GitLab

This does not actually has any effect for the build, but
is necessary to get rid of some warnings about ARC
in Xcode.

copy/paste

NSWindowStyleMask was introduced in 10.12 SDK, and used for methods like
`initWithContentRect:styleMask:` which before 10.12 used NSUInteger.
Continuing to use NSUInteger works but causes a warning when compiling
with 10.12 SDK.

Fixes #18105

This makes it easier to find the relevant code and
additionally has the benefit that imports can be more fine-grained
in the future, instead of just having one misc.h import,
where it is hard to know, without looking at it, what it is
actually needed for.

otherwise outputs timings less frame

the variable named str is not referred to after the lines in question,
as such this simplifies the implementation after the recent changes to
ParseFEC, ParseModulation, and ParseGuard.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

All three of ParseFec, ParseModulation, and Parseguard, are
potentially called with NULL (on unexpected (truncated) input), these
changes make sure that we do not dereference the passed pointer if
such happens.

Instead NULL is returned to signal that the parsing failed.

fixes: #18126
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

An xml-document containing a tag without a name is ill-formed, so this
case will never happen in practice (as the xml-reader will error out
before).

However, if it for some reason would happen, it will be handled due to
the fact that there will not be any handler matching "".
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This diagnostic is not helpful, and should definitely not use fprintf
(as such it is removed).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The variable in question is only needed within the preprocessor-if
that follows, as such it makes more sense to declare only declare it
if that is actually reached (while also supressing a warning
diagnostic related to the variable being unused).

An empty statement has been added after the label in order to comply
with the ISO C Standard, which mandates that a label preceedes a
statement (which a variable declaration is not).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

There can be more dialogs providers than qt, and qt might be compiled
even though the affected diagnostics are issued. To make the behavior
more relevant in/suitable for all circumstances:

 - the duplicate diagnostic has been removed, and;
 - mention of qt is removed.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The previous implementation suffered from a bunch of different issues,
mostly related to not properly checking errors and the current state
of the XML-parser.

This new implementation is somewhat more verbose, but manages things
in a correct manner:

 - prevents infinite loops on unexpected data
 - prevents memory corruption on unexpected data
 - prevents memory-leaks on error

fixes: #18124
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

An XML file containing a single root element without children would,
given the previous implementation, have its only element ignored.
Resulting in a memory-leak of the node in question, while also (more
importantly) making it impossible to access the associated data.

As XML does not allow documents such as the below (only a single root
is allowed):

   <?xml version="1.0">
   <ill-formed />
   <ill-formed></ill-formed>

Simply checking to see so that we are not popping away all our tags
are sufficient in order to fix this bug. The changes also make sure
that we do not invoke std::stack<...>::pop on an empty container
(which is undefined-behavior).

fixes: #18122
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The previous implementation would result in a diagnostic from gcc due
to it being unable to prove that b_bold and b_italic was always
initialized before usage.

These changes get rid of the diagnostic, while also simplifying the
implementation. A comment was added to make the purpose of the loop
clear.
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

The socks protocol does not allow username/password to be longer than
255 characters. Instead of silently truncating the entities to the
maximum limit we should simply reject them.

These changes also removes the XXX comment, as it is not strictly "we"
who does not support longer username/passwords.. it's the protocol
itself.
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

The code of course works as it should, but llvm will spit out a
warning diagnostic due to extraneous parenthesis (thinking that it
probably should be "=" instead of "==").
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

increases readability.
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

psz_value is given a value when the node type is XML_READER_TEXT, but
given that this is not guaranteed to ever happen before
XML_READER_ENDELEM there is nothing stating that psz_value is not
NULL.

fixes: #18125
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

The previous implementation would crash on malformed/unexpected input
due to xml_ReaderNextNode returning a value different than
XML_READER_STARTELEM, meaning that accessing "node" is ill-formed.

fixes: #18123
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

The relevant if-statement is meant to abort parsing if strdup fails,
and should therfore check the value of "psz_elname" (as node is always
not-NULL on the line in question).
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

see commits prior to a4b48614Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

The now removed lines of code are part of the legacy implementation
(before vlc_UrlParse was used), and are no longer required.
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

it is not set in other places and some compilers don't like a stray comma.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

MMRESULT is not defined in that case with MS headers
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

meets TIMING_SEI dependency on SPS

Some SEI requires parsing context from picture
header. We are then currently always wrong for
first frame or first frame before a change in those
params. We need to delay parsing.

allows to insert something else than sps/pps
and uses last block pointer

https://bugreports.qt.io/browse/QTBUG-53255
As a side note, Qt have been asked to include those patches included in Qt 5.6.x
branch, but it won't happen until 5.9 gets released, if ever.