When connecting to a remote libvirt instance, a VM may only be listening
on localhost for SPICE/VNC connections. In such a situation, virt-viewer
then tries to connect to localhost, which is not correct as this
'localhost' referred to the remote libvirt host it connected to.
This commit adds a couple of tests on the libvirt URI used and the
<graphics> listen address to error out in this situation.
Resolves: rhbz#1108523
In order to solve several problems with sizing and resizing displays, a
'map' handler was added to VirtViewerDisplay. The first time the map
handler runs, its queues a resize to attempt to ensure that the window
gets created at its desired size. Subsequent map events generate a call
to _make_resizable(), which was an attempt to ensure that the window was
always 'shrinkable' on the Microsoft Windows platform. Recent testing
suggests that this _make_resizable() is not actually necessary on
Windows anymore, since it is possible to shrink the display even when
this call is removed.
In addition, the call to _queue_resize() is a bit of an indirect
solution to the problem of ensuring the proper size at startup. What we
really want is to guarantee that the very first size request negotiation
returns the desired size rather than the minimum size. In order to do
this, we've added a flag to determine whether we've ever received a size
request, and if not, we return our desired size, even if 'dirty' is not
set.
Using %d as a format-specifier for intptr_t causes a warning with
mingw64:
virt-viewer-events.c: In function 'virt_viewer_events_add_handle':
virt-viewer-events.c:103:5: warning: format '%d' expects argument of
type 'int', but argument 5 has type 'intptr_t' [-Wformat=]
g_debug("Converted fd %d to handle %d", fd, _get_osfhandle(fd));
Add support to build the virt-viewer's msi using GTK3.
For the GTK3 build, in order to provide all used icons for Windows
systems we have to include manually all the icons we want to or add
adwaita-icon-theme as dependency. I've decided to go with the first
approach, what can be improved when we have "foreach" support in
msitools (https://bugzilla.gnome.org/show_bug.cgi?id=741296).
libgcc_s_sjlj-1 is needed by libgmp (on x86)
libgcc_s_seh-1 is needed by libgmp (on x86_64s)
libgmp-10.dll is needed by libnettle
libhogweed-2-5.dll is provided by libnettle
libnettle-4-7.dll is needed by gnutls
gnutls is needed by gvnc and libvirt
An interesting point here that worth to mention is the usage of /nonfatal
when including libgcc_s_{sjlj,seh}-1.dll. As we only have the _seh in x64
the build breaks trying to add "not found" files. A check for arch was
one option to solve the problem, but _sjlj may exist in x64 as well, when
using an old gcc. An explicit check if the file exists (in compile time)
was another idea, but for some reason the "-" part of the filename was
interpreted as a math operand, breaking the build.
With all that in mind, adding /nonfatal was the most convenient solution.
Setting the zoom-level using the command line option '--zoom' is not
working for vnc guests. This problem can be solved by emitting
the "display-desktop-resize" signal when vnc is initialized.
https://bugzilla.redhat.com/show_bug.cgi?id=1170071
SpiceSession in spice-gtk v0.27 will remove channels from session during
disconnect (and not when they are actually disposed). When no channels
are left, session-disconnected is emitted, and the VirtViewerSession
will be unref from the application. Use a weak reference to self to
avoid crashing after calling spice_session_disconnect()
As a workaround for existing clients, spice-gtk v0.27 will defer the
disconnection to idle time. But the fix still makes sense and would
prevent potentially future issues if spice-gtk changes back to sync
disconnection.
(the alternative of calling ref/unref would needlessly recreate a
SpiceSession with a call to create_spice_session(), which is something
we can avoid when leaving the application)
- do not overwrite err if ->initial_connect() sets it
- remove need for waitvm if the display server isn't yet started (note:
this function might be untested, I am not sure relying on libvirt events
is enough)
- remove need for waitvm if the display server isn't yet started (note:
this function might be untested, I am not sure relying on libvirt events
is enough)
Some refactoring to make the code easier to read, mostly code
movement/reindenting and introduction of a "wait" label which has the
same purpose as "done".
This also adds a "goto wait" within an if block, but this does not
change the initial code flow, just makes it more explicit.
