Rémi Denis-Courmont authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

As defined in RFC2397.

Thomas Guillem authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

cf. https://developer.apple.com/library/archive/documentation/DeveloperTools/Conceptual/DynamicLibraries/100-Articles/DynamicLibraryUsageGuidelines.html#//apple_ref/doc/uid/TP40001928-SW12

"""
When the library name is a filename (that is, when it doesn’t include
directory names), the dynamic loader searches for the library in several
locations until it finds it, in the following order:

 - $LD_LIBRARY_PATH
 - $DYLD_LIBRARY_PATH
 - The process’s working directory
 - $DYLD_FALLBACK_LIBRARY_PATH

When the library name contains at least one directory name, that is,
when the name is a pathname (relative or fully qualified), the dynamic
loader searches for the library in the following order:

 - $DYLD_LIBRARY_PATH using the filename
 - The given pathname
 - $DYLD_FALLBACK_LIBRARY_PATH using the filename
 """

Conclusion: DYLD_LIBRARY_PATH on macOS behave like LD_LIBRARY_PATH on
Linux but LD_LIBRARY_PATH on macOS has a different behavior than
LD_LIBRARY_PATH on Linux, so DYLD_LIBRARY_PATH should be used on macOS.
Specially since we can't really control if the library name will include
a path or not (relative or absolute).

This does not make the live555 module dependent on libopus, just the helper
functions in opus_header.{c,h}

Lyndon Brown authored 5 years ago and Jean-Baptiste Kempf committed 3 years ago

caused options to always be adjusted to value of 1 when saving prefs.

Lyndon Brown authored 5 years ago and Jean-Baptiste Kempf committed 3 years ago

"flat" is not in the choice list and is meaningless to the module, not
matching any of its known presets. it is thus equivalent to a default of
"none". (see `window_get_param()` in `modules/visualization/visual/window.c`
which describes this as falling back to "using rectangular window (i.e. no
window)").

it is a reasonable assumption that the author meant the default to be the
`"flattop"` choice, and just made a mistake here, however this is not
certain and the preference expressed in review was to preserve the existing
behaviour by switching to "none".

Lyndon Brown authored 5 years ago and Jean-Baptiste Kempf committed 3 years ago

the lowercase "auto" default option value did not match the uppercase
"Auto" in the choice list.

option values should typically be lowercase. this fixes the mistake with
the case of the choice value, whilst maintaining backwards compatibility
with existing CLI usage and from saved settings files. this was the
clear preference for fixing this resulting from the MR discussion.

Lyndon Brown authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

(no functional change)

 - better prepare for the next commit where we need to support both "Auto"
   and "auto".
 - avoid multiple `strdup()`.
 - avoid avoidable string comparisons.

Lyndon Brown authored 5 years ago and Jean-Baptiste Kempf committed 3 years ago

NULL does not map to an entry in the choice list.

also, mark the choice labels for translation.

Lyndon Brown authored 5 years ago and Jean-Baptiste Kempf committed 3 years ago

-1 was not in the choice list.

this adds the missing choice list entry corresponding to the default.
i've made a guess at the right label and explanation to use based upon
what seems to be happening with a -1 value in `stereo_to_mono()` and
from bits of git history relating to it.

Lyndon Brown authored 5 years ago and Jean-Baptiste Kempf committed 3 years ago

0 was not in the choice list.

it looks like 0 was interpreted by the code as the same as -1 via
`dvb_parse_transmit_mode()`, so not a big problem.

Pierre Lamot authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

fix: #26236

Note that this now uses `unsigned char` rather than `char`, consistent
with the expectation for the `%c` format string specifier. This also
avoids using a signed value to index an array.

This paves the way to split private and public properties from
module_config_t.

Private core headers don't need this.

Steve Lhomme authored 3 years ago and Rémi Denis-Courmont committed 3 years ago

This would make the detection of a canceled thread vs normal run impossible.

Not implemented in posix/thread.c because we don't get the return code in our
code. We could do it with an indirect call.

Prince Gupta authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

fixup 96053261

They are not used anymore since it was moved to the interop_dxva2
module.

This will use extensions from wgl directly into interop_dxva2, instead
of using those loaded by wgl.

The only users of those extensions are interop modules, so only
interop_dxva2 for extension related to WGL integration.

By moving the loading of those extension to the interop, we can have
them working even when the OpenGL platform context is provided by the
vglmem (vgl.c) module callbacks, allowing hardware acceleration even
when using the libvlc OpenGL callbacks.

Filters would likely be using only OpenGL extensions that are already
available, but could be using the same mechanism in the worst-case
scenario.

Refs #25234, #25256

Prince Gupta authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

fixes associated compile time warning

Prince Gupta authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

fixes undefined sanitizers warnings -
../../modules/gui/qt/maininterface/main_interface.cpp:283:19: runtime
error: load of value 190, which is not a valid value for type 'bool'
.... etc

Hugo Beauzée-Luyssen authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

This will help automatically cancel outdated pipelines instead of having
to do so manually

Benjamin Arnaud authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

fix #26037

Romain Vimont authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

The orientation matrix was not initialized if orientation was
ORIENT_NORMAL.

The regression has been introduced by commit
01db27e1: since the matrix has been
reduced to 2x3, all the matrix components were set by the macro
MATRIX_SET(), so the prealable initialization to "identity" has been
removed.

However, the switch did not cover the case ORIENT_NORMAL, which had
nothing to do if the matrix was already initialized to "identity".

In practice, this caused a problem only on Android with hardware
acceleration enabled, because interop_android.c is the only interop
which does not apply a vflip to the input format (see commit
1fee4df0).