| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
| |
This commit sucks bad, but everything else is worse is other ways.
Basically, the current vblank waiting time in the vo_wayland_wait_frame
function (calculated very carefully using presentation statistics) is
randomly too short. Some compositors are quite variable when they
actually return callback so our timeout expires too quickly and throws
everything off. The fix? Add an arbitrary 5% to the vblank value and
pray that nothing gets off that much. Why did they have to make
swapinterval 1/fifo mode indefinitely block? Fixes #9504.
|
|
|
|
|
|
|
| |
Better to avoid any wonky calculations on startup with garbage values.
The others end up being derived from last_ust/last_msc. refresh_interval
is referenced exactly once and could, in theory, result in some terribly
erroneous vblank time.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A read can be prepared on the wayland display FD that is never actually
read. This occurs when events are triggered on other FDs in the fd set.
This change cancels a prepared read if poll reported no events for it.
This fixes some hangs due to how nvidia's EGL implementation polls on
the wayland fd unlike mesa implementations. It is based on nvidia's
proposed fix for qt's similar message pump in
https://codereview.qt-project.org/c/qt/qtwayland/+/373473
Signed-off-by: Kurt Kartaltepe <kkartaltepe@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This possibility actually existed for years. The wayland protocol is
asynchronous and there's no restriction on when a compositor can send a
surface enter event. In mpv's case, the surface enter event is used to
set some vital things regarded geometry/scaling etc. However, this
implictly assumes that wl->current_output is actually initialized. The
vast majority of the time, vo_wayland_reconfig will happen first which
is where wl->current_output is, and should, be created. There's no
rule/law that the ordering of events will always occur in this order.
Plasma with certain auto-profile conditions can send the surface enter
event before mpv does its initial reconfig. That segfaults of course.
Just add a check to make sure we have wl->current_output here and return
if we don't. This assumes that the compositor will send us another
surface enterance event when mpv actually does the initial surface
commit and roundtrip request later. Wayland logs indicate this does
happen. Fixes #9492.
|
|
|
|
|
|
| |
It was never implemented before but it's trivial. As an aside, touch
events currently don't support modifiers either (is this a thing?). Well
if someone complains that can be done later. Fixes #9490.
|
|
|
|
|
|
|
|
|
| |
This was originally added in f2afae55e95b4b1eec1aeb828ba6ff1f0695d993
for unclear reasons (way to go me). This concept is clearly incorrect.
It doesn't matter what state the window is in. As soon as mpv detects a
scale change, it needs to reset the buffer scale of the window. Just
remove all this junk and put wl_surface_set_buffer_scale in
set_surface_scaling like it should be. Related issue: #9426.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Back when runtime updating of autofit/geometry was added for wayland and
x11 (commits: 4445ac828dca0298543103094357e64f8828ef56 and
ced92ba607ebd98687b26ef3d8c09d5f6e22cf4b respectively), the naive
assumption was that window-related geometry would always be available.
While this is true 99% of the time, this isn't a guarentee. It is
possible for certain things such as loading shaders to delay starting up
the player. This causes autofit/geometry options to be registered as a
runtime update and triggers VOCTRL_VO_OPTS_CHANGED. This ends up calling
some geometry-related functions but this happens before the actual
values are available. Hence, a nullptr was accessed which segfaults. At
least one user experienced this with a combination of options in wayland
but in theory the same thing could happen under x11.
The fix is simple. Just be sure to check that the required geometry is
available before doing any calculations. In wayland, this would be
wl->current_output. Additionally add an assert to set_geometry (we
should never use this function without wl->current_output) to be extra
sure. In x11, the check is on x11->window. Later when the reconfig for
each backend actually happens, the autofit/geometry set by the user
happens anyway so ignoring it in this case does no harm. Fixes #9381.
|
|
|
|
|
|
|
|
|
|
|
| |
In the reconfig event, the keepaspect option was checked before setting
the window_size geometry to the new params obtained from the vo. This is
incorrect. If a user disabled keepaspect on wayland, the video's size
would not change on a reconfigure event (i.e. loading a new video in the
playlist with a different size). No other windowing backend (x11, win32,
etc.) behaves like this or uses keepaspect in its code like wayland did
in this case. Clearly, this is not correct. Such functionality should be
handled by a separate option entirely. Just remove this if statement.
|
|
|
|
|
|
|
|
|
|
|
|
| |
The value of the border option should always match what the actual state
of the window is. Previously if a compositor rejected the request by
mpv, it did not correct itself. Also add some code to keep track of
decoration requests. Anytime the state is changed, make the last saved
request again (doesn't hurt and seems like intuitive behavior).
Unfortunately, this isn't foolproof since options only send callback if
the value is changed. (ex. on sway if the floating window has no border,
and then is titled, setting the border value to "yes" does nothing since
tiling the window already set the border value to "yes").
|
|
|
|
|
|
|
|
|
| |
In wayland_common, wl->window_size keeps track of the window's size when
it is not in the fullscreen or maximized state. GET_UNFS_WINDOW_SIZE is
meant to report the size except for when it is fullscreen (hence UNFS).
However, the actual function was merely returning the wl->window_size so
it was wrong for maximized windows. Workaround this by returning
wl->geometry instead when we have the maximized case. Fixes #9207.
|
|
|
|
|
| |
Set it to 24 if it couldn't be read frome $XCURSOR_SIZE for some reason.
Since it seems to be the default value for most distros/DE.
|
|
|
|
|
| |
Read $XCURSOR_THEME environment variable if set and if not then fall
back to "default" (/usr/share/icons/default and ~/.icons/default)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Normally in wayland, you receive a keymap event from the compositor
which is what allows the client to setup what is needed for xkb.
However, it turns out that this doesn't occur in the case of virtual
keyboards, such as from wtype, that come from the custom
virtual-keyboard protocol. What happens in this case is that mpv only
receives a keyboard entrance event. According to the wayland protocol
documentation [1], "the compositor must send wl_keyboard.modifiers event
after [the wl_keyboard.enter] event". It is possible for this to occur
before the physical keyboard is properly mapped (i.e: using a virtual
keyboard to start mpv). What this results in is a segfault once
xkb_state_update_mask is called in the modifiers event. The fix is to
simply not always assume we have created the xkb state if we get this
event and check for its existence first. Closes #9119.
https://wayland.freedesktop.org/docs/html/apa.html#protocol-spec-wl_keyboard
|
| |
|
|
|
|
|
|
| |
Right after the listener is constructed, request a mode based on the
value of the border option. If border, then request SSD. If no border,
request CSD. The decoration listener handles it from there.
|
|
|
|
|
| |
Oversight in 69c64f5. If there is no xdg_toplevel_decoration, don't
attempt to set the mode.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The current usage of the xdg-decoration protocol is quite a bit crude.
There's a specific listener struct for this like with most wayland
things, but mpv wasn't even using it. It also made no attempt to detect
if the compositor was telling mpv to use client side decorations. To
implement this correctly, first of all we need to create the decoration
listener object and add it like how most wayland things work. Secondly,
set_border_decorations needs to be changed to accept the uint32_t mode
given by the compositor. When we get the decoration event, pass the mode
obtained from the compositor to set_border_decorations. Update the mpv
border option, based on whether or not we have server side decorations
(yes if we have them; no if we don't). mpv can actually request either
client side or server side decorations with this protocol. This function
doesn't belong in set_border_decorations but rather it should be called
when the border option changes. Whether or not this request actually
works depends on the compositor (Plasma appears to allow it; sway does
not). If the compositor allows this request, then it gets handled by
configure_decorations. Note that the enum strangely starts at 1 hence
why there is an extra +1.
|
|
|
|
|
|
|
|
|
|
|
|
| |
So the resizing mechanism is actually supposed to match the video size
to the window size while preserving the aspect ratio. In wayland, the
current logic behaves as if --no-keepaspect-window was set. Fix this by
simply multiplying the height by the same scale factor the width is
multiplied by. Also get rid of the pointless (width > height) test (it
makes no difference in any case) as well as some unneccesary checks for
the keepaspect-window option. The use of ceil here is to make sure the
window coordinates can never possibly have be 0 due to truncation
(weston can still give you a 1x1 window which is fun). Fixes #9098.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There's currently some touch related code in mpv wayland, but clearly
nobody actually uses because it's a broken mess. Initially an attempt to
distinguish between two finger touches and one finger touch was made,
but there's not a good way to make this work. For whatever reason,
initiating either xdg_toplevel_resize or xdg_toplevel_move blocks any
other touch events from occurring (at least in plasma). Trying to call
these functions anywhere else is not really useful since the serial will
be invalid outside of the touch down events (well it would work in the
touch up event but that's just silly).
Anyways, let's just make this work sanely. Eliminate the touch entries
variable completely because it's pointless. Only one finger event is
ever considered at a time. Touches besides the initial one are all
ignored. If a user touches and drags within the touch edge radius, then
a resize event occurs. If the user touches and drags elsewhere on the
window, a move event occurs. A single tap displays the osc (which is
clickable if you tap again). A double tap toggles fullscreen.
Additionally, the default touch edge radius of 64 pixels is way too big
(at least I think so). Cut this in half to 32 which feels a lot better
(on a pinephone using plasma mobile anyway).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The way the window-scale option is supposed to behave wasn't really ever
documented until 25cfc81. It turns out that different OS/WMs may do
slightly different things. For wayland, we need to fix two things.
First, only return the wl->window_size in GET_UNFS (aka don't return the
fullscreen size). For SET_UNFS, we need to change the behavior for the
maximized window case. If the window is maximized, first attempt to
unmaximize it and send the wayland event. If the compositor allows this
to happen, then go ahead and set the new dimensions and resize it. In
the case that the attempt to unmaximize is not successful, then don't
attempt the resize and just save the window size dimensions for later to
be used when the user unmaximizes the window.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Regression from 24357cb. It's ugly but unfortunately keeping tracking of
the last toplevel width and height really is the best way to solve this
problem and removing it was a mistake. Compositors don't always send
width/height coordinates of the actual window. The easiest way to
trigger this is by changing window-scale/current-window-scale and then
unfocusing the window. The compositor will send another toplevel
configure event with coordinates of the window before the resize. Maybe
compositors could be smarter but multiple ones do this (sway, gnome,
plasma), so just keep the old workaround. The only difference this time
is that the toplevel width/height is saved at the very beginning which
also covers the case where it equals 0 (i.e. weston).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The source of many geometry woes. There's some loosely related toplevel
things that should be cleaned up/fixed. First of all,
VO_EVENT_LIVE_RESIZING is actually completely useless. It might have
been useful sometime in the past, but there's no point. It doesn't
"speed up" resizing in any way and appears to be originally for cocoa.
Just remove it.
Way back in the day, toplevel_width/height was added as a workaround for
when we got uncoorperative (i.e. wrong) width/height coordinates from
the compositor in this event. Basically it could happen due to numerous
reasons but a lack of atomic commits was part of the reason and also
mpv's geometry handling then was a lot rougher. We *shouldn't* need this
workaround anymore. The width/height values are only used exactly when
we need them. If mpv sets geometry on its own, it should still be the
right dimensions.
Related to the above, mpv never actually propertly handled the case
where width or height was equal to 0. According to the xdg-shell spec,
"If the width or height arguments are zero, it means the client should
decided its own window dimension." An example of a compositor doing this
is weston. It's, unsurprisingly, broken. Getting out of fullscreen or a
maximized state does not restore the old window size like it should. The
right way to handle this is to just return near the end of the function
if we have a 0 for either argument and before any geometry is set
(wl->geometry's width or height can never be zero). Luckily, state
changes are already being detected so they just trigger the goto when
needed.
Finally, e2c24ad mistakenly removed the VO_EVENT_EXPOSE. There are edge
cases where this is needed and it's safer to just force a redraw here
when the window gets activated again. Just force wl->hidden to false
first and then trigger the expose.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
More wayland weirdness. So previously, flipping a hidden state from true
to false was done in vo_wayland_wait_frame. In theory, this would be
after you get the frame callback and all those events so there's no
problem. However since the function also does a bunch of
flushing/dispatching/etc. to the default display queue so a lot of
unknown things can happen before we actually set the hidden variable
back to false. For example if a single image was paused and left on
another virtual desktop long enough (~5 minutes) while also not having
focus, switching back to that desktop could render it a black frame.
This edge case was supposed to be handled by the surface being activated
again in the toplevel event but apparently that doesn't always work. The
fix is to just delete all of that junk and set wl->hidden = false in the
frame callback. What's actually happening is kind of a mystery honestly.
Probably the compositor drops the buffers after a while as an
optimization (sensible) and forces a repaint if you switch back to the
virtual desktop. Somehow wl->hidden not being set to false would not
properly trigger a repaint (likely because it also sends a toplevel
event which does stuff) thus you just get a black window. If you just
make sure to set hidden in the frame callback, it appears like all of
these problems and edge cases are solved. Since this event must happen
first, that makes sense. That simplifies a lot of stuff and fixes some
subtle bugs at the same time so just go with this approach.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Another day, another wayland refactor. Way back when, dcc3c2e added
support for the hidpi-window-scale option (something you probably should
never set to no but whatever) to wayland. Well technically, it never had
any runtime toggling support (don't remember if detecting when vo_opts
changed was possible or not then; maybe not). Anyways in the process of
fixing that, I went ahead and refactored how this is all handled. The
key difference is that when hidpi-window-scale is disabled, wl->scaling
is directly set to 1 instead of forcibly setting
wl->current_output->scale to 1. Note that scaling operations don't
always require a geometry reset/resize so set_surface_scaling needs to
be separate from set_geometry. The logic here is kind of complicated but
it (should) be correct.
|
|
|
|
|
|
|
|
|
|
|
| |
The wl_surface lives for the entire lifetime of the vo. It's only
neccesary to set the scale initially and when the output scaling changes
(the surface moves to a different output with a different scale or the
output itself changes it scale). All of the calls that were being made
in the egl/vulkan resize functions are not needed. vo_wlshm wasn't
correctly rescaling itself before this commit since it had no logic to
handle scale changes. This should all be shared, common code in the
surface/output listeners.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A subtle regression from c26d833. On sway if mpv was set to be a
floating window in the config, set_buffer_scale would actually get
applied twice according to the wayland log. That meant a 1920x1080
window would appear as a 960x540 window if the scale of the wl_output
was set to 2. This only affected egl on sway (didn't occur on weston and
was too lazy to try anything else; probably they were fine). Since
wl->render is initially false, that meant that the very first run
through the render loop returns false. This probably caused something
weird to happen with the set_buffer_scale calls (the egl window gets
created and everything but mpv doesn't write to it just yet) which makes
the set_buffer_scale call happen an extra time. Since it was always
intended for mpv to initally render, this is worth fixing. Just chnage
wl->render to wl->hidden (again) and flip the bools around. That way,
the initial false value results in render == true and mpv tries to draw
on the first pass. This fixes the weird scaling behavior because
reasons.
|
|
|
|
|
|
| |
Not sure what I was on when I wrote this. wayland-app-id is supposed to
default to "mpv". Just set that in the vo_sub_opts and don't do this
weird m_config_cache_write_opt thing. Also make the doc entry nicer.
|
|
|
|
|
|
|
| |
Mostly a cosmetic change that (hopefully) makes things look better. Some
functions and structs that were previously being exported in the wayland
header were made static to the wayland_common.c file (these shouldn't be
accessed by anyone else).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This was originally just a bugfix for a race condition, but the scope
expanded a bit. Currently, the wayland code does a prepare_read ->
dispatch_pending -> display_flush -> read_events -> dispatch_pending
routine that's basically straight from the wayland client API
documentation. This essentially just queues up all the wayland events
mpv has and dispatches them to the compositor. We do this for blocking
purposes on every frame we render. A very similar thing is done for
wait_events from the VO.
This code can pretty easily be unified and split off into a separate
function, vo_wayland_dispatch_events. vo_wayland_wait_frame will call
this function in a while loop (which will break either on timeout or if
we receive frame callback from the compositor). wait_events needs to
just check this in case we get some state change on the wakeup_pipe
(i.e. waking up from the paused state).
As for the actual bugfix part of this, it's a slight regression from
c26d833. The toplevel config event always forced a redraw when a surface
became activated again. This is for something like displaying cover art
on a music file. If the window was originally out of view and then later
brought back into focus, no picture would be rendered (i.e. the window
is just black). That's because something like cover art is just 1 frame
and the VO stops doing any other additional rendering. If you miss that
1 frame, nothing would show up ever again. The fix in this case is to
always just force a redraw when the mpv window comes back into view.
Well with the aforementioned commit, we stopped doing
wl_display_roundtrip calls on every frame. That means we no longer do
roundtrip blocking calls. We just be sure to queue up all of the events
we have and then dispatch them. Because wayland is fundamentally an
asynchronous protocol, there's no guarantee what order these events
would be processed in. This meant that on occasion, a
vo_wayland_wait_frame call (this could occur multiple times depending on
the exact situation) would occur before the compositor would send back
frame callback. That would result in the aforementioned bug of having
just a black window. The fix, in this case, is to just do a
vo_wayland_wait_frame call directly before we force the VO to do a
redraw. Note that merely dispatching events isn't enough because we
specifically need to wait for the compositor to give us frame callback
before doing a new render.
P.S. fix a typo too.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Take two. f4e89dd went wrong by moving vo_wayland_wait_frame before
start_frame was called. Whether or not this matters depends on the
compositor, but some weird things can happen. Basically, it's a
scheduling issue. vo_wayland_wait_frame queues all events and sends them
to the server to process (with no blocking if presentation time is
available). If mpv changes state while rendering (and this function is
called before every frame is drawn), then that event also gets
dispatched and sent to the compositor. This, in some cases, can cause
some funny behavior because the next frame gets attached to the surface
while the old buffer is getting released. It's safer to call this
function after the swap already happens and well before mpv calls its
next draw. There's no weird scheduling of events, and the compositor log
is more normal.
The second part of this is to fix some stuttering issues. This is mostly
just conjecture, but probably what was happening was this thing called
"composition". The easiest way to see this is to play a video on the
default audio sync mode (probably easiest to see on a typical 23.976
video). Have that in a window and float it over firefox (floating
windows are bloat on a tiling wm anyway). Then in firefox, do some short
bursts of smooth scrolling (likely uses egl). Some stutter in video
rendering could be observed, particularly in panning shots.
Compositors are supposed to prevent tearing so what likely was happening
was that the compositor was simply holding the buffer a wee bit longer
to make sure it happened in sync with the smooth scrolling. Because the
mpv code waits precisely on presentation time, the loop would timeout on
occasion instead of receiving the frame callback. This would then lead
to a skipped frame when rendering and thus causing stuttering.
The fix is simple: just only count consecutive timeouts as not receiving
frame callback. If a compositor holds the mpv buffer slightly longer to
avoid tearing, then we will definitely receive frame callback on the
next round of the render loop. This logic also appears to be sound for
plasma (funfact: Plasma always returns frame callback even when the
window is hidden. Not sure what's up with that, but luckily it doesn't
matter to us.), so get rid of the goofy 1/vblank_time thing and just
keep it a simple > 1 check.
|
|
|
|
| |
The display scaling might change here, so we need to signal mpv's core.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is actually a very nice simplification that should have been
thought of years ago (sue me). In a nutshell, the story with the
wayland code is that the frame callback and swap buffer behavior doesn't
fit very well with mpv's rendering loop. It's been refactored/changed
quite a few times over the years and works well enough but things could
be better. The current iteration works with an external swapchain to
check if we have frame callback before deciding whether or not to
render. This logic was implemented in both egl and vulkan.
This does have its warts however. There's some hidden state detection
logic which works but is kind of ugly. Since wayland doesn't allow
clients to know if they are actually visible (questionable but
whatever), you can just reasonably assume that if a bunch of callbacks
are missed in a row, you're probably not visible. That's fine, but it is
indeed less than ideal since the threshold is basically entirely
arbitrary and mpv does do a few wasteful renders before it decides that
the window is actually hidden.
The biggest urk in the vo_wayland_wait_frame is the use of
wl_display_roundtrip. Wayland developers would probably be offended by
the way mpv abuses that function, but essentially it was a way to have
semi-blocking behavior needed for display-resample to work. Since the
swap interval must be 0 on wayland (otherwise it will block the entire
player's rendering loop), we need some other way to wait on vsync. The
idea here was to dispatch and poll a bunch of wayland events, wait (with
a timeout) until we get frame callback, and then wait for the compositor
to process it. That pretty much perfectly waits on vsync and lets us
keep all the good timings and all that jazz that we want for mpv. The
problem is that wl_display_roundtrip is conceptually a bad function. It
can internally call wl_display_dispatch which in certain instances,
empty event queue, will block forever. Now strictly speaking, this
probably will never, ever happen (once I was able to to trigger it by
hardcoding an error into a compositor), but ideally
vo_wayland_wait_frame should never infinitely block and stall the
player. Unfortunately, removing that function always lead to problems
with timings and unsteady vsync intervals so it survived many refactors.
Until now, of course. In wayland, the ideal is to never do wasteful
rendering (i.e. don't render if the window isn't visible). Instead of
wrestling around with hidden states and possible missed vblanks, let's
rearrange the wayland rendering logic so we only ever draw a frame when
the frame callback is returned to use (within a reasonable timeout to
avoid blocking forever).
This slight rearrangement of the wait allows for several simplifications
to be made. Namely, wl_display_roundtrip stops being needed. Instead, we
can rely entirely on totally nonblocking calls (dispatch_pending, flush,
and so on). We still need to poll the fd here to actually get the frame
callback event from the compositor, but there's no longer any reason to
do extra waiting. As soon as we get the callback, we immediately draw.
This works quite well and has stable vsync (display-resample and audio).
Additionally, all of the logic about hidden states is no longer needed.
If vo_wayland_wait_frame times out, it's okay to assume immediately that
the window is not visible and skip rendering.
Unfortunately, there's one limitation on this new approach. It will only
work correctly if the compositor implements presentation time. That
means a reduced version of the old way still has to be carried around in
vo_wayland_wait_frame. So if the compositor has no presentation time,
then we are forced to use wl_display_roundtrip and juggle some funny
assumptions about whether or not the window is hidden or not. Plasma is
the only real notable compositor without presentation time at this stage
so perhaps this "legacy" mechanism |