GitLab

Propagating Python exceptions from callbacks through the C context back to the
original caller does not currently happen, is nontrivial/unsafe to implement,
and not desirable at this point any more as by now we have established the
current behaviour. So remove the catching of ZeroDivisionError in the tests.

https://bugzilla.gnome.org/show_bug.cgi?id=616279

These methods now return a context manager object. Within the __exit__ method
thaw_notify() and handler_unblock() are called respectively. This allows
statements like the following:

with obj.freeze_notify():
    obj.props.width = 100
    obj.props.height = 100
    obj.props.opacity = 0.5

This does not affect standard usage of these methods.

https://bugzilla.gnome.org/show_bug.cgi?id=672324

Signed-off-by: Martin Pitt <martinpitt@gnome.org>

https://bugzilla.gnome.org/show_bug.cgi?id=620943

Based on original patch from Bryan Silverthorn.

https://bugzilla.gnome.org/show_bug.cgi?id=524719

Based on original patch by Juanje Ojeda <jojeda@emergya.es>.

https://bugzilla.gnome.org/show_bug.cgi?id=646667

When creating a new instance using Type() and trying to access
__grefcount__ before calling the subclass's __init__ function, there
used to be a segmentation fault because we were trying to access the
not yet created object. Now raise a proper exception instead.

https://bugzilla.gnome.org/show_bug.cgi?id=640434

Co-authored-by: Martin Pitt <martinpitt@gnome.org>

They will basically cause a crash if misused, and you can always use a
python member attribute instead.

https://bugzilla.gnome.org/show_bug.cgi?id=641944

Signed-off-by: Martin Pitt <martinpitt@gnome.org>

Commit 832f16f9 fixed a lockup with multiple GLib.MainLoops. Add corresponding
test case.

https://bugzilla.gnome.org/show_bug.cgi?id=663068

Used when overriding methods like gtk_container_forall wich pass in a
callback that needs to be executed on internal children:
    def do_forall(self, callback, userdata):
        callback(self.custom_child, userdata)

https://bugzilla.gnome.org/show_bug.cgi?id=644926

Co-authored-by: Tomeu Vizoso <tomeu.vizoso@collabora.co.uk>
Co-authored-by: Simon Schampijer <simon@laptop.org>
Signed-off-by: Martin Pitt <martinpitt@gnome.org>

https://bugzilla.gnome.org/show_bug.cgi?id=668903

Signed-off-by: Martin Pitt <martinpitt@gnome.org>

3.2.x is built from the pygobject-3-2 branch now, and 3.2.0 is released. So
continue with 3.3.x on master.

- Ignore *.o, backup files, and generated Makefiles in all subdirectories
- Ignore *.pyc files.
- Do not ignore .gitignore, we actually want to track this.

- Symlink *.py files from srcdir into builddir during build, as Python does not
  accept the extensions and modules in different paths.
- "make clean" should remove *.pyc files
- tests/runtests.py: Look for tests in srcdir, not in builddir

Also reorganize the pyflakes check, since target dependencies do not
take the exit status of the shell command into account.

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

https://bugzilla.gnome.org/show_bug.cgi?id=672627

By running the whole source tree via the indent.py script found
in the Python distribution.

failIf → assertFalse, failUnless → assertTrue

Caught by the previous commit of making deprecations fatal.

As we dropped the static bindings a while ago, there is no need any more to run
the tests in two phases (static/GI). Now just run them all in one go,
simplifying tests/Makefile.am.

As this changes the order of the tests, defining $GSETTINGS_SCHEMA_DIR now
needs to happen even further, so move it from tests/test_overrides.py to
tests/runtests.py.

With current glib, gsettings now fails to find the gschemas.compiled during the
tests. Move the setting of $GSETTINGS_SCHEMA_DIR before the module import,
which makes this work again.

This widget has been removed in Gtk+ 3, add a small wrapper to the
compat module to make at least basic pygtk programs that use it work.

https://bugzilla.gnome.org/show_bug.cgi?id=672589

https://bugzilla.gnome.org/show_bug.cgi?id=672578

And add a target to make check that runs pyflakes.

https://bugzilla.gnome.org/show_bug.cgi?id=672578

Instead of setting warnings/criticals to fatal in individual test modules, do
it in runtests.py, so that it applies to all tests.

We currently have some tests which are known to generate CRITICALs (now marked
with FIXME), and some WARNINGs (as they test behaviour with known-bad values).
For these, warnings/criticals are now explicitly permitted.

Alberto Mardegan authored Mar 20, 2012 and Martin Pitt committed Mar 21, 2012

Marshalling of interfaces got broken with commit
7746d218.

Also, do not abort on unsupported types, but log a critical failure and
continue.

https://bugzilla.gnome.org/show_bug.cgi?id=668903

Commit bf8c9583 changed the List/TreeStore overrides to use
insert_with_valuesv(), but supplied all columns instead of just those which are
not None. With this, None values cause warnings like

(runtests.py:12375): Gtk-WARNING **: /build/buildd/gtk+3.0-3.3.20/./gtk/gtkliststore.c:851: Unable to convert from (null) to gboolean

Update the tests to make warnings fatal, to catch this better.

Change _convert_row() to skip the None entries and return the list of not-None
columns, and use the latter instead of a simple range(n_columns). This matches
the behaviour before bf8c9583, where columns with None values were skipped
as well.

https://bugzilla.gnome.org/show_bug.cgi?id=672463

Tests currently give

(runtests.py:15072): GdkPixbuf-WARNING **: GdkPixbufLoader finalized without calling gdk_pixbuf_loader_close() - this is not allowed. You must explicitly end the data stream to the loader before dropping the last reference.

Fix this by calling close().

Use 3.1.93 for now, this will most likely become 3.2.0 as it is.