Fatih Uzunoğlu authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

ListView.delayRemove should apply to the
delegate, which has id: delegate, and not
the root.

Thomas Guillem authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

And fix potential data races, that could have become criticial with the
new medialibary (more than one samba access in //). Indeed, the SMBCCTX
and SMBCFILE were stored statically and fetch from an unique fd without
any lock mechanism.

And favor our original VAAPI backend.

Fixes #26361

Fixes #26367

Alexandre Janniaux authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

This improves the documentation by exposing a C++-idiomatic way to
handle the resources without leaking for most users copying this part of
the code to their application.

Alexandre Janniaux authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

C++14 is needed to use std::make_unique<> in particular, and is already
a requirement for libvlc anyway.

Fix #26356

Lyndon Brown authored 5 years ago and Hugo Beauzée-Luyssen committed 3 years ago

...and indeed ignoring them.

as per the previous commit, we want to get rid of cat hints from option
sets, instead looking up cats from subcats. we have already implemented in
the previous commit the lookup mechanism, and started ignoring cat hints
from plugins, but still relied upon them from the core set for constructing
the base tree (and reacted to them in building core panels).

here we adjust the base tree construction code to completely remove our
reliance upon cat hints. we also now completely ignore them when building
core panels (no difference results from this).

we are now in a position to purge the cat items themselves from the entire
infrastructure along with all remaining artefacts, though that will be done
separately later.

note that the macos code needed no such change here, it was taken care of
completely in the previous commit.

Lyndon Brown authored 5 years ago and Hugo Beauzée-Luyssen committed 3 years ago

having `set_category()` + `set_subcategory()` pairs of calls to setup the
location in the preferences tree where a plugin node needs to be attached
to, with corresponding pairs of option records in the plugin option sets,
is silly, inefficient and bug prone. all we need is the subcat records
and we can look up the corresponding cat from the data table when it
comes to time to build the tree.

here we take a step forward to address this, adding cat data to the subcat
data table, and a lookup function making use of it.

note, qt and macos tree construction are done a little differently from
each other, so while the change here is entirely sufficient to rid the core
option set of category hints as far as the macos code is concerned, more
work in a subsequent commit will be needed for qt. thus the purge of
`set_category()` must be postponed for a later commit and so for now we
must just ignore those records.

a subcat table entry was added for `SUBCAT_HIDDEN` since for now the qt
code could potentially look this up and we should preserve the existing
behaviour for it for now.

Lyndon Brown authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

with respect to plugin vs. module.

before we adjust it in the next commit.

Lyndon Brown authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

these 'index-of' helpers find the index for the given cat/subcat. this
allows us to de-duplicate the code from the other helper functions (both
for those that already exist and ones to be added in further commits).
they loop based upon array length rather than looking out for some
particular condition of varying properties in an end marker record,
allowing us to also drop those.

i used an index-of design here rather than have functions that return
pointers because there will be a secondary use for index-of helpers in
later commits.

Lyndon Brown authored 5 years ago and Hugo Beauzée-Luyssen committed 3 years ago

...to prepare for divergence in the data held for categories vs.
subcategories.

we replace here the generic helpers:
 - `config_CategoryNameGet()`
 - `config_CategoryHelpGet()`

with:
 - `vlc_config_cat_GetName()`
 - `vlc_config_cat_GetHelp()`
 - `vlc_config_subcat_GetName()`
 - `vlc_config_subcat_GetHelp()`

the new names conform to the format requested for the
`vlc_config_subcat_IsGeneral()` function added in MR 313.

opportunity was taken to remove the old style prefixes from the attributes
of `config_category_t`, affecting a few additional lines, avoiding
introducing inconsistency or new instances of the old style with the
addition of `config_subcategory_t`.

Lyndon Brown authored 5 years ago and Hugo Beauzée-Luyssen committed 3 years ago

it is used only to define the structure of `categories_array` which
lives in `vlc_config_cat.h`; it belongs next to that.

Lyndon Brown authored 5 years ago and Hugo Beauzée-Luyssen committed 3 years ago

this stuff is used by prefs, and is not even used by help currently.

Benjamin Arnaud authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

fix #26275

This is a manual backport of
https://github.com/protocolbuffers/protobuf/commit/01a05a53f40ca2ac5f0af10c6cc0810bee39b792

Alexandre Janniaux authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

The test was failing randomly because the decoder was asked to decode
more data and its output format was reset to VLC_CODEC_UNKNOWN, which is
not an allocatable format for picture_NewFromResource.

Abandon the picture allocation in decoder_decoder(dec, picture) callback
to let the scenario handle the allocation, and in particular discard it
if it already did the test. The new test_finished boolean will force the
decoder callback to no-op. It is a bit redundant with the display_opened
callback where it could loop doing the update until it is correctly
closed but much clearer.

I'm not sure yet, but it seems that the easiest method to control the
behaviour of the test is to have a custom demux to push data at will.
It's adding a lot of code and is trickier to manage so here we fix the
test first.

Fixes #26374

Rémi Denis-Courmont authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

Normally closing a logger should not incur new log messages, but...
the early logger does exactly that, and this prevents it from looping
infinitely into itself. This situation occurs if no proper logger was
installed until the log switch was destroyed.

Fixes #26363.

This effectively reverts
commit e8700cbd.

Thomas Guillem authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

Fixes #22909

Regression from c9ab8bd8

Fixes #26367

This commit introduces an integration test infrastructure which mocks
the external components (filters, decoders, window, display) used by the
video pipeline and starts this pipeline using the usual input item
properties and player.

The test will spawn an interface to have an initial vlc_object_t without
going though the private layers of libvlc (avoiding casting the instance
to private type, etc), so as to avoid encouraging code using such casts.

It uses this infrastructure to prevent regression from the ticket #26345.
By ensuring that the second failed decoder update won't crash, we can
check that an already started vout won't lead to double-join/free like
it was fixed in a7c99d05.

The test itself will later be extended to add more sensible cases.