(cherry picked from commit c4ad4065)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

See the API https://docs.microsoft.com/en-us/windows/win32/api/mmdeviceapi/nf-mmdeviceapi-activateaudiointerfaceasync



It requires a recent mingw-w64 with the added API.

We request the IAudioClient asynchronously and return the found client or NULL
once the async call as completed.
The code originates from the vlc-winrt project with some modifications.

Do not rely anymore on the local "winstore-client" variable to cache the
IAudioClient. A client is queried/used between each Start/Stop calls.

(cherry picked from commit 32349b35) (edited)

edited:
- the structure in this branch already exists as a a typedef

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

And some more cleaning

(cherry picked from commit d0b5eb1c) (rebased)

rebased:
-

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Fixed since c7644611

(cherry picked from commit 3cdbeb2c) (edited)

edited:
- in this branch the -O2 forcing was only done for iOS

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

No need to force fvisibility=hidden explicitly anymore.

Log why the -O2 is needed (this still feels like an odd fix)

(cherry picked from commit 41a66587) (edited)

edited:
- the -O2 patch is only applied for iOS on this branch

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Introduced in 2012 for x86 Android builds. It should be enabled directly by
CMake when building on modern systems.

libmatroska also throws some exception and was always compiled without this flag.

(cherry picked from commit 107a97ce)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

No need to force fvisibility=hidden explicitly anymore.

(cherry picked from commit 615d63eb)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

We build all contribs as static libraries.

The setting exists since CMake 3.0. For targets built with older CMake settings
the value is discarded.

(cherry picked from commit 9f518873) (rebased)

rebased:
- mysofa uses pthreadGC2 in this branch

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

For other targets it's not needed and creates a warning.

(cherry picked from commit 542d7a79) (rebased)

rebased:
- the MSYS detection is different in this branch

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

It won't work with Winstore builds and we don't use it anyway.

(cherry picked from commit 852d9576)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

(cherry picked from commit 98bdfe4b) (rebased)

rebased:
- this branch didn't have other patches applied

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

And remove upstreamed patches.

(cherry picked from commit 877d8301)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

It's used by newer gnutls. But gives the following error:

error: 'memset_s' is only available on macOS 10.9 or newer [-Werror,-Wunguarded-availability]
   (void) memset_s (s, len, '\0', len);
          ^~~~~~~~
 /Applications/Xcode9.2.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.13.sdk/usr/include/string.h:145:9: note: 'memset_s' has been marked as being introduced in macOS 10.9 here, but the deployment target is macOS 10.7.0

Fixes this compilation error:
system/fastopen.c:134:9: error: 'connectx' is only available on macOS 10.11 or newer [-Werror,-Wunguarded-availability]
                ret = connectx(fd, &endpoints, SAE_ASSOCID_ANY, CONNECT_RESUME_ON_READ_WRITE | CONNECT_DATA_IDEMPOTENT, NULL, 0, NULL, NULL);
                      ^~~~~~~~
/Applications/Xcode9.2.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.13.sdk/usr/include/sys/socket.h:713:5:
note: 'connectx' has been marked as being introduced in macOS 10.11 here, but the deployment target is macOS 10.7.0

In this patch __builtin_available() is assumed to be avalaible in the toolchain
which is the case in our 3.0 and 4.0 toolchains. A cleaner patch should detect
it in configure.ac. But we can't do autoconf changes in gnutls as it never
works properly.

The OS values match exactly the values found in
https://opensource.apple.com/source/xnu/xnu-4570.41.2/bsd/sys/socket.h



(cherry picked from commit c4a23f4c)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Backport a UNICODE fix that was merged after the GnuTLS code freeze.

Forbid call to GetHandleInformation(). fcntl never goes in that code section in
Winstore builds. It's only used for linux random number generation.

Do not allow calling SecureZeroMemory which is forbidden which is forbidden
and not available as an inline function in some cases.

(cherry picked from commit a4841494)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

ecc_scalar_random is public in the nettle they embed, but the symbol is not
mangled in the original nettle package, so it's not mangled here either. It
leads to the nettle symbol behind defined twice.

The bug is actually in nettle, not the script that imports nettle in gnutls.

(cherry picked from commit 1685914d)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

We only want the static library

(cherry picked from commit 2ddaceb6)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

So that patches changing the same file still apply regardless of this change.
And we do it after all patches are done.

(cherry picked from commit f442ecf5)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Steve Lhomme authored 8 years ago and Steve Lhomme committed 4 years ago

It's more flexible after updates.

(cherry picked from commit 3e69d914)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

(cherry picked from commit 799db8c8)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Marvin Scholz authored 5 years ago and Steve Lhomme committed 4 years ago

(cherry picked from commit 78df0f10)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Martin Storsjö authored 5 years ago and Steve Lhomme committed 4 years ago

The aarch64 assembly in gnutls still needs to be disabled on windows
in general though; the issue is that it unconditionally uses ELF
specific directives (.type, .size, .section), so the disabling isn't
related to any specific issue in clang, but only that gnutls' assembly
for this architecture isn't portable.

Add a comment to clarify the issue.

(cherry picked from commit 38eda60a) (rebased)

rebased:
- this branch doesn't have the NACL fixes (0d4d18e7)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Martin Storsjö authored 6 years ago and Steve Lhomme committed 4 years ago

This was a stealth release to fix packaging issues (relevant for
building on windows) in the 3.6.7 tarball, see
https://lists.gnupg.org/pipermail/gnutls-help/2019-April/004506.html

.

Remove the gnutls-winstore-ntop patch which conflicts with how
upstream now handles the inet_pton function.

(cherry picked from commit 49b4aa93)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Rémi Denis-Courmont authored 6 years ago and Steve Lhomme committed 4 years ago

(cherry picked from commit 0ec35e2d)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

(cherry picked from commit a252d4d3)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

(cherry picked from commit 86272c44)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

(cherry picked from commit 809ccfa5)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

David authored 6 years ago and Steve Lhomme committed 4 years ago

connectx is available since macOS 10.11 and iOS 9, so this is not
needed anymore.

Note, this cannot be backported to 3.0 branch.

(cherry picked from commit dae2fff6) (edited)

edited:
- older gnutls patches were backported before

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

David authored 6 years ago and Steve Lhomme committed 4 years ago

getentropy is correctly annotated since the macOS 10.13 SDK (at
least), and the gnutls check includes the correct header on macOS.
Therefore this patch is not needed anymore.

This also fixes current build failure as it avoids touching
configure.ac now.

(cherry picked from commit 58d4ea7d)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Hugo Beauzée-Luyssen authored 6 years ago and Steve Lhomme committed 4 years ago

SIZE_MAX is in stdint.h, which is already included

(cherry picked from commit 0878789f)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

That's what the buildbot does and it's not compatible with idn2 in mingw64
which requires detection with pkg-config

(cherry picked from commit 8c5044be) (rebased)

rebased:
- disabling acceleration for x64 was merged differently

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Hugo Beauzée-Luyssen authored 6 years ago and Steve Lhomme committed 4 years ago

And drop upstreamed or now unrequired patches

(cherry picked from commit 636aa214) (edited)

edited:
- remove vasnprintf-android-no-percent-n.patch which is merged in 3.6.6

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

Hugo Beauzée-Luyssen authored 6 years ago and Steve Lhomme committed 4 years ago

(cherry picked from commit f754a71f)
Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

(cherry picked from commit 9c7e8fd6)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

After a flush, i_first_render_host_time is reset to 0 and i_render_host_time
should not be touched since the playback has not started again yet. This caused
the i_first_render_host_time to be never setup.

Regression from f507868d
Fixes #24876

(cherry picked from commit 1530679a)
Signed-off-by: Thomas Guillem <thomas@gllm.fr>

(cherry picked from commit 52ff049d)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Alexandre Janniaux authored 4 years ago and Steve Lhomme committed 4 years ago

std::lower_bound doesn't return an iterator to the std::end value if the
element is not found, but like its name says, to the first value that is
greater or equal than the value wanted, meaning it could be the next one
if no processor has been bound to the ebml id sent to this dispatcher.

Instead of asserting, it should really be checking whether the ebmlid
matches.

Regression from c7644611.

(cherry picked from commit 20ed34f4)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>

This is already the case for CONTRIBFLAGS.

(cherry picked from commit bd558800)

Signed-off-by: Steve Lhomme <robux4@ycbcr.xyz>