I had delayed reformatting mplayer.c as I wanted to split it, but
since I didn't come up with a good way to do that I'll clean up the
messy formatting now.
Parsing of commands with string arguments had several problems. First,
escape handling modified the original string, which broke keyboard
command definitions with strings containing escapes. Second, in an
argument like "\\" the second quote was incorrectly considered escaped
even though the preceding '\' was itself escaped. Third, an
unterminated quoted string resulted in the argument being set to NULL
even if this violated the declared the minimum number of arguments to
the command type; this could cause a crash in the code executing the
parsed commands. Fix by rewriting the string argument parsing. Also
change int/float argument parsing so that the whole command is
rejected on error.
Initialize mpctx->last_chapter_seek to -2 instead of -1. This changes
get_current_chapter() return value to -2 for files which have no
chapters. -2 is used by some commands related to chapters to recognize
files without chapters and return failure without any effect in that
case.
Ordered chapter code tries opening files to find those matching the
SegmentUID values specified in the timeline. Previously this scan did
a full initialization of the Matroska demuxer for each file, then
checked whether the UID value in the demuxer was a match. Make the
scan code instead provide a list of searched-for UIDs to the demuxer
open code, and make that do a comparison against the list as soon as
it sees the UID in the file, aborting if there is no match.
Also fix units used in "Merging timeline part" verbose message.
Video filter help output was printed twice since it used a MSGT type
that was not suppressed during preparsing. Fix. This removes a
"ID_VIDEO_FILTERS" line that was printed --identify, but I'm not aware of
anything using that and other things like audio filters have no such
ID_ output either. Also change some printf() calls to mp_msg() in code
printing filter-specific help.
A trailing separator in string list options was ignored after recent
commit e873d703e9 ("options: change option parsing to use bstr"),
which broke uses such as "-vo vdpau,". Fix.
The command line parsing code recorded the next commandline argument,
if any, as the parameter of any option recorded to playtree. Thus a
command line like "mplayer file -fs -aid 1" would record option "-fs"
with a bogus argument "-aid". Historically this triggered no visible
problems because such bogus arguments were silently ignored when
interpreting the options later. However after recent commit 507fa7e2c2
("options: indicate ambiguous option parameters explicitly")
parameters to flag options are no longer ignored, and the bogus values
now triggered parsing errors. Add a check to stop recording parameters
for old-style single-dash options if m_config_check_option() says the
option did not consume any arguments.
Seeking while paused could result in the current audio pts being
reported incorrectly due to relevant variables not being reinitialized
after the seek until more audio was played. When playing audio-only
files, this meant that current overall playback position could be
reported incorrectly which in turn could break further seeks. Improve
things on two levels: First, store the seek target position and use
that as the current playback position for audio-only files until
things can be reinitialized. Second, try to reinitialize audio
decoding enough to know its current pts even while paused. Also avoid
printing the actual huge negative value of MP_NOPTS_VALUE on the
status line when pts could not be determined.
Change written_audio_pts() and playing_audio_pts() to return
MP_NOPTS_VALUE if no reasonable pts estimate is available. Before they
returned some incorrect value typically around zero (but not
necessarily exactly that).
Recent commit 5d5ca22a6d ("options: commandline: accept --foo=xyz
style options") left some bad code under "#ifdef MP_DEBUG" in
playtree.c, which caused a compilation failure if configured with
"--enable-debug". Fix this. Having the "#ifdef MP_DEBUG" there was
completely unnecessary; it only increased the risk for this kind of
problems for no real benefit - executing the asserts under it would
have no noticeable performance or other penalty in default builds
either. Remove several cases of such harmful "#ifdef MP_DEBUG".
After commit 39e373aa8d ("options: allocate dynamic options with
talloc") dynamically allocated options must be allocated with talloc.
stream_cdda allocated a string used with the option machinery using
strdup(). Fix. Also silence a warning.
Code mapping X keyboard events to internal player key names remapped
the incoming KeySym values in a way that only kept information about
the lowest byte of the value and whether or not the value was below
256. This caused collisions between values. Remove this obviously
broken mapping and use the raw KeySym values instead. I'm not familiar
enough with X key handling to tell whether there was any valid
motivation whatsoever for the mapping, but the implementation was
certainly broken; if something else breaks it'll have to be fixed
later.
If parsing an option fails, print a string corresponding to the parse
function return value (M_OPT_MISSING_PARAM etc). The primary
motivation is that the parsing code already outputs messages
explaining most problems, but does not itself print anything in the
missing parameter case. Before double-dash --options such errors were
rare (or rather they resulted in the next commandline argument being
silently misinterpreted as an argument to the previous option
instead); but now an argument like "--ss" should give a better
indication about the problem than just "Error parsing option".
Allow writing commandline options with two leading dashes. In this
mode a parameter for the option, if any, follows after a '=';
following separate commandline arguments are never consumed as a
parameter to a previous double-dash option.
Flag options may omit parameter and behave like old single-dash
syntax. "--fs=yes", "--fs=no" and "--fs" are all valid; the first two
behave like configuration file "fs=yes" and "fs=no", and last is the
same as old "-fs" (same effect as "--fs=yes").
A "--" argument on the command line is used to indicate that all
following arguments should be interpreted as filenames, not options.
Specifying it as the last argument was considered an error. I see no
particular reason to forbid the following filename list to be empty,
nor do other programs with similar functionalitly I know about treat
it that way. So just ignore a "--" with no more arguments.
Make per-file loop option start from --ss position, not always 0.
Do looping in more cases; before looping was only done when
encountering real end of file, now it also happens for example at
--endpos or --frames limits. Also move the --ss option to the option
struct.
Command line options like "-foo xyz" are ambiguous: "xyz" may be a
parameter to the option "foo" or an unrelated argument. Instead of
relying on the struct m_config mode field (commandline/file) pass
parameters to specify ambiguous mode explicitly. Meant for "--foo"
options which are never ambiguous on command line either.
Rename the BSTR() function to bstr(). The former caused a conflict
with some Windows OS name, and it's no longer a macro so uppercase
naming is less appropriate.
The --ass-force-style option was only applied when the main libass
library handle was created. Thus any per-file option changes later had
no effect. Do the ass_set_style_overrides() call in per-file
initialization instead so that possible changes will be applied. Also
move the option variable to the option struct.
Current libass will crash (usually) if you set style overrides to a
nonempty value, then an empty one. It'll be easier to trigger this bug
after this commit, but the problem is not on mplayer2 side. The fix is
trivial so hopefully there will be a fixed libass soon.
Commit df899f59be removing a write outside a buffer triggered another
problem, as for some reason the code did not 0-terminate its read
buffer in the specific case that it had encountered an EOF, and as a
result could parse contents left in the buffer for a second time.
Usually this resulted in parsing error messages. Fix the problem by
rewriting the offending code in a less hacky form.
Recent commit 82b8f89bae ("input: rework event reading and command
queuing") had a problem in the handling of commands generated by
autorepeat code (this code is only used with joystick and Apple IR
remote input). Fix (still untested though).
Do the global initialization of libavcodec and libavformat
(avcodec_register_all(), av_register_all()) immediately on program
startup and remove the initialization calls from various individual
modules that use libavcodec/libavformat functionality.
Libavutil eval API seems to be enough for vf_qp, so enable the filter
if the API is available. Also clean up some left over lines for other
filters in Makefile.
Window events or user commands could trigger a call to resize() before
config() had been called. This caused at least error messages. Add a
check to make resize() exit without doing anything in this case. It
will be called again later when the output is configured.
Setting O_NONBLOCK on a file descriptor also affects all other fds
that share the same underlying open file description, and in case of
stdin such sharing is likely. Making stdin nonblocking can also make
stdout nonblocking (they may be the same connection to a terminal),
and it can also affect other processes (in "program1 | program2", the
shell may give the same terminal connection to program1 as stdin and
to program2 as stdout, thus program1 making its stdin nonblocking also
turns program2's stdout nonblocking).
To avoid these problems stop making fd 0 nonblocking. After the
previous commit this should no longer cause problems as long as
select() does not spuriously report the fd as readable.
Rework much of the logic related to reading from event sources and
queuing commands. The two biggest architecture changes are:
- The code buffering keycodes in mp_fifo.c is gone. Instead key input
is now immediately fed to input.c and interpreted as commands, and
then the commands are buffered instead.
- mp_input_get_cmd() now always tries to read every available event
from every event source and convert them to (buffered) commands.
Before it would only process new events until one new command became
available.
Some relevant behavior changes:
- Before commands could be lost when stream code called
mp_input_check_interrupt() which read commands (to see if they were
of types that triggered aborts during slow IO tasks) and then threw
them away. This was especially an issue if cache was enabled and slow
to read. Fixed - now it's possible to check whether there are queued
commands which will abort playback of the current file without
throwing other commands away.
- mp_input_check_interrupt() now prints a message if it returns
true. This is especially useful because the failures caused by
aborted stream reads can trigger error messages from other code that
was doing the read; the new message makes it more obvious what the
cause of the subsequent error messages is.
- It's now possible to again avoid making stdin non-blocking (which
caused some issues) without reintroducing extra latency. The change
will be done in a subsequent commit.
- Event sources that do not support select() should now have somewhat
lower latency in certain situations as they will be checked both
before and after select()/sleep in input reading; before the sleep
always happened first even if such sources already had queued
input. Before the key fifo was also handled in this manner (first
key triggered select, but if multiple were read then rest could be
delayed; however in most cases this didn't add latency in practice
as after central code started doing command handling it queried for
further commands with a max sleep time of 0).
- Key fifo limiting is more accurate now: it now counts actual
commands intead of keycodes, and all queued keys are read
immediately from input devices so they can be counted correctly.
- Since keypresses are now interpreted immediately, commands which
change keybindings will no longer affect following keypresses that
have already been read before the command is executed. This should
not be an issue in practice with current keybinding behavior.
Remove some unnecessary typedefs and remove pointless mp_ prefix from
some internal struct names.
Change the type of the "close_func" pointers from "void f(int fd)" to
"int f(int fd)" so that using standard close() there is valid.
Delete some useless assert() statements.
Move internal MP_MAX_KEY_DOWN define from input.h to input.c.
Commit aba8a1838a which added 9- and 10-bit formats failed to change
the definition of the IMGFMT_IS_YUVP16() macro (which is misnamed btw,
it matches 9, 10 and 16 bits). This prevented vo_gl from accepting
input in supported 9 and 10 bit colorspaces. Fix.
After commit 39e373aa8d ("options: allocate dynamic options with
talloc") dynamically allocated options must be allocated with talloc.
Code implementing -use-filename-title still set opts->vo_wintitle to a
value from strdup(), triggering an abort when the option was freed.
Fix.
Remove the copy of the "lowres" field that vd_ffmpeg kept in its
private struct and use the value from AVCodecContext directly instead.
The copy gave no benefit and it could be set to the wrong value if
someone used "-lavdopts o=lowres=X" (which would change the real value
but not the copy).
Support for -lavdopts skipframe had been broken since commit bc767c2a9
as framedrop logic now set the same field and thus overwrote the value
set from the option. Change that code to set it to the original value
instead of 0 when not dropping frames.
libsmbclient uses dynamically linked libtalloc.so which has symbols
with names matching those in the internal talloc.c. This name
collision caused a crash under stream_smb when trying to play
anything with smb://. Add a "#pragma GCC visibility" hack to talloc.h
to hide the internal symbols from being visible to the outside
libtalloc.so. Hopefully this doesn't break any platform worse than
possibly causing a compiler warning about an unknown pragma (at least
GCC and clang recognize it).
Before there was no attempt to set warning flags sanely when compiling
with clang. Set some reasonable defaults that cut down noise and
enable various non-default warnings that are enabled with GCC too.
I'm not sure whether clang is supposed to support more of the options
now used with GCC - it accepts some of those options but they
apparently have no effect; I didn't find any real documentation about
the individual warnings.
This reverts commit 1e544365bf.
I was careless when cherry-picking the reverted commit - the only
effect it had was add an unnecessary libavutil dependency to a HOST_CC
compiled file.
The name of the project is "libass". "ASS" alone refers only to the
subtitle format, not the library, and --disable-ass did not completely
disable handling of subtitles in this format - only advanced rendering
with libass. Thus --disable-libass is a better name.
