- Jan 23, 2022
-
-
Rémi Denis-Courmont authoredThe pkg-config file correctly supplies -pthread for LDFLAGS, but libtool discards it, leading to linkage failure if libupnp is linked statically.
-
We don't depend on OpenGL types for the signature anymore.
-
Create an header file containing the interop definition and operations, so that the entrypoint can also be implemented out-of-tree and doesn't depend on code from the modules/ directory.
-
The implementations are now independant of the vlc_gl_api_t structure and the OpenGL virtual table. The rationale for removing vlc_gl_api_t is that interop implementations are already required to load specific function for most of them (like the eglCreateImageKHR function) and having vlc_gl_api_t in the public module API require exposing the OpenGL vtable and types to the public scope, which is pretty ard to justify and port/maintain for every platform.
-
Avoid using vlc_gl_api_t to determine whether npot are usable or not and directly inline the same check from gl_api.c.
-
Allocate a private part for the interop containing the actual OpenGL vtable used by the interop "core" code.
-
Directly load the function in the call. This is not very pretty, but much easier to write than storing the pointer in a private part of the importer.
-
The API type is already available in the vlc_gl_t object.
-
This improves compatibility with the core profile for OpenGL and removes usage of the extension field from vlc_gl_api_t, to remove dependency from the interop to the vlc_gl_api_t.
-
Checking an extension is different between OpenGL compatibility profile or ES2 profile, and OpenGL core profile. Separating the call in a different function simplify the handling of those extension checks, and provides better compatibility with the core profile.
-
-
-
This removes the dependency from the vtable in vlc_gl_api_t, for future removal of the api from the interop API.
-
This removes the dependency from the vtable in vlc_gl_api_t, for future removal of the api from the interop API.
-
The surface type doesn't matter, especially since initialization for X11 is done by the decoder device now. All that matter is that the extension is available for the given context, providing access to the vdpau interop functions from this context.
-
This removes the dependency from the vtable in vlc_gl_api_t.
-
config.h is ought to be included in source files, not headers.
-
interop and picture_t definition are not needed in the header.
-
Only glBindTexture is used from the interop code, so remove the dependency from the vtable in vlc_gl_api_t altogether. The rationale for removing vlc_gl_api_t is that interop implementations are already required to load specific function for most of them (like the eglCreateImageKHR function) and having vlc_gl_api_t in the public module API require exposing the OpenGL vtable and types to the public scope, which is pretty ard to justify and port/maintain for every platform.
-
Rémi Denis-Courmont authored
libmpg123, xcb are not rebuildable in the current CI setup. upnp builds but breaks the VLC build (refs #24239).
-
Rémi Denis-Courmont authored
fontconfig cannot be cross-compiled onto Linux with current contribs. Until this is fixed, we need to turn it off to get CI out of its metastable state.
-
Rémi Denis-Courmont authored
The code ostensibly intended to append to PKG_CONFIG_LIBDIR. But in the case that it was set from the command line, the code had no actual effects: indeed, Make will not override a supplied variable unless 'override' is explicitly passed. Forcefully appending values would have been wrong in any case, as that have left literally no ways to fully override PKG_CONFIG_LIBDIR. Concretely it would have become completely impossible to prevent pkg-config searching in /usr/lib. So rather than attempt to fix the code, this patch makes the current behaviour official and removes the useless code.
-
Rémi Denis-Courmont authored
-
Rémi Denis-Courmont authored
This moves the PKG_CONFIG and PKG_CONFIG_LIBDIR initialisation where PKG_CONFIG_PATH is. Note that the moved code is predicated on HAVE_CROSS_COMPILE either way. No functional changes.
-
Rémi Denis-Courmont authored
Traditionally the presence of the header file was not considered sufficient, though this was somewhat forgotten with the prevalence of pkg-config. However most libraries have portable headers, while the shared library is platform-dependent. It is common for the header to be present while the library is absent with "multilib" installations (i.e. Linux installation with multiple architectures).
-
Rémi Denis-Courmont authored
Traditionally the presence of the header file was not considered sufficient, though this was somewhat forgotten with the prevalence of pkg-config. However most libraries have portable headers, while the shared library is platform-dependent. It is common for the header to be present while the library is absent with "multilib" installations (i.e. Linux installation with multiple architectures).
-
Rémi Denis-Courmont authored
The bin/ directory is shared between all contrib targets. Whilst cleaning one target, only the executables for that target should be erased.
-
Rémi Denis-Courmont authored
Newer versions of GNU/make mark the job server pipe file descriptors with the close-on-exec flag. This prevents the file descriptors from being leaked into other programs executed by the children processes of GNU/make. But of course, the flip side is that the file descriptors are closed even if the program is also a make-like tool that wants to participate in the job counting and access the job server. To support recursive make, GNU/make will clear the close-on-exec flag in a child process that is about to execute a command: - referencing the variable MAKE, or - prefixed by a plus sign '+'. In this particular case, the command is $(CARGO_VENDOR_SETUP), not $(MAKE), so this does not work out of the box. Add an explicit plus.
-
AOM_USAGE_ALL_INTRA isn't being defined while AOM_USAGE_GOOD_QUALITY is, refs https://aomedia.googlesource.com/aom/+/refs/tags/v3.0.0/aom/aom_encoder.h#1001
-
overlooked in dcebb25b.
-
fixes #26510.
-
should always be pts == dts
-
-
-
Rémi Denis-Courmont authored
-
- Jan 22, 2022
-
-
Rémi Denis-Courmont authored
-
Rémi Denis-Courmont authored
Note that a newer version of FFmpeg is necessary to get actual platform-specific acceleration. This merely allows plain C build.
-
fixes #26274
-
-
Rémi Denis-Courmont authored
The whole Requires.private stanza, not just its value, is expanded by config.status from the configure script. So the conversion from private to public must be done after expansion.
-
