We achieve this with two changes:
- Move the Shell.get_thumbnail call in DocInfo from _init
to getIcon, so that it isn't executed until it's actually
needed.
(If caching the output of said call permanently is desired
we could still do it on the first getIcon invocation, but
I don't believe this is necessary given that looking up an
already generated icon is pretty fast and this also gives
us an updated icon in case the file changes.)
- More importantly, we ommit the get_thumbnail call in case
the URI doesn't start with file://. Looking up, for example,
an http:// URI is very slow, and doesn't give us an icon anyway.
http://bugzilla.gnome.org/show_bug.cgi?id=586539
The list of all applications was based on the items that show up in
the menus, and might not have contained all applications returned
by the AppMonitor for the most used applications request. One example of
such an application was Evince. This resulted in a crash when an
application we did not prepare the info for was returned as one of the
matches by the AppMonitor, so we now include all applications returned
by the AppMonitor in the list, in addition to the ones from the menus.
Also mark apps as stale when we catch a "changed" signal from the AppMonitor
to ensure that the cache is refreshed.
With commit a3d35af444 variable
itemInfo in js/ui/appDisplay.js was changed to a new object
(AppInfo from js/misc/appInfo.js) but some of the code in
js/ui/appDisplay.js wasn't updated accordingly. This commit
fixes that and makes the search box in the overlay usable
again.
This lets us share the recent-app-tracking, recent-file-tracking, and
icon-drawing code between the overlay and the sidebar, without the
sidebar having to poke into AppDisplayItem and DocDisplayItem's guts.
Currently function shell_get_thumbnail_for_recent_info located in
src/shell-global.c is used to get thumbnails for recently used files.
However, it only works if you have a GtkRecentInfo object for the file,
even though the thumbnail generation code doesn't depend on it. This commit
renames the function to shell_get_thumbnail and makes it generic so that it
just takes two strings: a filename and a mimetype.
When there is no root pixmap, the result will be a CoglMaterial
with a layer with no texture, and that causes a crash.
Work around this by supressing painting the root pixmap actor
when there is no root pixmap.
http://bugzilla.gnome.org/show_bug.cgi?id=585196
- clutter_actor_get_transformed_position()/size() return floats
- clutter_stage_get_actor_at_pos() takes a pick mode
- ClutterTimeline no longer has a concept of frames
- ClutterUnit is now replaced by float
- cogl_texture_new_from_data() signature changed
http://bugzilla.gnome.org/show_bug.cgi?id=585013
With clutter changes, we can now no longer clone an actor unless
it is part of a stage. So, we hide the root pixmap source, and
add it.
This means that the logic to free the source actor when the
last clone disappears no longer applies, since the stage will also
reference it; so we just leave the actor around permanently.
http://bugzilla.gnome.org/show_bug.cgi?id=585012
We need a foreach_with_internals() function that includes the
background_texture/background_rectangle actors, so that states
will properly be updated on map/unmap/etc.
http://bugzilla.gnome.org/show_bug.cgi?id=585007
- ClutterUnit is now replaced with float
- allocate() now takes flags rather than absolute_origin_changed boolean
- cogl_texture_new_from_data() signature changed
http://bugzilla.gnome.org/show_bug.cgi?id=585007
This removes the border and padding and aligns the box to the top. Perhaps we
could push the padding down into the active area of the icon since it does look
a bit nicer with it.
http://bugzilla.gnome.org/show_bug.cgi?id=583139
The animation time for the overlay is a bit long. We're using 0.5 second.
From my quick tests of "another system" it takes about half that time to
animate. The other system feels about right.
http://bugzilla.gnome.org/show_bug.cgi?id=583572
The code here is significantly cleaner if we use the data Metacity
already has cached and validated, rather than talking to X directly.
Also some preparatory work for extending the monitor API by
clarifying the name of the (current) main entry point.
Change the "overlay" actor to be a group of 4 actors that we can
rearrange so as to have a hole in the middle (to cover up the whole
screen except for the highlighted icon). For non-iconified windows, we
still highlight them the old way (raising them above the overlay),
because we don't want square highlights around shaped windows.
Make the indicator in the pop-up move faster.
Quickly fade in the "overlay" window when starting, rather than
showing it abruptly. Destroy it right away rather than just hiding it
when the AltTabHandler is destroyed.
Fix the font size to be the same as the "Activities" button.
Fix a warning when tabbing past an iconified window.
To avoid loading applications from two different systems, use
ShellAppSystem solely. This unifies the initial load and the
reload.
Extend ShellAppSystem to also load settings/preferences, and
ensure they appear in the search.
Drag and drop was not working since the (non-reactive, invisible)
chrome group was on top of most of the screen. Explicitly set
a 0x0 size on that group to resolve the problem.
We need to include the icon size in the data we pass to on_pixbuf_loaded()
so that we can include it in the key we use when caching the icon
Currently we save icons in the cache at size 0 but look them up at size 48
so no caching occurs.
http://bugzilla.gnome.org/show_bug.cgi?id=583503
Recent gjs changes have made it log most types of messages (including
log() calls) only to a configured log file and not to stderr. So,
set GJS_DEBUG_OUTPUT=stderr to make *stderr* the log file and get
our log() calls back.
The goal of the workspace view is to identify and select windows. Transparency
results in the blending of the window and the background (and icons on the
desktop). I think the transparency is counterproductive.
We used it in the initial mockups kind of as a lark - it wasn't well thought
out.
http://bugzilla.gnome.org/show_bug.cgi?id=582647
Add a ShellTextureCache class which loads (and can cache)
pixmap->texture conversions. This fixes a problem with the
async code in ClutterTexture that it was lower priority
than animations, and also ensures we're really only
loading these pixbufs once in the icon case.
Adds an explicit "chrome" layer for the panel (and later the sidebar),
managing the input region and struts for them, and hiding them when
fullscreen windows are present or the user enters the overlay.
http://bugzilla.gnome.org/show_bug.cgi?id=581771
We need to reset the popularity GSList* at start, and everytime we load data about a another activity. This also implies fixing the typo about (activity != -1).
Thanks to Dan Winship. http://bugzilla.gnome.org/show_bug.cgi?id=582215
In addition to pressing the Activities button, the overlay can be entered
by pressing the System key or Alt+F1. We want the button to look pressed in
in these cases too.
Rather than having main.js manage this, put it into overlay.js, and
have the overlay object emit signals that other code can watch to do
things when the overlay is showing/shown/hiding/hidden.
If a file already has a failed thumbnail indicator, don't try to
thumbnail it, and if thumbnailing fails, create a failed thumbnail
indicator. (In particular, so that if thumbnail creation times out, we
don't keep retrying and eventually timing out every time the user
opens the overlay.)
http://bugzilla.gnome.org/show_bug.cgi?id=568081
Using appExec for the last application that registered the document was failing
in certain cases, such as for the Open Office for which the application
execution string is being registered as "soffice %u" in ~/.recently-used.xbel.
In general, using the default application for the mime type seems to be a more
predictable way to open documents. We still fall back to using the appExect
for the last application if the default application for a given mime type was
not found.
