccee6f64.
But we should not use binaries from there as most of the time they
can't run locally.

native tools we may use are in BUILDPREFIX/bin (or BUILDBINDIR) which
is always in the path.

This reverts commit 7292c978.

Most of the time we can't run cross-compiled binaries on the host machine.
We do not use the -config programs as it may have been the case in the past.

We have native binaries in the path since fe47c2f2.

So we don't have to hardcode the PATH.

When using a static Rust libstd the linker can pick the runtime it wants.
It will correspond to the same runtime we use for the rest of the code.

See !5675 (comment 447873)

gnullvm targets most importantly use LLVM libraries and ABI. So libgcc is replaced
with libunwind, the linker is LLD, it uses native TLS instead of emutls, function
sections are enabled, and there's probably something else that I forgot. UCRT
is another difference, but a smaller one since the changes from my previous
sentence cause builds with GCC/Binutils to break.

windows-gnu targets are built with MSVCRT, but that shouldn't be a big deal as
long as you stick to the default static std library (in other words, as long as
you don't use -C prefer-dynamic). I can't guarantee it will always work, but
t's much better compared to using windows-gnullvm targets, which have a 100%
chance of failing when using them with GCC/Binutils. Unless you want to migrate
to LLVM as well (windows-gnullvm targets by default will pick Clang from PATH
as the linker), but that would be a much bigger change than this PR implies.

And only windows-gnu when not compiled with LLVM.

It's stricly equivalent and avoids a negative test.

The PATH set in HOSTVARS_CMAKE is not kept when calling CMAKEBUILD.
We need that path to have access to the "cross compiled" fxc2.

We need to write in binary mode all the time, or at least in binary output mode.

Pierre Lamot authored 9 months ago and Steve Lhomme committed 9 months ago

qml imports policy have changed recently, some MR have been rebased and merged
without updating the imports

Claudio Cambra authored 9 months ago and Steve Lhomme committed 9 months ago

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>

Claudio Cambra authored 9 months ago and Steve Lhomme committed 9 months ago

macosx: Return VLCMediaLibraryParentGroupType directly from rowToVideoGroup rather than an NSInteger

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>

Claudio Cambra authored 9 months ago and Steve Lhomme committed 9 months ago

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>

Martin Storsjö authored 9 months ago and Steve Lhomme committed 9 months ago

_UCRT is not a compiler internal define. Some build scripts may be
invoking the compiler with an explicit -D_UCRT, but if not, the
toolchain default is set by the toolchain headers. Therefore, include
crtdefs.h to pull in default headers for checking whether it is set
or not.

Also check for older forms of mingw-w64, which didn't use the
define _UCRT to signal the type of CRT.

Claudio Cambra authored 9 months ago and Steve Lhomme committed 9 months ago

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>

Claudio Cambra authored 9 months ago and Steve Lhomme committed 9 months ago

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>

Claudio Cambra authored 9 months ago and Steve Lhomme committed 9 months ago

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>

Claudio Cambra authored 9 months ago and Steve Lhomme committed 9 months ago

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>

Fatih Uzunoğlu authored 9 months ago and Steve Lhomme committed 9 months ago

These are not exposed to QML at the moment.

Vikram Kangotra authored 9 months ago and Steve Lhomme committed 9 months ago

Ensure mutex is unlocked if extension index exceeds array
size to prevent deadlock.

This is necessary to work on machines with Python 3.12.
libplacebo fails to detect python with meson 1.2.0.

1.2.3 is the first to support Python 3.12 [1]

We download 1.4.2 since 3f852923.

[1] https://mesonbuild.com/meson-python/reference/meson-compatibility.html

We will never have the same tarball on all platforms using
different versions of tar, different order of files, different owner, etc.

Since we generated the tarball we don't have to verify
the checksum, we can assume it's OK, assuming cargo vendor is safe.

If we ever delete the tarball, the marker is deleted so we check
the download properly from the download if possible.

Just like the regular tarball checksum.

The vendor tarball checksum is located in a <package>/vendor-SHA512SUMS
and the rule to use is called .sum-vendor-<package>.