The decoder initially read from the socket, decoded the video and sent
the decoded frames to the screen:
+---------+ +----------+
socket ---> | decoder | ---> | screen |
+---------+ +----------+
The design was simple, but the decoder had several responsabilities.
Then we added the recording feature, so we added a recorder, which
reused the packets received from the socket managed by the decoder:
+----------+
---> | screen |
+---------+ / +----------+
socket ---> | decoder | ----
+---------+ \ +----------+
---> | recorder |
+----------+
This lack of separation of concerns now have concrete implications: we
could not (properly) disable the decoder/display to only record the
video.
Therefore, split the decoder to extract the stream:
+----------+ +----------+
---> | decoder | ---> | screen |
+---------+ / +----------+ +----------+
socket ---> | stream | ----
+---------+ \ +----------+
---> | recorder |
+----------+
This will allow to record the stream without decoding the video.
The order of cleanup was not the reverse as the initialization order. As
a consequence, recorder_destroy() could theoretically be called even if
recorder_init() failed.
The deprecated avcodec_decode_video2() should always the whole packet,
so there is no need to loop (cf doc/examples/demuxing_decoding.c in
FFmpeg).
This hack changed the packet size and data pointer. This broke recording
which used the same packet.
The extradata buffer is owned by libav, so it must be allocated with
av_malloc(), not SDL_malloc().
This fixes a crash on Windows during avformat_free_context().
Windows does not support UTF-8, so pushing a file with non-ASCII
characters failed.
Convert the UTF-8 command line to a wide characters string and call
CreateProcessW().
Fixes <https://github.com/Genymobile/scrcpy/issues/422>
Implement recording to Matroska files.
The format to use is determined by the option -F/--record-format if set,
or by the file extension (".mp4" or ".mkv").
Some containers force their own time base. For example, matroska
overwrite time_base to (AVRational) {1, 1000}.
Therefore, rescale our packet timestamps to the output stream time base.
Suggested-by: Steve Lhomme <robux4@ycbcr.xyz>
It is very convenient when I play mobile game and watch video at the
same time.
Tested on Linux mint Cinnamon as well as Windows 10.
Signed-off-by: Yu-Chen Lin <npes87184@gmail.com>
To clean up the device, the client executed "adb shell rm" once the
server was guaranteed to be started (after the connection succeeded).
This implied to track whether the installation state, and failed if an
additional tunnel was used in "forward" mode:
<https://github.com/Genymobile/scrcpy/issues/386#issuecomment-453936034>
Instead, make the server unlink itself on start.
Mouse events position were unsigned (so negative values could not be
handled properly).
To avoid issues with negative values, mouse events outside the device
screen were ignored (commit a7fe9ad779).
But as a consequence, drag&drop were "broken" if the "drop" occurred
outside the device screen.
Instead, use signed 32-bits to store the position, and forward events
outside the device screen.
Fixes <https://github.com/Genymobile/scrcpy/issues/357>.
Never create a "struct point" with a position possibly outside the
device screen (i.e. in the black borders area), and do not transmit such
events.
This fixes an assertion failure on mouse wheel events outside the device
screen area.
The client passes parameters to the server via "adb shell" arguments.
Use "-" instead of "" when no crop is specified to avoid empty
arguments, which are not handled the same way on all devices.
Fixed <https://github.com/Genymobile/scrcpy/issues/337>.
Configuration packets produced by MediaCodec have no valid PTS, and do
not produce frame. Do not queue their (invalid) PTS not to break the
matching between frames and their PTS.
Several frames may be read by read_packet() before they are consumed
(returned by av_read_frame()), so we need to store the PTS of frames in
order, so that the right PTS is assigned to the right frame.
AVStream.codec has been deprecated in favor of AVStream.codecpar.
Due to the FFmpeg/Libav split, this happened in two separate versions:
- 57.33.100 for FFmpeg
- 57.5.0 for Libav
This partially reverts commit f00c6c5b13.
On Ctrl+C, we need to execute cleanup code. For instance, if recording
is enabled, we need to write MP4 file trailer on exit.
Custom SDL signal handlers were disabled because it leaded to process
hanging on Ctrl+C during network calls on initialization, but now it
seems to work correctly, the network calls return immediately on signal.
Since PTS handling has been fixed, the recorder do not associate a PTS
to a wrong frame anymore, so PTS of "configuration packets" (which never
produce a frame), are never read by the recorder. Therefore, there is no
need to ignore them explicitly, so we can remove the MediaCodec flags
completely.
The PTS was read from the socket and set as the current one even before
the frame was consumed, so it could be assigned to the previous frame
"in advance".
Store the PTS for the current frame and the last PTS read from the
packet header of the next frame in separate fields.
As a side-effect, this fixes the warning on quit:
> Application provided invalid, non monotonically increasing dts to
> muxer in stream 0: 17164020 >= 17164020
To handle special chars, text is handled as text input instead of key
events. However, this breaks the separation of DOWN and UP key events.
As a compromise, send letters and space as key events, to preserve
original DOWN/UP events, but send other text input events as text, to be
able to send "special" characters.
Fixes <https://github.com/Genymobile/scrcpy/issues/87>.
Suggested-by: pete1414
Suggested-by: King-Slide <kingslide@gmail.com>
The common command.c handled process errors from system-specific int
values (errno).
Rather, expose a new enum process_result to handle error cause in a
generic way.
There are many user who encounters missing adb.
To stop things happens again, we check it and show
sexy response to user.
Signed-off-by: yuchenlin <npes87184@gmail.com>
The current_process field was never reset after an installation is
complete. As a consequence, installer_stop() attempted to terminate it,
leading to a warning, at best.
In the Android input header file, an enum has a value taking more than
31 bits, leading to the following warning:
ISO C restricts enumerator values to range of ‘int’
Since we don't use it, remove it.
A missing initialization (fixed by the previous commit) leaded to kill
unexpected process.
In order to prevent consequences of similar errors in the future, never
call kill() with a non-positive PID.
See <https://github.com/Genymobile/scrcpy/issues/182>.
The current_process field of struct installer was not initialized.
Since the installer instance is static, its default value was 0.
The call to installer_stop() then called kill(0, SIGTERM) (on Linux),
which sent SIGTERM to every process in the process group. In particular,
the scrcpy process was killed.
As a consequence, the last cleanup steps, like disabling "show touches",
were not executed.
Fixes <https://github.com/Genymobile/scrcpy/issues/183>.
Meson decided to crossbuild for Windows as soon as
meson.is_cross_build() returned true. This made non-Windows crossbuilds
fail.
Instead, add an explicit option "crossbuild_windows".
Fixes <https://github.com/Genymobile/scrcpy/issues/165>.
On Windows, an application is either console or gui, it cannot be both.
Scrcpy should be both: it outputs important information to console, but
we still want to be able to ignore the console and launch it without a
visible cmd.exe window.
Therefore, build two binaries:
- scrcpy.exe
- scrcpy-noconsole.exe
Build the Windows binary from mingw on Linux, using the official
prebuilt binaries for ffmpeg, SDL2 and adb.
MSYS2 and all its packaged dll are not necessary anymore.
In practice, proc_show_touches may not be used uninitialized, since it
checks the flag options->show_touches, but the compiler can't know that,
so initialize it to avoid the warning.
I could not make Ctrl+'+' and Ctrl+'-' work for every keyboard on every
platform.
Instead, use Ctrl+UP and Ctrl+DOWN (like in VLC) to change the volume.
Fixes <https://github.com/Genymobile/scrcpy/issues/103>.
SDL_CreateTexture() is called both during initialization and on frame
size change.
To avoid inconsistent changes to arguments value, factorize them to a
single function create_texture().
Double-clicks were not sent to the device anymore since the
"double-click on black borders" feature.
When a double click occurs inside the device screen, send the event to
the device normally.
Fixes <https://github.com/Genymobile/scrcpy/issues/97>.
Enabling "show touches" involves the execution of an adb command, which
takes some time.
In order to parallelize, execute the command as soon as possible, but
reap the process only once everything is initialized.
If --show-touches is set, then the option must be disabled on quit.
Since it executes an adb command, it takes some time, so close the
window beforehand so that the close window button does not seem
unresponsive.
On H.264 stream EOF, the eof_reached flag is set, but av_read_frame()
still provides a frame, so check the flag only afterwards.
As a side-effect, it also fixes a memory leak (the very last packet was
not unref).
Request SDL not to replace the SIGINT and SIGTERM handlers, so that the
process is immediately terminated on Ctrl+C.
This avoids process hanging on Ctrl+C during network calls on
initialization.
Some of them accepted a timeout, but it was not used since
commit 9b056f5091 anymore.
On Windows and MacOS, resizing blocks the event loop, so resizing events
are not triggered:
- <https://bugzilla.libsdl.org/show_bug.cgi?id=2077>
- <https://stackoverflow.com/a/40693139/1987178>
As a workaround, register an event watcher to render the screen from
another thread.
Since the whole event loop is blocked during resizing, the screen
content is not refreshed (on Windows and MacOS) until resizing ends.
"adb reverse" currently does not work over tcpip (i.e. on a device
connected by "adb connect"):
<https://issuetracker.google.com/issues/37066218>
To work around the problem, if the call to "adb reverse" fails, then
fallback to "adb forward", and reverse the client/server roles.
Keep the "adb reverse" mode as the default because it does not involve
connection retries: when using "adb forward", the client must try to
connect successively until the server listens.
Due to the tunnel, every connect() will succeed, so the client must
attempt to read() to detect a connection failure. For this purpose, when
using the "adb forward" mode, the server initially writes a dummy byte,
read by the client.
Fixes <https://github.com/Genymobile/scrcpy/issues/5>.
The serial is needed for many server actions, but this is an
implementation detail, so the caller should not have to provide it on
every call.
Instead, store the serial in the server instance on server_start().
This paves the way to implement the "adb forward" fallback properly.
The SDL mouse wheel event seems inconsistent between horizontal and
vertical scrolling.
> Movements to the left generate negative x values and to the right
> generate positive x values. Movements down (scroll backward) generate
> negative y values and up (scroll forward) generate positive y values.
<https://wiki.libsdl.org/SDL_MouseWheelEvent#Remarks>
Reverse the horizontal.
Fixes <https://github.com/Genymobile/scrcpy/issues/49>.
The text input control_event was initially designed for mapping
SDL_TextInputEvent, limited to 32 characters.
For simplicity, the copy/paste feature was implemented using the same
control_event: it just sends the text to paste.
However, the pasted text might have a length breaking some assumptions:
- on the client, the event max-size was smaller than the text
max-length,
- on the server, the raw buffer storing the events was smaller than the
max event size.
Fix these inconsistencies, and encode the length on 2 bytes, to accept
more than 256 characters.
Fixes <https://github.com/Genymobile/scrcpy/issues/10>.
Paste computer clipboard to the device on Ctrl+v.
The other direction (pasting the device clipboard to the computer) is
not implemented. It would require a communication channel from the
device to the computer, other than the socket used by the video stream.
The High DPI support is enabled by default, so that the renderer use the
full definition of High DPI screens.
However, there are still mouse coordinates problems on some MacOS having
High DPI support (but not all), so expose a way to disable it.
The decoder sometimes returned a non-zero value on error, but not on
every path.
Since we never use the value, always return 0 at the end (like in the
controller).
SDL_MouseWheelEvent does not provide the mouse location, so we used
SDL_GetMouseState() to retrieve it.
Unfortunately, SDL_GetMouseState() returns a position expressed in the
window coordinate system while the position filled in SDL events are
expressed in the renderer coordinate system. As a consequence, the
scroll was not applied at the right position on the device.
Therefore, convert the coordinate system.
See <https://stackoverflow.com/questions/49111054/how-to-get-mouse-position-on-mouse-wheel-event>.
Use high DPI if available.
Note that on Mac OS X, setting this flag is not sufficient:
> On Apple's OS X you must set the NSHighResolutionCapable Info.plist
> property to YES, otherwise you will not receive a High DPI OpenGL
> display.
<https://wiki.libsdl.org/SDL_CreateWindow#flags>
On user request to quit, two kinds of blocking calls must be interrupted
on the server:
1. the reads from and writes to the socket;
2. the call to MediaCodec.dequeueOutputBuffer().
The former case was handled by calling shutdown() on the socket from the
client, but the latter was not managed.
There is no easy way to wake this call properly, so just terminate the
process from the client (i.e. send SIGTERM on Linux) instead.
The server is copied to /data/local/tmp/scrcpy-server.jar and executed
on the device.
As soon as we are connected, we can unlink (rm) it from /data/local/tmp,
to keep the device clean.
The server is currently a JAR, but it may ba an APK or a DEX, so the
variable name should not contain the type.
Rename the environment variable, the Meson options and the C
definitions.
Expose net_recv_all() and net_send_all(), equivalent of net_recv() and
net_send(), but that waits/retries until the requested length has been
transferred.
Use these new functions where it was (wrongly) assumed that the
requested length had been transferred.
SDL_net is not very suitable for scrcpy.
For example, SDLNet_TCP_Accept() is non-blocking, so we have to wrap it
by calling many SDL_Net-specific functions to make it blocking.
But above all, SDLNet_TCP_Open() is a server socket only when no IP is
provided; otherwise, it's a client socket. Therefore, it is not possible
to create a server socket bound to localhost, so it accepts connections
from anywhere.
This is a problem for scrcpy, because on start, the application listens
for nearly 1 second until it accepts the first connection, supposedly
from the device. If someone on the local network manages to connect to
the server socket first, then they can stream arbitrary H.264 video.
This may be troublesome, for example during a public presentation ;-)
Provide our own simplified API (net.h) instead, implemented for the
different platforms.
The syntax was correct, but less readable, and it unnecessarily zeroed
the fields other than "type".
Create the event properly, from a separate method.
screen_render() should not be called on initialization:
1. it is useless, since the window is hidden until the first frame;
2. it writes an empty texture (probably green) to the renderer.
The "screen control" handled user input, which happened to be only
used to control the screen.
The controller and screen were passed to every function. Instead, group
them in a struct input_manager.
The purpose is to add a new shortcut to enable/disable FPS counter. This
feature is not related to "screen control", and will require access to
the "frames" instance.
The device serial was provided as a positional argument:
scrcpy 0123456789abcdef
Instead, expose it as an optional argument, -s or --serial:
scrcpy -s 0123456789abcdef
This avoids inconsistency between platforms when the positional
argument is passed before the options (which is undefined).
Since Meson 0.44, subproject_dir may not be '.' anymore. This implies we
must move app/ and server/ to a subprojects/ directory, which requires
to also change some gradle files.
Instead, just use subdir(), with options to disable building of the app
or the server.
The client was built with Meson, the server with Gradle, and were run by
a Makefile.
Add a Meson script for the server (which delegates to Gradle), and a
parent script to build and install both the client and the server to the
system, typically with:
meson --buildtype release build
cd build
ninja
sudo ninja install
In addition, use a separate Makefile to build a "portable" version of
the application (where the client expects the server to be in the
current directory). Typically:
make release-portable
cd dist/scrcpy
./scrcpy
This is especially useful for Windows builds, which are not "installed".
The current meson version is able to generate a config.h from a
configuration data object without any template.
However, older versions of meson require a template, so provide it for
compatibility.
The SDL clean up does not crash anymore on exit, probably since the
memory corruption caused by calling SDLNet_TCP_Close() too early has
been resolved.
On startup, the client has to:
1. listen on a port
2. push and start the server to the device
3. wait for the server to connect (accept)
4. read device name and size
5. initialize SDL
6. initialize the window and renderer
7. show the window
From the execution of the app_process command to start the server on the
device, to the execution of the java main method, it takes ~800ms. As a
consequence, step 3 also takes ~800ms on the client.
Once complete, the client initializes SDL, which takes ~500ms.
These two expensive actions are executed sequentially:
HOST DEVICE
listen on port | |
push/start the server |----------------->|| app_process loads the jar
accept the connection . ^ ||
. | ||
. | WASTE ||
. | OF ||
. | TIME ||
. | ||
. | ||
. v X execution of our java main
connection accepted |<-----------------| connect to the host
init SDL || |
|| ,----------------| send frames
|| |,---------------|
|| ||,--------------|
|| |||,-------------|
|| ||||,------------|
init window/renderer | |||||,-----------|
display frames |<++++++-----------|
(many frames skipped)
The rationale for step 3 occuring before step 5 is that initializing
SDL replaces the SIGTERM handler to receive the event in the event loop,
so pressing Ctrl+C during step 5 would not work (since it blocks the
event loop).
But this is not so important; let's parallelize the SDL initialization
with the app_process execution (we'll just add a timeout to the
connection):
HOST DEVICE
listen on port | |
push/start the server |----------------->||app_process loads the jar
init SDL || ||
|| ||
|| ||
|| ||
|| ||
|| ||
accept the connection . ||
. X execution of our java main
connection accepted |<-----------------| connect to the host
init window/renderer | |
display frames |<-----------------| send frames
|<-----------------|
In addition, show the window only once the first frame is available to
avoid flickering (opening a black window for 100~200ms).
Note: the window and renderer are initialized after the connection is
accepted because they use the device information received from the
device.
SDLNet_TCP_Close() not only closes, but also release the resources.
Therefore, we must not close the socket if another thread attempts to
read it.
For that purpose, move socket closing from server_stop() to
server_destroy().
Replace screen_update() by a higher-level screen_update_frame() handling
the whole frame updating, so that scrcpy.c just call it without managing
implementation details.
Do not wait 100ms anymore to let the server print any exception: we
justly want to ignore them.
Moreover, there is no nanosleep() on Windows, so this solve another
problem.
When the video stream socket is closed and read_packey() returns -1,
av_read_frame() still returns 0.
To detect EOF, check the flag eof_reached in the AVIOContext.
This avoids garbage errors on closing.
Expose frames_offer_decoded_frame() and frames_consume_rendered_frame()
so that callers are not exposed to frame swapping (between the decoding
and rendering frames) details.
The skip_frames flag was a non-configurable runtime flag. Since it is
not exposed to the user, there is no need for a (possible) runtime cost.
For testing purpose, we still want it to be configurable, so make it a
compilation flag.
We encounter some problems with SDL2_image on MSYS2 (Windows), so
implement our own XPM parsing which does not depend on SDL_image.
The input XPM is considered safe (it's in our source repo), so do not
check XPM format errors. This implies that read_xpm() is not safe to
call on any unsafe input.
Although less straightforward, use SDL_CreateRGBSurfaceFrom() instead of
SDL_CreateRGBSurfaceWithFormatFrom() because it is available with SDL
versions older than 2.0.5.
The purpose of handle_shortcut() was to group all actions together,
whether they are initiated from a text input event or a keycode event.
However, it did not handle the case where it was initiated from a mouse
event (a right-click must turn the screen on), since the action was
identified by the shortcut char.
Instead, expose one function per action, to be called directly from
where the event is handled.
The right-click is almost useless on Android, so use it to turn the
screen on.
Add a new control event type (command) to request the server to turn the
screen on.
The server is built as an APK to simplify the build, but in fact this is
a simple jar (it is not even signed).
In order to avoid confusion, rename it to .jar, so that users do not try
to "adb install" it.
Also rename it from "scrcpy" to "scrcpy-server" to distinguish from the
client-side.
The server path may be customized using SCRCPY_APK. If its basename is
different from "scrcpy.apk", it will be pushed with a different name,
so the execution would fail.
Therefore, force the push target filename.
On rotation, scrcpy resize the window to match the new rotation.
However, in fullscreen mode, setting the window size does not change the
windowed size on X11, so the behavior is incorrect.
To avoid the problem, apply the resize only after fullscreen is
disabled.
For readability, indent "case" in switch blocks.
Replace:
switch (x) {
case 1:
// ...
case 2:
// ...
case 3: { // a local scope block
int i = 42;
// ...
}
}
By:
switch (x) {
case 1:
// ...
case 2:
// ...
case 3: { // a local scope block
int i = 42;
// ...
}
}
