This is a bit cleaner. Also don't repeat the chroma shift calculations
over and over, but store the image size instead, which is simpler and
will give us a chance to fix display of non-mod-2 image sizes.
gl_video.c contains all rendering code, gl_lcms.c the .icc loader and
creation of 3D LUT (and all LittleCMS specific code). vo_opengl.c is
reduced to interfacing between the various parts.
Do this instead of stuffing all x11/cocoa/win32/wayland specific code
into gl_common.c. The cocoa specific parts could probably go directly
into cocoa_common.m, possibly same with wayland.
Also redo how the list of backends is managed. Get rid of the GLTYPE_
constants. Instead of having a big switch() on GLTYPE_, each backend
entry has a function pointer to setup the MPGLContext callback (e.g.
mpgl_set_backend_x11()).
Makes the code a bit simpler to follow, at least in the "modern"
decoding path (update_video_nocorrect_pts() is used with old demuxers,
which don't return proper packets and need further parsing, so this code
looks less simple now).
Good news: MPV worked fine even without the fixes, but pointer size
mismatch warnings aren't the nicest things to leave lying around.
Fix macro that assumed HWND is uint32_t-sized.
Win64 is also a special butterfly and is an LLP64 platform on amd64
CPUs, while all the other amd64 platforms are LP64. Cygwin decided to go
with the other platforms, and thus sizeof(long) != sizeof(int), and in
cygwin's windows headers LONG is int-sized. Fix an mp_msg that assumed
LONG is long.
The vo_x11_clearwindow_part() function assumed that the video is always
centered. Replace it with a new vo_x11_clear_background() function
instead, which essentially does the same as the old function. It takes
the video rectangle instead of just the video size, and doesn't have to
make the assumption that the video rectangle is centered.
Also make vo_x11 use it (seems advantageous).
This was probably enabled to guarantee that panscan is always reset in
windowed mode. However, the window size should be exactly the video size
in windowed mode, unless in cases where the user forcibly changed the
window size (e.g. --geometry). In the former case, panscan will have no
influence at all, and in the latter case we want it to have influence.
If that's what the user asked for, there's no reason to introduce
special cases to ignore it on fullscreen.
The old behavior is perhaps accidentally due to the fact that aspect
calculations used to be disabled in windowed mode, rather than a
deliberate decision.
The code that is changed is responsible for scaling the video size to
display size, so that the resulting video rectangle is letter-boxed
inside the display window. This is before panscan calculations, which
can actually enlarge the video and make it larger than the display size
again. (src_dst_split_scaling() in vo.c takes cares of clipping the
video size to window size.)
I'm not sure why this rounding is done, as using panscan controls can
introduce odd sizes again. The rounding has been part of the code since
the initial commit. On the other hand, this rounding can slightly
influence the aspect ratio of the displayed image to the worse. It
forces the image to be scaled by an additional pixel, without actually
correcting the display size into the other direction.
Although video sizes are usually at least aligned on 2 (and often more),
odd sizes can still happen when playing e.g. anamorphic DVDs.
Remove the additional rounding.
(Note that we still round the _source_ image position and size when the
displayed image is larger than the screen, e.g. when panscan is used.
This is needed by some VOs so that the image source rectangle starts on
full chroma pixels. Maybe this rounding should be moved to the
respective VOs, which includes at least vo_direct3d.)
Remove lots of weird logic and dead code.
The only difference is that when specifying a monitor aspect ratio, it
will always upscale and never downscale.
The draw_osd change is a bit tricky: I guess originally, there was some
intention not to second-guess the generic aspect code, but the result is
the same and the code is more confusing.
The rescaling is rather silly. vo_get_src_dst_rects() doesn't return an
uncropped image, so the texture coordinates have to be recalculated,
which looks more complicated, but is actually what the other OpenGL VOs
also do.
Tested and fixed by Stefano Pigozzi.
Instead of manually calculating aspect ratio etc., use
vo_get_src_dst_rects(). This has the advantage that we can drop some
special cases in aspect.c later.
Simplify the code, and move resizing the X images from draw_image() to a
new resize() function.
Do some other simplifications. The gXErrorFlag check was actually dead
code, even in mplayer-svn. Special-casing on WinID is not needed
anymore (we created an embedded window, and its size is updated by
x11_common.c).
I introuced this regression in 8fc0b618d5. The backend would go into
incosistent state caused by calling `vo_cocoa_fullscreen` where it wasn't
needed.
Fixes#44
This only matters for those who want to use vf_pp. The old API is marked
as deprecated, and doesn't work on Libav. It was broken on FFmpeg, but
has recently started working again - the fields in question were not un-
deprecated though. Instead you're supposed to use a new API, which does
exactly the same thing (what...?).
Also don't pass the QP table with mp_image_copy_attributes() - it
probably does more harm than it's useful.
By the way, with -vf=dlopen=TOOLS/vf_dlopen/showqscale.so, it appears
the table as output by recent FFmpeg is offset by 1 macroblock in X
direction and 2 macroblocks in Y direction, which most likely will
interfere with normal vf_pp operation. However, this is not my problem.
The only real reason for this commit is that we can finally get rid of
all libav* related deprecation warnings. (Though they are constantly
deprecating APIs, so this will not last long.)
The old names have been deprecated a while ago, but were needed for
supporting older ffmpeg/libav versions. The deprecated identifiers
have been removed from recent Libav and FFmpeg git.
This change breaks compatibility with Libav 0.8.x and equivalent
FFmpeg releases.
avcodec_encode_video() was deprecated, and was finally removed from
Libav and FFmpeg git.
This change breaks compatibility with Libav 0.8.x. Thank the Libav
developers, not me.
Removes the local ret because it was only in an assert statement and also
removes the window size assignments, because they are already done in
config_wnidow_wayland.
Fix a regression introduced in commit 979ce46c64 causing a window to take up
more space than what the display allows.
Add keepCentered:YES, so that the video area is always clipped to the current
visible frame (even when using scale).
Fixes#38.
Latest nvidia drivers ignore the application setting, so this switch
makes even less sense than before. It's still possible to control this
with VO specific suboptions.
Separate the video output options from the big MPOpts structure and also only
pass the new mp_vo_opts structure to the vo backend.
Move video_driver_list into mp_vo_opts
Removes almost every global variabel in vo.h and puts them in a special struct
in MPOpts for video output related options.
Also we completly remove the options/globals pts and refresh rate because
they were unused.
Commit 9efe32120 stops recreating the hidden window used for OpenGL's
extensions autodetection. The `cocoa_common` code assumed the hidden flag could
only come when creating a window which is now not true anymore.
Move some code around so that hidden, fullscreen and window position are
respected on subsequent calls of config.
Apparently newer Mesa versions changed their <GL/glx.h> header, and
unconditionally define GLX_CONTEXT_MAJOR_VERSION_ARB and others. This
clashed with gl_header_fixes.h, a header which quarantines bad hacks
to make compilation possible on systems with outdated GL headers.
Specifically, our header was included before glx.h, so the hacks were
always active, and somehow Mesa's glx.h used to deal with this by not
redefining existing identifiers.
Fix the gl_header_fixes.h logic so the hacks are checked after including
glx.h.
When paused, --cursor-autohide worked with a precision of 500ms, which
is the main loop's default sleep time when paused. Cursor hiding is
polled in x11_common, and the main loop never called the X11 code at
the right time. Fix this by allowing the VO to set a time when it
should be called next.
This was done so because the X11 code had a hard to track down issue
with some window managers, and caused the VO window to be placed
incorrectly. This was fixed in the previous commit. Consequently, we can
remove this bad hack.
vo_opengl creates an invisible window with VOFLAG_HIDDEN in order to
test whether the OpenGL context is useable. The visible window is
created at a later point. This has been broken forever (in vo_gl,
now called vo_opengl_old, it could be avoided by disabling auto-
detection explicitly using the "yuv" sub-option). Avoiding
VOFLAG_HIDDEN only mitigates the issue, and a related bug can still
happen with some window managers (see below).
As a hack, code was added to vo_gl to destroy the (hidden) window so
that all state was destroyed. Later, more hacks were added to deal with
some caveats that came with recreating the window, such as probing for
the context up to 4 times.
Attempt to fix the problem properly. There were two problems: first,
the window was not resized to video size before mapping. This was the
main cause for the placement issue, e.g. mapping the window as 320x200,
and then resizing it. Second, mpv tried to force the window position
with XSetWMNormalHints and PPosition with values it just read with
XGetGeometry. This messes up the window manager's default placement.
It seems to be a race condition, and behavior is different across
various WMs too: with IceWM, the window manager's placement is usually
preferred, and with Fluxbox, mpv's position is preferred. mpv's default
position is centering the window on the screen, which is much nicer for
video in general than typical WM default placement, so it's possible
that this bug was perceived as a feature. (Users who want this have to
use --geometry="50%:50%", doing this by default is probably not safe
with all WMs.)
Since the old code was hard to follow and full of issues, it's easier
to redo it. Move general window creation stuff out of the
vo_x11_config_vo_window function, and move the resize logic into it.
This has been tested on IcwWM, Fluxbox, awesome, Unity/Compiz.
Store the window state (position and size) in vo_x11_state, instead of
in vo->dx/dy/dwidth/dheight. The VO variables are overwritten by vo.c on
every vo_config() call, which is extremely not helpful.
Now vo->dx/dy are mostly unused (except for passing the position forced
by the --geometry option), and vo->dwidth/dheight are set for the VO,
and otherwise read for resize detection only.
In the long term, the way vo_config() handles the --geometry option
should be changed, and vo->dx/dy should be removed.
Remove some useless stuff: VO_EVENT_MOVE and VO_EVENT_KEYPRESS were
generated, but unused. Wayland changes by Alexander Preisinger.
It was once used for vo_sdl (the old one based on SDL 1.2), since SDL
apparently lost the GL state when switching to fullscreen. The new
vo_sdl (using SDL 1.3) doesn't use or need this. It's dead code, so
get rid of it.
