Upon import of uuid, that module uses ctypes.util to find libuuid/libc,
which involves forking a shell process which then forks and runs ldconfig.
ldconfig performs a whole bunch of disk accesses. This is a slowdown.
uuid.getnode() is also documented to be slow. Lets avoid using it in
the activity launch path; we don't need such a high degree of
randomness.
This makes activities launch 1-2 seconds faster on XO-1.
The new bundle helpers were being auto-loaded with sugar3.bundle due
to their placement in __init__.py. This was causing Gi to be imported,
breaking GTK2 activity launches.
Move the helpers to a dedicated module.
At the moment our tests requires to be run inside a clean
dbus/xvfb session, which is not always available where the
check target is used (for example in distcheck). Also when
running distcheck in buildbot we don't really want to run
tests again, since we just run them in the check step.
We might revisit this in the future but using a non standard
target is the easiest solution for now.
The bundle API is a bit confusing and inconsistent. Fix up the
ActivityBundle and ContentBundle classes to be more consistent and to
work with the ongoing improvements being made in bundleregistry.
The last change brak compatibility with the use in activities
without set the filter_type parameter, because None is not
a allowed value in the dbus call.
Signed-off-by: Gonzalo Odiard <gonzalo@laptop.org>
It needs to be called as early as possible before unit tests
has a chance to import gtk.
Hopefully will fix random test failures we are seeing in buildbot.
Constants are defined to select the different filter_type
values. A comment was added to document the use.
Signed-off-by: Gonzalo Odiard <gonzalo@laptop.org>
This ensures the test itself is not recognized as a client
by the registry. I'm not sure how that happens since we are
not loading gtk, but it does. Not being a real gtk client with
a mainloop we wasn't answering dbus calls and causing the registry
to hang and fail.
Development continued there for a while. Now I will remove it
and used the toolkit implementation instead.
Changes:
* Retry faster
* Handle GLib errors
* Don't always dump the tree
* Support master atspi api
