GitLab

Make sure that the password length is under the maximum lenght. If not
report it as an authentication failure with an adapted message.

Disable audio if no backend is available and configure is used with
the option '--with-audio=auto'

Commit 33d40162 uses
g_output_stream_write_all_async/finish which is on gio 2.44 version.
This breaks the build on older systems even if --disable-webdav is used.

Using #USE_PHODAV to wrap those functions that will not be called
without --enable-webdav.

Updating the display causes weird behaviors in virt-viewer, like
re-resizing to a previous monitor config when, after changing the
display resolution (through System > PReferences > Displays). the agent
reconnects.
It is not perfect yet, because when the agent reconnects the guest
resizes to the previous config and switch to the proper one, but this
seems more like an agent bug than a spice-gtk one.

Since spice-gtk requires glib 2.28, we can now fix a small FIXME.

Since G_TYPE_CLASS_GET_PRIVATE is a bit expensive, it's still worth to
cache it in klass->priv. However, there is no good place I can think of
to put this. (channel_class_init() is called only once, and not per
each subclass)

For historical reasons, the code was placed under gtk/ subdirectory.
If it was always bugging you, bug no more!

Remove outdated items:
- ChannelBaseAudio?,
- GDI backend wouldn't work well with scaling (moved to fdo bug)

Fix fdo link

The struct is g_new0() already.

Fail early during initialization if the usbclerk service can't be
reached.

This changes the current behaviour that would retry to connect to
usbclerk every time to perform an operation. Now, it will try each time
the UsbDeviceManager is initialized.

Use regular g_return precondition checks.

Throw an error when initialization failed.

To be consistent with GLib idioms, the code should use a common prefix for errors.

The current code unref() the device too early, it must be unref
after it is no longer needed.

We want the exported functions to follow the gio async idom. Currently,
both install() and uninstall() are paired with install_finish(). In the
following patch we introduce uninstall_async() uninstall_finish() using
that common internal op_finish() function. Then we correctly pair
associated async/finish() calls.

Using g_pollable_input_stream_create_source to generage several dummy
GSources in order to check if giopipe sets all of them to be dispatched.

This test check for zombie GSources during a write_all/read_chunk
operation.

Concurrent write is not supported and should fail. The GIO error is
G_IO_ERROR_PENDING

Both functions are basic the same so let's keep the generic one.

Client's webdav can request less data (8192) then the amount sent by
guest's webdav. Using g_output_stream_write_all_async in order to avoid
losing data.

PipeInputStream and PipeOutputStream should not fail when creating
GPollableStream source as this currently does not work with default
write_all and read_all functions;

In order to avoid creating zombie GSource in create_source of both
PipeInputStream and PipeOutputStream, we track all created GSources and
set them to be dispatched when data is available to read/write. It is
worth to mention that concurrent write/read is not possible with current
giopipe and only the last created GSource will read the data as it is
dispatched first.

Disabling the proxy avoids usage of GProxyResolver to determine
the necessary proxy protocol and to do the proxy negotiation.

GSocketClient uses the system proxy by default, and it may end up using
the system HTTP proxy with bad results if CONNECT is not
supported. spice-gtk uses it's own SPICE_PROXY instead, and doesn't rely
on GSocketClient default proxy but GProxyAddress instead. Disabling the
default proxy solve a wrong proxy from being used.

It may be worth to revisit this change if GSocketClient can be told to
ignore proxies that are not eligible for Spice connections. (HTTP could
be though, in which case it would be a user configuration issue)

Fixes:
https://bugzilla.redhat.com/show_bug.cgi?id=1040679

g_simple_io_stream_new() do not steal references from us, unref the
streams to avoid the leaks.

Leak found by Victor Toso.

==30787== 2,808 (648 direct, 2,160 indirect) bytes in 9 blocks are definitely lost in loss record 479 of 482
==30787==    at 0x93AEBFD: g_type_create_instance (gtype.c:1849)
==30787==    by 0x939128A: g_object_new_internal (gobject.c:1774)
==30787==    by 0x9392AA0: g_object_newv (gobject.c:1921)
==30787==    by 0x939338B: g_object_new (gobject.c:1614)
==30787==    by 0x403B30: make_gio_pipe (giopipe.c:441)
==30787==    by 0x403C8E: spice_make_pipe (giopipe.c:465)
==30787==    by 0x4029B5: fixture_set_up (pipe.c:42)
==30787==    by 0x9844EC8: test_case_run (gtestutils.c:2123)
==30787==    by 0x9844EC8: g_test_run_suite_internal (gtestutils.c:2185)
==30787==    by 0x984509A: g_test_run_suite_internal (gtestutils.c:2196)
==30787==    by 0x98453FA: g_test_run_suite (gtestutils.c:2249)
==30787==    by 0x9845430: g_test_run (gtestutils.c:1553)
==30787==    by 0x402012: main (pipe.c:312)

There is no strong reason to have signal handlers in spicy (a test
tool). Clean quit can be achieved through the File/Quit menu.

Fixes:
https://bugzilla.redhat.com/show_bug.cgi?id=1197175

Commit No. 63bf0027 changed number of
arguments of function playback_stop(), and didn't change call of this
function on line 628.

This would break building when using --with-audio=gstreamer

Check g_output_stream_write_all() return value to silence a coverity
CHECKED_RETURN warning

Client usually connect to channel-new to connect their
handlers, such as open-fd.

The usbmanager channel-new handler will call channel_connect() on usbredir
channels, which may call open-fd.

However, open-fd can be emitted before the client had a chance to
connect their handlers (from the channel-new callback).

Connecting after the default handler solves this case.

Fixes:
https://bugzilla.gnome.org/show_bug.cgi?id=748665

Since channel can become session-less, let's check they still have the
expected session when running the open-host idle.

Ensure playback is stopped when the channel is destroyed.

We are going to reuse playback_stop() in following commit.