This adds a explicit method for cleaning up the animation. Previously,
animations expected the next_frame to be called where frame=end.
That guarantee was never provided by the api. This resulted in
animations sometimes being in odd states. For example, Browse activity
had an animation that overrode the Gtk "draw" signal. If the signal
was not unbound, the animated object was mistakenly rendered after
the completion of the animation, obstructing the user's view.
This api is explicit and allows cleaner code in the animation
implementation. The clean up code goes in do_stop rather than in a
conditional of next_frame.
Many activities already have the repository url pointing to their GitHub
page - which also serves as a bug tracker for many of the projects. The
repository url is not part of the spec, but the bug tracker url is.
This commit adds the 1st pass generator for these files. It also adds
documentation about the required fields in the "activity.info"
file, as AppStream requires more metadata than most activities currently
include.
When created via "git submodules add", a submodules may have the
"/.git" path be a file rather than a directory. The bundlebuilder
previously thought that all submodules had "/.git" as a directory.
The unversioned import results in the following log noise:
/usr/lib/python2.7/site-packages/sugar3/graphics/window.py:26: PyGIWarning: GdkX11 was imported without specifying a version first. Use gi.require_version('GdkX11', '3.0') before import to ensure that the right version gets loaded.
from gi.repository import GdkX11
Problem #1, buddy menu appears on Sugar startup on physical hardware, or
when mouse is centre of display.
Problem #2, Terminal toolbar steals keyboard focus, when mouse is left
in home view search box at activity start time.
Regressions were introduced in the port from Gtk2 to Gtk3.
Cause is enter-notify-events that arrive because cursor is already
present in widgets as they are realised and mapped.
Fix is to capture cursor position when the widget is being created, and
ignore enter-notify-event if the same position is seen.
The log message "PaletteWindowWidget popup get_visible True" is not
useful, and is removed. It can be made to occur in Terminal by repeated
right-click on any of the first three toolbar button.
Previously there was the concept of having 2 popdowns for a palette.
The first popdown showed the of tooltip the palette, and the 2nd
showed the content of the palette. However, this was confusing
to new users, and made users wait for little reason.
This commit removes the 2nd popdown. An example of this is when you
mouse over an icon to invoke the palette. Previously, it showed
the tooltip, then made you wait, then showed the content of the
palette. With this patch, the whole palette is shown at the
first popdown, making a snappy UX.
Steps to reproduce:
1. Long press on home view favorite icon (an event icon)
Expected: Palette invoker where finger is
Actuall: Palette invokes in top left corner
This is because the long press controller emits coords relative
to the widget's window, and EventIcons sometimes have their own
windows. We can't change the long press controller, as it is a
public api. But we fix this by translating the coords to the
root window coords in the CursorInvoker. Using root coords is
consistent with the right click handleing in the cursor invoker.
Steps to reproduce:
1. Long press activity icon on homeview
Expcected: Palette invokes, icon not activated
Actual: Palette invokes, icon also activated, activity launched
This was because the "activate" signal was emitted before the
palette invoker got it's signal, meaning the palette invoker couldn't
scilence the event.
Steps to reproduce:
1. Long press activity icon on homeview
Expcected: Palette invokes, icon not activated
Actual: Palette invokes, icon also activated, activity launched
This was because the "activate" signal was emitted before the
palette invoker got it's signal, meaning the palette invoker couldn't
scilence the event.
This will allow .desktops for all of the activities to be
packaged. In the future, we could ship a special launcher script
so that the user can choose the jobject to run, but for now
we just open the activity using the command specified in the
activity.info file.
If the user has a "random" telepathy connection manager
installed and running on the system (eg. telepathy idle being
active if they are chatting on irc with GNOME's Polari), activities
will crash on startup.
Steps to reproduce the crash:
1. Run Polari (eg. via the Terminal activity)
2. Join an irc channel in Polari (eg. #sugar on freenode)
3. Try to open an activity from the home view
- notice that opening any sugar3 activity will crash
Currently, the ToolInvoker code passes the button child in such a
way that the WidgetInvoker queues only the child's draw. In previous
Gtk+ versions, this magically resulted in redrawing the parent as well.
However, Gtk+ 3.20 is probably trying to save power and not randomly
redraw widgets. To fix this, we must queue the redraw of the widget
that we want to redraw explicitly.
- Handle lack of GSettings gracefully.
- Still requires sugar-datastore.
- Supports avoiding X11 docks/panels.
- Provides icons for Activity windows.
Try it outside Sugar. Go to an Activity directory and run 'sugar-activity'.
Tested it with Terminal, Finance, Write, Browse, Memorize under XFCE4.
Tested in Gnome under OLPC-OS.
Also works from Sugar Terminal Activity.
Does not affect regular Sugar operation.
This is patch v.2 -
Addresses most concerns:
- Removed commented code, sorry.
- Changed code to use profile.get_nickname and get_color where possible.
Couldn't the launcher just pass this info?
Maybe the launcher could set the activity root as well?
- It is intended to be usable from the command line also.
Should put sugar version in the environment
- It is intended to work even without Sugar Shell installed.
Why don't we always set the icon?
- On XO it might use some memory. I was concerned to degrade
performance.
Also, imports should be at the top of the file?
- Also a concern about performance on XO.
This way it is only loaded in this use case.
Maybe it is insignificant -moved as requested.
It would be nice if the changes to the POT for sugar-toolkit-gtk3
could be incorporated in this pull request, please.
- There were no changes to POT files as part of this patch. Maybe
it is worth translating low level command line tools, not sure.
Suggest packaged activities might also provide .desktop files.
- Intriguing but not sure within scope of this patch. You mean generate
a .desktop file automatically as an option? Sounds nice!
Suggest sugar-activity might also accept path to unpacked bundle.
- Implemented!
Sugar depended on Rainbow for clearing the activity instance/ and tmp/
directories. But Rainbow is no longer used downstream.
- remove support for Rainbow,
- avoid race when creating directories; don't check they exist before we
create.
Git submodules are only listed by their root directory in the
output of "git ls-files". Therefore, we must handle this case
so that the files from the module are included in the install
and dist_xo commands.
To reproduce the issue, you could do something like the following:
cd pippy
git clone https://github.com/samdroid-apps/collabwrapper
git add collabwrapper
python setup.py install # or just osbuild run
CHANNEL_TYPE_TUBES was deprecated since telepathy-gabble 0.17.25 and
is no longer present on modern distributions. This breaks collaboration
for _all_ activities, regardless if they use tubes or not.
Therefore, check if the channel creation failed due to support and allow
channel creation process to continue.
This patch fixes collaboration for the activities that do not use tubes
channel in modern distributions, but also allow to use tubes when running
on older distributions.
Signed-off-by: Martin Abente Lahaye <tch@sugarlabs.org>
In Gtk+ 3.20, you need to use the css name to select elements,
rather than the gtype name. Therefore, these must be added.
The css name must be set before the class instances are created, as
it effects the class rather than the instance. This is why it must
be places after the class definition.
A rework of the TreeViewInvoker used by the journal and activity list.
Fixes a persistent journal entry palette that reappears in journal or
home view. https://bugs.sugarlabs.org/ticket/4918
Also fixes an over-sensitive journal icon; with the journal mostly
empty, rapidly move the mouse from the bottom of the screen into the
lowest icon and then out again. Before this patch, the palette was
shown. After this patch, the mouse must rest in the icon.
Changes made:
- override _ensure_palette_exists in parent class Invoker,
- remove reference to MouseSpeedDetector, it is no longer required,
- remove enter and leave events; they are for the TreeView as a whole,
and are not required,
- properly detect that None is returned by get_path_at_pos, when the
coordinates are outside the TreeView cells.
Not fixed:
- moving the mouse out of the journal icon invoker does not popdown the
palette, but it does when moving the mouse out of the activity icon in
the activity list; so this is a problem beyond the invoker,
Tested on Fedora 18 and Ubuntu 15.10.
Test cases:
- check a click will activate item,
- check a right-click will show palette,
- check a mouse hover will show palette,
- check a brief mouse passing will not show palette,
- check if mouse leaving palette will popdown.
Instead of getting the position of the mouse later in the code,
this commit uses the coords given to sugar by Gtk that say where
the event happened. This means that if the system is busy, or
the user is fast with the mouse, the palette will pop up where
they originally clicked.
When disk is full, Sugar will not start, because the backup log
directory cannot be created.
X session logs in tmpfs show this exception:
Traceback (most recent call last):
File "/usr/lib/python2.7/runpy.py", line 162, in _run_module_as_main
"__main__", fname, loader, pkg_name)
File "/usr/lib/python2.7/runpy.py", line 72, in _run_code
exec code in run_globals
File "/usr/lib/python2.7/site-packages/jarabe/main.py", line 20, in <module>
logger.cleanup()
File "/usr/lib/python2.7/site-packages/sugar3/logger.py", line 113, in cleanup
os.mkdir(backup_dir)
OSError: [Errno 28] No space left on device: '/home/user/.sugar/default/logs/1448258018'
Fix is to ignore the exception.
We've been here before! Was previously fixed in 2010:
https://github.com/sugarlabs/sugar/commit/4cde481https://bugs.sugarlabs.org/ticket/1720http://dev.laptop.org/ticket/9623
But new regression introduced in 2012 during reorganisation of
code by Daniel Narvaez; the exception was no longer handled:
https://github.com/sugarlabs/sugar/commit/0e45f9dhttps://github.com/sugarlabs/sugar/commit/19db9c5
Reported-by: Nathan Riddle <nathanr333@charter.net>
Tested-by: James Cameron <quozl@laptop.org>
Create a private class for the button, timeout and response. Use the
private class as parent for both NotifyAlert and TimeoutAlert.
Extends the spurious response fix (cfeabdc) to NotifyAlert.
Also use g_timeout_add in preference to g_timeout_add_seconds, because
the latter is less precise in implementation.
Tested by changing the Register alert temporarily.
For an object of class TimeoutAlert, an OK response occurs on timeout,
even if user has clicked a button.
Remove the timeout when the user clicks a button.
A new method _response is added which overrides the parent method.
Gtk 3.8 introduces the tick callback [1]. This allows widgets to be
signaled before each frame, simmilar to requestAnimationFrame in the
browser.
This patch adds an optional widget argument to the Animator class so
tick based animation can be used. This is much more efficent than
using timeouts, as we get a more appropriate frame rate for the user.
[1] https://developer.gnome.org/gtk3/stable/GtkWidget.html#gtk-widget-add-tick-callback
As the ticket describe, create a object in the Journal
is no longer optional. The change do not break activities,
because the optional parameter is not removed, just ignored.
Right now is not possible do "from sugar3.datastore import datastore"
outside of GLib loop, because the import try to connect to a DBus signal,
and throw a exception.
That was not a problem in Sugar, because we always have a Glib loop
on all our activities, but is a problem when we want import the classes
to generate docs with pydoc. Many other classes import datastore,
then this is a blocker.
This patch removes the initialization when imported,
will be done when needed.
To solve ticket #4673, we calculate the palette size using the size
of all the widgets [2]. That solves the problem on Palettes based on
_PaletteMenuWidget but on _PaletteWindowWidget produce a wrong value
when the secondary_label is not null, and as a consequence
the palette position is wrong.
The solution is do the calculation only for _PaletteMenuWidget based
palettes.
[1] https://bugs.sugarlabs.org/ticket/4673
[2] b9d6b628a9
With the introduction of 'activate' signal on [1], and the use of that
signal in Sugar Home, we introduced a bug, now, pressing the secondary button
starts the activity instead of open the palette.
This patch solves the issue checking the putton pressed before send the signal.
[1] a19cf9ed27
When is used to clone a activity from sugar,
the options parameter is set to None.
The other uses of no_fail have False as default value
but this case was missing.
The effect of this error is that a .xo is not created in the Journal,
when the user select 'Duplicate' in the View Source window.
Commit a19cf9e changed the logic for handling CanvasIcon clicks. Therefore,
the visual state indicators must be updated to fit the new logic.
Test Case: Hold the mouse down on a home view icon. Move the mouse
in and out of the icon. See that the states correspond with what
happens if you release the mouse.
The mouse detector have a variable to store the id for the timeout.
The stop() method check if the id is null but do not set the id to null
after remove the GObject timeout, then try to remove the timeout multiple times.
The efect can be seen if a user move the mouse slowly over the icons
in the Sugar home, in shell.log we see many lines like:
sugar3/graphics/palettewindow.py:443: Warning: Source ID 2464 was not found
when attempting to remove it
The industry convention for mouse driven menu options is for them
to be activated when two conditions are met:
* a button down event occurs with the pointer inside the option, and;
* a button up event occurs with the pointer inside the option.
This issue was already solved on the PaletteMenuItem,
but the EventIcon have the same problem. This change add a 'activate' event,
that control that the two conditions are meet.
The code in Sugar need use this event instead of button-release-event.
When we use set_icon_widget(), GtkToolButton set a private property
contents_invalid = TRUE [1], and gtk_tool_button_construct_content [2]
is called. Then if the label widget not exist, a GtkLabel is added.
This is a problem for us, by example in the Clipboard buttons in the frame.
By adding a empty Gtk.Box instead of a label, we avoid the label creation.
But as this is a internal implementation of Gtk, can change in the future,
in fact the issue is not visible with Gtk < 3.16
This patch is based in a pr sent by Sam Parkinson
[1] https://github.com/GNOME/gtk/blob/master/gtk/gtktoolbutton.c#L1415
[2] https://github.com/GNOME/gtk/blob/master/gtk/gtktoolbutton.c#L357
When a programmer need open one activity from other [1]
usually need display in the user interface information about the activity
to open, like the name or the icon.
This method allow get this information, previously only available on jarabe.
[1] http://wiki.sugarlabs.org/go/Features/Start_activity_from_another_activity
The palette using a Gtk.Menu, had a separator displayed using a custom widget,
but that was not well displayed (the line didn't had the palette width
due to margin on the palette) and was broken on Gtk >= 3.16
Instead of use that widget to draw the separator line between at the bottom
of _HeaderItem, as we do with the toolbar buttons when the palette is displayed.
Historically, distro packagers have been using bunblebuilder, via
"setup.py install", to populate packages directories and files.
The install procedure uses ActivityBundle.install_mime_type to install
custom mime types, by creating symlinks and updating the system mime
types database via the update-mime-database tool.
When ActivityBundle.install_mime_type is executed during packages
creation, the symlinks end-up broken in the final package and the use
of update-mime-database populates the final packages with a new copy
of the database files which could clash with the system copy.
This patch adds a new option, called --skip-install-mime to "setup.py
install" to skip the execution of ActivityBundle.install_mime_type and
avoid the issues mentioned above.
Activity packagers should use this new option. The creation of symlinks
and the execution of update-mime-database should be done in post install
packaging steps.
Signed-off-by: Martin Abente Lahaye <tch@sugarlabs.org>
If is not initialized, the check for the espeak plugin will fail.
This patch also check for the espeak plugin at the beginnig
avoiding check every time the function enabled() is called.
There are two different palette widgets in Sugar.
_PaletteMenuWidget is a Gtk.Menu and _PaletteWindowWidget is a Gtk.Window.
Only the palettes where the widget is a Gtk.Window can do present()
After add the directory to the IGNORE_DIRS array,
we need make the standard procedure to get the list off files
for the bundle, get_files_in_git honor the configured ignored
files and dirs. That was not a problem until now,
because that files was already not added to git,
but in the case of the screenshots, are included in git,
but we don't want include them in our bundles.
This new invoker is able to handle all the palettes on a treeview,
removing the need of one invoker per cell renderer (CellRendererInvoker).
This simplifies the code and makes it more efficient.
Also removes the logic from the CellRendererIcon, which
should only care about drawing itself. [1]
Is important to note than in Gtk3 a CellRenderer is not a GtkWidget
then can't know when the mouse is over it or have the usual events
used by other invokers, making the implementation CellRendererInvoker
very complicate.
This commit also removes the invoker of CellRendererIcon.
The ScrollingDetector logic is simplified too,
because now there are only one invoker instead of one by renderer
and the example code updated.
[1] https://developer.gnome.org/gtk3/stable/GtkCellRenderer.html
Detecting if the treeview is scrolling we can improve the performance of
the CellRendererIcon by avoiding unneeded calculations.
The example scrollingdetector.py shows how this can be used. The
changes are backward compatible.
Signed-off-by: Manuel Quiñones <manuel.por.aca@gmail.com>
Signed-off-by: Gonzalo Odiard <godiard@sugarlabs.org>
On commodity hardware without olpc-powerd, there is unnecessary
filesystem access.
On XO laptop hardware there are unnecessary errors in log for every
object delete:
ERROR root: Inhibit Suspend: Could not delete file
/var/run/powerd-inhibit-suspend/1773
Exception AttributeError: "'NoneType' object has no attribute
'endswith'" in <bound method PowerManager.__del__ of
<sugar3.power.PowerManager instance at 0xa15962c>> ignored
The Clock activity in speaking mode is a good reproducer.
Following changes are made:
- move the directory check to __init__, and set self._path to None if
olpc-powerd is not present,
- on inhibit_suspend, use self._path, which avoids a check of the
directory,
- on restore_suspend or __del__, avoid a call to os.unlink if
olpc-powerd is not present.
Use a more efficient method to check if the pointer is inside the cell,
because we have the cell_area in the do_render method.
Signed-off-by: Manuel Quiñones <manuel.por.aca@gmail.com>
Signed-off-by: Gonzalo Odiard <godiard@sugarlabs.org>
This commit adds the `launch_bundle` function in `sugar3.activity.
activity` which accesses the same function in the shell is over dbus.
This is means the shell process launches the bundle. Activities
should not launch child processes as per the Rainbow security model
[1].
[1] http://wiki.laptop.org/go/Rainbow
If a translator add a '\n' char in the summary translation
ConfigParser can't parse the linfo file.
Remove them when the file is created and show a message to warning
to the developer.
When we ported sugar to Gtk3 we removed the support of the fields
exec and service_name due they were deprecated for a long time [1].
But this is still a problem in deployments, and we already added
compatibility for the service_name field.
Here [3] there is a long thred were we discussed this issue.
[1] ae85dd62d4
[2] 32f97b48c3
[3] http://lists.sugarlabs.org/archive/sugar-devel/2013-October/045404.html
As described in SL#4307 Gtk.Clipboard.set_with_data is not
introspectable, therefore it can't be accessed though python
Gtk bindings.
Therefore, as a workaround for this issue, we provide acesss
to this method though a custom SugarExt binding.
This solution uses Carlos Garnacho annotations as described in
https://bugzilla.gnome.org/show_bug.cgi?id=656312#c18
Fixes SL#4307.
Signed-off-by: Martin Abente Lahaye <tch@sugarlabs.org>
Now when the espeak plugin is not installed there will be an
error (logging.error) in the log. Instead of the ugly message
of Gst
Also, if Gst isnt installed it debug a error too.
Fixes#3345
The BundleBuilder currently crashes on a malformed translation file. This is
good for humans (makes them fix their issues) but crashes bots who need it
compiled and don't care about that single translation file.
This patch adds a --no-fail option which stops the builder from crashing
To display a list of translated language names in the UI,
we add a list of the available languages, because the espeak
language names are not translated.
Sam Parkinson