As found out by @philipl, failing to pass this from the VkInstance to
the VkDevice is bad style. We might want to override the get_proc_addr
pointer in the future.
20c4036daa switched to hiding all symbols
by default and exporting the ones we want with MPV_EXPORT.
However, the mpv_stream_cb_add_ro function was missed and it is part of
the client API. Export it too. Fixes#9971.
Our logging here today is very poor. We don't make it clear what
formats we are probing, or even that a certain format failed in most
cases. In the case where we do log the error, we don't make it clear
which format it was that failed.
The end result is that we have no idea what the possible and final
format spaces are, which makes it very hard to debug whether things are
working correctly, or to work on supporting additional formats.
The documentation's current discussion of hwdec usage is out of date,
and unnecessarily pessemistic when applied to modern hardware. The
reality is that modern APIs on modern hardware produce reasonable
results and there's no need to pretend otherwise. The current language
that tries to drive people away from using hwdecs at all leads to them
making bad choices when they do try to use it.
Let's also make it clearer that users should use vo=gpu with hwdecs
rather than vo=vaapi or vo=vdpau. Even the existing admonitions have
proven insufficient, so let's strengthen that language.
This code fails if we have DR buffers, but none of them correspond to
the current frame. Normally only happens if e.g. changing the decoder at
runtime, since DR buffers are not properly reinit in that case.
(Arguably a separate bug)
The current map_frame() code fails to clean up after itself on the
failure paths. But if map_frame returns false, no cleanup code is ever
attempted. Add the relevant calls to clean up state manually,
throughout.
So, turns out the approach in 7f67a553f doesn't work for all codecs. In
particular, sometimes lavc will internally allocate a new AVBuffer that
just points at the old AVBuffer but has a different opaque field for
some reason. In these cases, the DR metadata doesn't survive the
round-trip through libavcodec.
I explored several alternative ways of solving this problem, including
adding new mp_image fields, but in the end none of them survived the
round-trip through AVFrame and back. The `priv` and `opaque` fields
in respectively `mp_image` and `AVFrame` are also too heavily overloaded
to be of much help.
In the end, bite the bullet and use the same approach as done in
`vo_gpu`, which is to just keep track of a list of all allocations. This
is a really ugly way of doing things IMO, but ultimately, completely
safe.
Previously, when mpv was invoked with unsupported hwdec value, such
as --hwdec=foobar, there was no indication that it doesn't exist.
The reason it's not validated during options parsing is that the name
is only evaluated when selecting hwdec for actual decoding, by
matching it against runtime list of names from ffmpeg.
Additionally, when selecting hwdec for decoding, matching the name
came after filtering by codec, hence overall never-matched-name did
not necessarily indicate it's unsupported (doesn't exist at all).
Now we check the name before filtering by codec, and when done,
warn if no hwdec with that name exists at all.
This means that an unsupported name will now generate such warning
whenever we try to choose a hwdec, i.e. possibly more than once.
It's much better than no notification at all, and arguably adequate
for a sort of configuration error (linked ffmpeg has no such hwdec
name) which we don't validate during option parsing.
Some decoders, notably hevcdec, will unconditionally memset() the entire
AVBufferRef based on the AVBufferRef size. This is bad news for us,
since it also overwrites our `struct dr_buf`.
Rewrite this code to make it more robust - keep track of the DR buf
metadata in a separate allocation instead. Has the unfortunate downside
of technically being undefined behavior if `opaque` is not at least 64
bits in size, though, but avoids this issue.
Maybe there's a better way for us to unconditionally keep track of DR
allocation metadata. I could try adding it into the `mp_image` itself.
Maybe on a rainy day. For now, this works.
Fixes#9949
Might fix#9526
Previously only status<0 was considered as error, but status>0 is
also an error (the process exit code). Change to status != 0.
This likely makes little to no difference in practice, because if
stdout is empty or can't be parsed as JSON then it's considered
an error anyway, but still, this is more correct.
Also, on error, add the complete subprocess result to the verbose log.
There are two major ways of going about this:
1. Expose the native ra_gl/ra_pl/ra_d3d11 objects to the pre-existing
hwdec mappers, and then add code in vo_gpu_next to rewrap those
ra_tex objects into pl_tex.
2. Wrap the underlying pl_opengl/pl_d3d11 into a ra_pl object and expose
it to the hwdec mappers, then directly use the resulting pl_tex.
I ultimately opted for approach 1 because it enables compatibility with
more hardware decoders, specifically including ones that use native
OpenGL calls currently. The second approach only really works with
cuda_vk and vaapi_pl.
I was initially hesitant to link to an external article from the manpage
but a lot of other places in the man page already do something like
this, and it really makes more sense to have this as a wiki article
rather than trying to keep the manpage up-to-date by hand, since this
evolves quickly and the two are expected to converge over time.
This avoids decoding/caching more frames in advance than necessary. In
particular, this is very important for hwdec, which generally can't have
too many decoded frames in a pool at the same time.
So I can reuse it in vo_gpu_next without having to reinvent the wheel.
In theory, a lot of the stuff could be made more private inside the
hwdec code itself, but for the time being I don't care about refactoring
this code, merely sharing it.
We don't need to hold on to buffers longer than necessary. Doesn't
matter for vo_gpu but greatly matters for vo_gpu_next, since it persists
hwdec mapped textures for longer periods.
Unfortunately, only provides benefits for hwdecs which do explicit
copies in their decode path, which currently just means cuda and
d3d11va.
There's really nothing vulkan-specific about this hwdec wrapper, and it
actually works perfectly fine with an OpenGL-based ra_pl. This is not
hugely important at the time, but I still think it makes sense in case
we ever decide to make vo_gpu_next wrap OpenGL contexts to ra_pl instead
of exposing the underlying ra_gl.
Easiest way is by just using a designated struct initializer. If we
don't, `p->images` ends up inheriting random data, which leaks into e.g.
eglDestroyImageKHR.
It's a small miracle this never blew up before. Or maybe it did. Who
knows.
A few years ago, in 25e70f4743, we
disabled the vavpp deinterlacing auto-filter on the basis that it
caused crashes on _some_ hardware with _some_ driver version(s). But
since then, the situation has improved. There is still a limitation
where you can't turn deinterlacing on on the fly with AMD, but it
doesn't crash anymore (That is #7388).
So, given that AMD users have to set up the deinterlacing filter
manually either way, let's re-add the auto-filter for Intel users.
Some youtube_dl extractors retrieve URLs which contain other URLs
inside of them, for example Funimation, like this:
https://example.com/video?parameter=https://example.net/something
The url_is_safe function uses a pattern to match the protocol at the
start of the URL. Before this commit, this pattern was not compliant
with the URL spec (see the definition of "A URL-scheme string"):
https://url.spec.whatwg.org/#url-writing
Therefore it would match any characters, including "://", until the
last occurence of "://" in the string. Thus the above URL would match
https://example.com/video?parameter=https
which is not in safe_protos so the video will not play.
Now the protocol can only start with a letter and only contain
alphanumerics, "." "+" or "-" as the spec says, so it will only match
the first protocol in the URL ("https" in the above example.)
Previously the URL also had to contain "//" after the ":". Data URLs
do not contain "//": https://datatracker.ietf.org/doc/html/rfc2397
so now the pattern does not look for "//", only ":".
libplacebo v198 fixed this properly by adding the ability to flip planes
directly, which is done automatically by the swapchain helpers.
As such, we no longer need to concern ourselves with hacky logic to flip
planes using the crop. This also removes the need for the OSD coordinate
hack on OpenGL.
These helpers, for some reason, decided to round the returned values up
to multiples of the nearest plane alignment. This logic makes no sense
to me, and completely breaks any sort of oddly-sized mp_image.
This logic was introduced, presumably in error and without real
justification, as part of a major refactor commit (caee8748), As far as
I can tell, removing it again doesn't regress anything.
Fixes several serious bugs including buffer underflows and GPU crashes
in vo_gpu and vo_gpu_next.
Page 2 displays the frame timing info for each type (currently "Fresh"
and "Redraw"), but since they're unordered keys of a map, their
iteration order and hence their order on screen could (and did) change
depending on $things, which was annoying.
haasn thinks none of them should be considered more important (and
therefore maybe worthy of being displayed on top), so we go with
lexicographic order, which currently means that "Fresh" is first.
Fixes#9901
This fixes two potential divisions by 0 at generate_graph(...):
- If v_avg is (given and) 0.
- if v_max is 0.
The former doesn't seem to happen in practice because v_avg is only
used at one generate_gpah call, where it's apparently not 0.
The latter triggers if all the graph values are 0 (hence v_max is
also 0).
The implication of these divisions by 0 is an invalid y-value which
ends up at the ASS coordinates string for the graph inner content.
On linux the value ends as "nan" (luajit) or "-nan" (lua 5.1/5.2), and
on Windows it's "nan" (luajit) or "-1.#IND00" (lua 5.1/5.2), maybe due
to msvcrt's snprintf.
All these strings are wrong as ASS numbers, but due to luck in how
libass parses the coordinates, "nan" and "-nan" result in empty graph
(which looks OK for "all values are 0"), while "-1.#IND00" is parsed
as -1, which breaks the graph rendering (affects other graphs too).
One example of "all values are 0" is at page 0 (internal performance
graphs) on Windows - because the cpu metrics don't work.
So this fixes at least page 0 on Windows with lua 5.1/5.2.
While at it, move the scale calculations to one place, which now
avoids division by 0, instead of duplicating this calculation.
In the future, we can consider improving the generate_graph API:
- It needs one peak value, but takes 3 (v_max, v_avg, scale) which
are meshed into one final value.
- v_avg is only used in 1 of 6 call sites, but at the middle of the
arguments, so all other call sites need to pass explicit "nil".
- "scale" is arbitrary and used to leave some space at the top of the
graph. 5 places use 0.8, one uses 0.9. Could probably be unified.
Render subs at the output resolution, rather than the video resolution.
Uses the new APIs found in libplacebo 197+, to allow controlling the OSD
resolution even for image-attached overlays.
Also fixes an issue where the overlay state did not get correctly
updated while paused. To avoid regenerating the OSD / flushing the cache
constantly, we keep track of OSD changes and only regenerate the OSD
when the OSD state is expected to change in some way (e.g. resolution
change). This requires introducing a new VOCTRL to inform the VO when
the UPDATE_OSD-tagged options have changed.
Fixes#9744, #9524, #9399 and #9398.
The stop-screensaver option is currently limited to a simple yes/no
option. While the no option does always disable mpv trying to stop the
screensaver, yes does not mean the screensaver is always stopped. The
screensaver will be enabled again depending on certain conditions (like
if the player is paused). Simply introduce a new value for this option,
always, which does exactly what the name implies: the screensaver will
always be disabled.
- Make it clearer that playback_only affects subprocess' behavior when
the playback of the current playlist entry terminates, rather than
when mpv quits.
- Explain when status is positive and when it is negative.
- Replace "exited gracefully" in status' and error_string's
documentation with "terminated normally" so it can't be misinterpreted
as exiting successfully.
- Reword the playback_only warning
Historically, we have treated hwdec interop loading as a completely
separate step from loading the hwdecs themselves. Some hwdecs need an
interop, and some don't, and users generally configure the exact
hwdec they want, so interops that aren't relevant for that hwdec
shouldn't be loaded to save time and avoid warning/error spam.
The basic approach here is to recognise that interops are tied to
hwdecs by imgfmt. The hwdec outputs some format, and an interop is
needed to get that format to the vo without read back.
So, when we try to load an hwdec, instead of just blindly loading all
interops as we do today, let's pass the imgfmt in and only load
interops that work for that format. If more than one interop is
available for the format, the existing logic (whatever it is) will
continue to be used to pick one.
We also have one callsite in filters where we seem to pre-emptively
load all the interops. It's probably possible to trace down a specific
format but for now I'm just letting it keep loading all of them; it's
no worse than before.
You may notice there is no documentation update - and that's because
the current docs say that when the interop mode is `auto`, the interop
is loaded on demand. So reality now reflects the docs. How nice.
Pass channel volumes to `pw_stream_set_control` as array.
This is correct calling conventions and prevents
right channel muting every time ao-volume property is changed.
Terminate `pw_stream_set_control` calls with 0.
This is an annoying special case only really needed because of the
`vflip` filter. mpv handles this by directly adjusting the plane
transform. The libplacebo API, unfortunately, does not allow passing the
required information for this to work smoothly.
Long-term I plan on adding support for plane flipping in libplacebo
directly, but for the meantime, we will have to work-around it by moving
the flipping to the whole-image `crop` instead. Not an ideal solution
but better than crashing.
Fixes#9855
We're getting bug reports that the recent change to add extra usage
flags to swapchain buffers (for gpu-next) breaks mpv on some Windows 7
systems, and it seems like this is only happening with flip-model
swapchains.
Creating swapchains with DXGI_USAGE_UNORDERED_ACCESS should be valid. At
least, it's not specifically disallowed, unlike some other flags[1]. So,
just disable it for flip-model swapchains in Windows 7, rather than
disabling it everywhere.
[1]: https://docs.microsoft.com/en-us/windows/win32/direct3ddxgi/dxgi-usage
See the previous commit for the full explanation. Basically, luajit 2.0
has a bad pc file on macos that causes libmpv to fail during build. The
workaround was, if the os was darwin and luajit was found, to save a
full luajit dep and a partial luajit dep with the link args removed. The
partial dep was used for compiling libmpv, and the full dep was used for
the actual mpv executable. This worked and was needed for the CI to pass
but it sucked. Since the previous commit now makes the CI grab lua 5.1,
we don't need all this crap anymore. Just delete it and treat the
dependency normally.
This does effectively mean that building libmpv with luajit 2.0 on macOS
will no longer work with the meson build. However libraries not being
built correctly is not a mpv-specific issue. The waf build will succeed
for some reason, but it has known issues and it would be better if it
just failed honestly. An upstream developer said years ago that that
macOS users should use the 2.1 branch (and there's no release of
course). In any case, no macOS user should be building mpv with luajit
2.0, so we shouldn't be going out of our way to support this.
https://github.com/mpv-player/mpv/issues/7512https://github.com/LuaJIT/LuaJIT/issues/521#issuecomment-562999247
LuaJIT is still actively developed, but upstream is allergic to making
new releases for whatever reason. The last tagged release was in May of
2017, so we probably shouldn't expect a new release anytime soon. Now
for mpv, this doesn't really matter except in the case of macOS where
2.0.5 is actually a bit broken (and of course the CI uses luajit). More
specifically, the 2.0.5 pc is broken and has a "-pagezero_size 10000"
flag which causes libmpv to fail (only executables are allowed to use
this). This magically works on waf. It's possible that it just happens
to ignore the link arguments. However on the meson build, this is broken
and led to a really ugly hack using a partial dependency so both mpv and
libmpv succeed. Fortunately, the 2.1 luajit branch fixes this.
Unfortunately, there's no actual release.
Instead, just use Lua 5.1. Note that lua 5.1 is technically deprecated
in homebrew, but the chances of this going away is pretty slim since
everyone knows that new lua versions are not backwards compatible.
Anyways, using 5.1 works fine and lets us get rid of a terrible hack in
the meson build. People really shouldn't be using 2.0 LuaJIT anyway.
We have a few examples with vf=rotate=90, but libavfilter's rotate
accepts radians - not degrees, so a value of 90 is valid but misleading
because it's not 90 degrees.
Change the value of 90 at those examples.
Also replace "vf-add=hflip" with "vf add hflip", since that example
lists commands, not options, to run.
Drag and drop in wayland is weird and it seems everyone does this
slightly differently (fun). In the past, there was a crash that
occured (fixed by 700f4ef5fa) which
involved using the wl_data_offer_finish in an incorrect way that
triggered a protocol error (always fatal). The fix involved moving the
function call to data_device_handle_drop which seemingly works, but it
has an unfortunate side effect. It appears like GTK applications (or at
least firefox) close the pipe after this function is called which makes
it impossible for mpv to read data from the fd (well you could force it
open again in theory but let's not do that). Who knows if that was the
case when that commit was made (probably not because I'd think I would
have noticed; could just be a dummy though), but obviously having broken
dnd for a major application isn't so fun (this works with QT and
chromium anyway).
Ideally one would just simply check the pipe in data_device_handle_drop,
but this doesn't work because it doesn't seem the compositor actually
sends mpv the data by then. There's not actually a defined event when
you're supposed to be able to read data from this pipe, so we wait for
the usual event checking loop later for this. In that case,
wl_data_offer_finish needs to go back into check_dnd_fd, but we have to
be careful when calling it otherwise we'd just commit protocol errors
like before. We check to make sure we even have a valid wl->dnd_offer
before trying to indicate that it is finished and additionally make sure
there is a valid dnd_action (after checking the fd, it's always set back
to -1).
This doesn't fix everything though. Specifically, sway with
focus_follows_mouse (the default) and GTK as the data source still
doesn't work. The reason is that when you do a drag and drop in sway
with that option on, a new wl_data_device.data_offer event is sent out
instantly after the drop event. This happens before any data is sent
across the fd and before mpv even has a chance to check it. What GTK
does, when getting this new data_offer event, is close the pipe
(POLLHUP). This means mpv can't read it when we reach the event loop and
thus no data is ever read and broken drag and drop. From the client
side, this isn't really fixable since the wayland protocol doesn't have
a clear indication of when clients are supposed to read from the fd and
both the compositor and data source are doing things totally out of our
control. So we'll consider this weird case, "not our bug" at least. The
rest should work.
By receiving the graphics API-specific contexts from the generic
pl_gpu instance, it is not necessary to store them separately.
As the current priv struct stores nothing else, this allows its
removal. This removal cleans up related compiler warnings regarding
unused variables in case of opengl and d3d11 being disabled in
mpv/libplacebo.
Functions to receive the underlying vulkan,gl,d3d11 structure were
added in libplacebo version 4.182. Our current requirement for
gpu-next is 4.190, and thus we can freely utilize these helper
functions.
We only wish to touch actual video frames, which should have an
allocated image attached to them, so just check the frame type
early, and exit by passing through such non-video frames to further
filters in the chain without attempting to process them.
Fixes a crash in case of non-video (EOF/NONE) frames being passed
onto the filter when the dovi option was set to false since
05ccc51d53 .
While testing support for the exotic formats used by Intel vaapi for
4:2:2 and 4:4:4 surfaces, I realised that we were enumerating all
endpoints and checking formats for them. The problem with this is
that decoding (VLD) endpoints are only a subset of what vaapi exposes.
All the encoding endpoints are there too, and there is also the
None profile that we don't care about, but which generates ffmpeg
warnings if you try and examine it.
So, let's only look at VLD endpoints. This will speed things up a
little bit and make the logging less noisy.
This speeds up chapter name access while hovering the seekbar:
- Fetch the chapter-list property using observation rather than on
every render while hovering the bar. Property access, especially
with potentially lots of values (chapters), can be expensive-ish,
and can involve locking, so the less per-render the better.
- Sort the list once (on change) to simplify the iteration logic when
searching the chapter which matches the mouse position. It's still
O(N) (no binary search currently), but with less work.
The cached sorted list is stored at the state. While there are other
places which access this property and could use this cache, for now we
don't touch them, because they're not performance-critical (on init
for the chapter markers, on ch_{prev,next} button click for OSD).
Caching properties using observation instead of using mp.get_property
can indeed speedup other places too, but it should be part of a system
rather than implemented per-property. Maybe some day.
When utils.get_user_path was added, the expand-path command didn't
exist. Now it does, so remove the C code, make it a trivial wrapper.
Keep this function for backward compat to not break scripts, but
technically it's not required anymore.
Niklas Haas
LaserEyess
Dudemanguy
Philip Langdale
Avi Halachmi (:avih)
James Ross-Gowan
George Brooke
Guido Cella
Alex B
Thomas Weißschuh
Jan Ekström
sfan5