After an evaluation, GNOME has moved from Bugzilla to GitLab. Learn more about GitLab.
No new issues can be reported in GNOME Bugzilla anymore.
To report an issue in a GNOME project, go to GNOME GitLab.
Do not go to GNOME Gitlab for: Bluefish, Doxygen, GnuCash, GStreamer, java-gnome, LDTP, NetworkManager, Tomboy.
Bug 647778 - applications loading isn't fast enough
applications loading isn't fast enough
Status: RESOLVED OBSOLETE
Product: gnome-shell
Classification: Core
Component: general
3.2.x
Other Linux
: Normal enhancement
: ---
Assigned To: gnome-shell-maint
gnome-shell-maint
: 613814 647825 649102 651879 663754 663841 delay-bug 669992 675073 684112 684113 690298 690894 (view as bug list)
Depends on: pango-threadsafe
Blocks:
 
 
Reported: 2011-04-14 14:28 UTC by G. Michael Carter
Modified: 2021-07-05 14:09 UTC
See Also:
GNOME target: ---
GNOME version: ---


Attachments
appDisplay: Create application actors incrementally (1.87 KB, patch)
2011-04-23 15:56 UTC, Colin Walters
needs-work Details | Review
appDisplay: Process application display in chunks instead of all at once (2.83 KB, patch)
2011-04-25 19:42 UTC, Colin Walters
needs-work Details | Review
collection of stack traces (210.48 KB, text/plain)
2011-04-25 19:57 UTC, Colin Walters
  Details
appDisplay: Process application display in chunks instead of all at once (3.01 KB, patch)
2011-04-25 20:52 UTC, Colin Walters
committed Details | Review
duplicated items (382.52 KB, image/png)
2011-05-03 09:52 UTC, Gary Ching-Pang Lin
  Details
Force loading high-res bitmaps for non-recolored icons (1.33 KB, patch)
2012-10-05 09:41 UTC, Bastien Nocera
reviewed Details | Review
Force loading high-res bitmaps for non-recolored icons (1.33 KB, patch)
2012-10-11 09:54 UTC, Bastien Nocera
none Details | Review

Description G. Michael Carter 2011-04-14 14:28:16 UTC
When you first click on the applications tab you can sit for a good 5 seconds with a blank screen.  I'm assuming this is search the HD for .desktop files.   

Since the default is the "All" category I imaging this will get worse with slower hard drives and more applications.

As a suggestion, could you show a waiting icon or load the icons as they are found.  I've seen a few users think the gnome-shell is broken at that point.
Comment 1 Dan Winship 2011-04-15 13:18:19 UTC
*** Bug 647825 has been marked as a duplicate of this bug. ***
Comment 2 Colin Walters 2011-04-22 12:47:40 UTC
Okay, so I have this bug on my netbook for sure (Fedora 15 with partial jhbuild).  Admittedly I didn't do any profiling at first and just guessed it was I/O, and ended up massively cleaning up the application code in bug 648149.

Didn't help a bit.  So I sat down to actually profile and debug this.  Now, one thing I observed is that if I did Alt-F2 r, each time it would be very slow, but I didn't really observe the hard disk light much at all.  So it isn't I/O.

Not disk?  Okay, must be CPU then.  Fire up perf.  Nothing stands out - the biggest CPU usage is 4% in librsvg, and the rest is a large scattering of 1-2% and less.

Not disk, not CPU bound?  Hmm...we must be blocking on some system call then.  Owen hypothesized that we were spending a lot of time uploading textures - sounds very plausible.

I haven't yet looked for tools to debug DRM specifically, but I thought I'd try strace.  Here's what I used:

mkdir /tmp/gs-apps; rm -f /tmp/gs-apps/*; strace -ttt -o /tmp/gs-apps/trace -ff -T -p $(pidof gnome-shell)

So the results are very interesting and surprising - we're not blocking on DRM ioctls.  The pixbuf loading threads are blocking for on the order of *seconds* on a mutex.

In other words, the problem is thread contention.  So I wrote a quick script to parse these strace results and sort the system calls by latency.  Here's the result on one of the StTextureCache worker threads:

syscall: 'futex' args: '0x940ce4c, FUTEX_WAIT_PRIVATE, 7, {14, 999511941}' result: '0' latency: 1.068685
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 260, {0, 499746524}' result: '-1 ETIMEDOUT (Connection timed out)' latency: 0.499973
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 47, {0, 488190520}' result: '-1 ETIMEDOUT (Connection timed out)' latency: 0.488351
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 192, {0, 499720505}' result: '0' latency: 0.265901
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 251, {0, 499665123}' result: '0' latency: 0.149814
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 33, {0, 499751269}' result: '0' latency: 0.143782
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 233, {0, 499725599}' result: '0' latency: 0.116903
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 217, {0, 499642591}' result: '0' latency: 0.107817
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 239, {0, 494809806}' result: '0' latency: 0.102327
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 70, {0, 499673843}' result: '0' latency: 0.101916

Note that it blocked for a full *second* on a mutex we'll call "mutex A", and repeatedly blocked on "mutex B" for a full tenth of a second.

Here are the results from a second pixbuf worker thread:

syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 263, {0, 499750266}' result: '-1 ETIMEDOUT (Connection 
timed out)' latency: 0.499968
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 194, {0, 499776663}' result: '0' latency: 0.280369
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 106, {0, 499652374}' result: '0' latency: 0.102555
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 236, {0, 499638466}' result: '0' latency: 0.100484
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 221, {0, 499721528}' result: '0' latency: 0.096608
syscall: 'futex' args: '0x9336094, FUTEX_WAIT_PRIVATE, 186, {0, 499662017}' result: '0' latency: 0.092422

Note the two worker threads are heavily contending for mutex B!  The second thread appears not to block on mutex A though.

So what we really need now are stack traces.
Comment 3 Colin Walters 2011-04-22 12:58:02 UTC
Without stack traces yet, one thing that does occur to me as a possibility is pango; see for example pango/pango/pangocairo-render.c:_pango_cairo_do_glyph_string.  It grabs a lock for the full duration of drawing.  We know this happens both in the main thread as well as the worker threads when they're loading an SVG.
Comment 4 Colin Walters 2011-04-23 15:37:57 UTC
So, I changed StTextureCache to not create new threads, and we spend the same total amount of time (about 5 seconds here).  So bad mutex locking isn't the issue.

If I rerun strace again, I see basically only the main thread active, and we spend *long* amounts of time blocked in poll():

'_llseek' total time: 0.04905699999999935
'read' total time: 0.058808999999999834
'gettimeofday' total time: 0.20409699999998782
'ioctl' total time: 0.26747200000000637
'poll' total time: 5.0851340000000045

The active fd returned each time is 7, which I'm pretty sure is the X server.

As far as CPU now, sysprof shows ~15% of our time doing CSS style matching, 5% scaling pixbufs, and a lot of 1-2% scattered around.

It could be that we're just doing too much work in one go.
Comment 5 Colin Walters 2011-04-23 15:56:16 UTC
Created attachment 186522 [details] [review]
appDisplay: Create application actors incrementally

This is a hack to work around how slow we are at processing apps
right now.  We process 5 applications at a time with a 150ms delay,
which makes the system feel *much* more interactive.
Comment 6 Colin Walters 2011-04-23 15:59:32 UTC
The above is sort of a hack; creating even 100 application actors shouldn't be slow.  But still, it's low risk, high gain for now - 3.0.1 material I think.
Comment 7 Owen Taylor 2011-04-24 22:44:56 UTC
Strangely, though this was definitely happening for me earlier, now, even if I 'echo 3 > /proc/sys/vm/drop_caches' and restart the shell, switch to applications is more or less instant (certainly less than 0.25 seconds... obviously my laptop is much faster than your netbook, but that wouldn't explain the differences between what I was seeing earler, and this) That makes me wonder if the cause for the problems isn't some particular bad icon that is in SVG format or something.

Certainly if you have a reproducible long hang, it should be tracked down while you are reproducing it.

 - Can you make it hang every time you switch to the overview by dumping the StIconCache, destroying the application tab, or dropping other cached information we have around?

 - If you can make it hang every time, can you write JS code to make the shell go into an infinite loop of switching tabs (maybe using the perf code, you can see examples of switching tabs already there), then get backtraces of where it's spending it's time by ssh'ing and gdb'ing?
Comment 8 Owen Taylor 2011-04-25 00:36:04 UTC
Review of attachment 186522 [details] [review]:

So, on my system, this takes the display that comes up instantly (small fraction of the second) and instead makes the apps pop in in 5 app chunks over ~2 seconds (66 apps / 5 apps per chunk * 0.150 seconds per chunk) ... this looks way worse.

If anything, something like:

        this._pendingAppIdleId = Meta.later_add(Meta.LaterType.BEFORE_REDRAW, Lang.bind(this,this._addPendingApps));

add 5 apps per *frame* makes a lot more sense to me than adding 5 apps, pausing for an arbitrary 150ms, then adding 5 more apps, and so forth. But it's still pretty visible that the apps fill in incrementally.

If the above doesn't work for you, then that's interesting in itself, because it might mean whatever is causing the blocking:

 - Isn't the creation of the actor (which is happening in a limited fashion per frame)
 - Isn't loading the icon (which shouldn't block anything, even if it's slow)
 - But is the work that is being done at completion of the load thread (which could all collapse together from multiple frames if the loads finished at the same time)

Perhaps clutter st-texture-cache.c:pixbuf_to_cogl_handle()
Comment 9 Colin Walters 2011-04-25 14:53:20 UTC
(In reply to comment #7)
> Strangely, though this was definitely happening for me earlier, now, even if I
> 'echo 3 > /proc/sys/vm/drop_caches' and restart the shell, switch to
> applications is more or less instant (certainly less than 0.25 seconds...
> obviously my laptop is much faster than your netbook, but that wouldn't explain
> the differences between what I was seeing earler, and this) That makes me
> wonder if the cause for the problems isn't some particular bad icon that is in
> SVG format or something.

How many apps do you have installed? 

Shell.AppSystem.get_default().get_flattened_apps().length
 
> Certainly if you have a reproducible long hang, it should be tracked down while
> you are reproducing it.

That's what I was trying to do with the above.

>  - Can you make it hang every time you switch to the overview by dumping the
> StIconCache, destroying the application tab, or dropping other cached
> information we have around?

I'll work on this.
Comment 10 Kristoffer Kleine 2011-04-25 15:06:14 UTC
When I hit the Applications tab the first time it takes about 1.5 seconds until it is shown (though it feels longer).

I have 135 apps shown there and my system specs are:
Intel Core2 Duo CPU E8400 3.00GHz
Nvidia GeForce 9600 GT with 512mb memory
Comment 11 Dan Winship 2011-04-25 15:08:16 UTC
I get the "hang" too. Not 5 seconds, but 2-3. Intel video, 107 apps.

It doesn't appear to be related to filesystem caching at all; if I go to the apps view, then immediately restart gnome-shell and go the apps view again, it's slow the second time too.
Comment 12 Colin Walters 2011-04-25 15:20:47 UTC
The question now is: does this scale roughly linearly with number of apps, or are particular types of apps (e.g. SVG icons?) more likely to slow things down?
Comment 13 G. Michael Carter 2011-04-25 15:24:23 UTC
I added the seconds to my clock and in the apps view I noticed the clock stops for the 5 or so seconds.

Don't know if that's any help tracking down the issue.
Comment 14 Owen Taylor 2011-04-25 17:06:58 UTC
(In reply to comment #9)
> (In reply to comment #7)
> > Strangely, though this was definitely happening for me earlier, now, even if I
> > 'echo 3 > /proc/sys/vm/drop_caches' and restart the shell, switch to
> > applications is more or less instant (certainly less than 0.25 seconds...
> > obviously my laptop is much faster than your netbook, but that wouldn't explain
> > the differences between what I was seeing earler, and this) That makes me
> > wonder if the cause for the problems isn't some particular bad icon that is in
> > SVG format or something.
> 
> How many apps do you have installed? 
> 
> Shell.AppSystem.get_default().get_flattened_apps().length

98

So, not way less than what other people are reporting. For a demo, I made some attempt to remove apps with bad/ugly icons, so it's possible that if it is a problem with particular a particular icon or icons, I might have removed them.
Comment 15 Colin Walters 2011-04-25 19:42:46 UTC
Created attachment 186613 [details] [review]
appDisplay: Process application display in chunks instead of all at once

Layout here can be slow for multiple reasons; better to avoid blocking
the whole UI until we're done with layout.
Comment 16 Colin Walters 2011-04-25 19:51:27 UTC
So after debugging this more, we don't think there is a smoking gun.  

Some individual issues:
https://bugzilla.gnome.org/show_bug.cgi?id=627083
https://bugzilla.gnome.org/show_bug.cgi?id=648638
Comment 17 Colin Walters 2011-04-25 19:57:14 UTC
Created attachment 186614 [details]
collection of stack traces

For reference, a collection of stack traces.
Comment 18 Owen Taylor 2011-04-25 20:35:21 UTC
Review of attachment 186613 [details] [review]:

This seems to work OK for me - there are definitely some visual artifacts as the apps load - for me I get labels appearing first, then icons then the scrollbar resizes, but it's likely better than a hang.

::: js/ui/appDisplay.js
@@ +115,3 @@
+    _addPendingApps: function() {
+        var i;
+        var startTimeMillis = new Date().getTime();

Our coding standard is to use 'let' .. with 2000+ uses of let in our code, it doesn't make sense to use var in one place even if you are the opinion that we should be using var and avoiding non-standardize extensions.

@@ +118,3 @@
+        for (i = 0; i < this._pendingAppIds.length; i++) {
+            var currentTimeMillis = new Date().getTime();
+            if (currentTimeMillis - startTimeMillis > 75) {

this should really be a constant broken out rather than a random number embedded into the code.

@@ +127,3 @@
+        if (this._pendingAppIds.length > 0) {
+            this._pendingAppIdleId = Meta.later_add(Meta.LaterType.BEFORE_REDRAW,
+                                                    Lang.bind(this, this._addPendingApps));

shouldn't you just return true here, rather than adding a new id, then implicitly removing the later by return None which is boolean false?

@@ +148,3 @@
+        this._pendingApps = apps;
+        if (this._pendingAppIdleId)
+            Mainloop.source_remove(this._pendingAppIdleId);

Not a mainloop source

@@ +149,3 @@
+        if (this._pendingAppIdleId)
+            Mainloop.source_remove(this._pendingAppIdleId);
+        this._pendingAppIdleId = Meta.later_add(Meta.LaterType.BEFORE_REDRAW,

pendingAppLaterId or addPendingAppsId or something, it's not an idle.
Comment 19 Colin Walters 2011-04-25 20:52:18 UTC
Created attachment 186618 [details] [review]
appDisplay: Process application display in chunks instead of all at once

Updated for comments
Comment 20 Owen Taylor 2011-04-25 21:16:18 UTC
Review of attachment 186618 [details] [review]:

Just two minor stylish things. Good to commit with these fixed.

::: js/ui/appDisplay.js
@@ +120,3 @@
+            let currentTimeMillis = new Date().getTime();
+            if (currentTimeMillis - startTimeMillis > MAX_APPLICATION_WORK_MILLIS) {
+                break;

Very minor points, but excess brackets and this really should be at the end of the block not the beginning to save an extra syscall ... we always want to do at least one app.
Comment 21 Colin Walters 2011-04-25 21:20:22 UTC
Comment on attachment 186618 [details] [review]
appDisplay: Process application display in chunks instead of all at once

Attachment 186618 [details] pushed as 7813c5b - appDisplay: Process application display in chunks instead of all at once
Comment 22 Colin Walters 2011-04-25 21:20:54 UTC
Keeping this bug open, since I'm not really happy with the speed as is now.
Comment 23 Rui Matos 2011-05-01 18:47:48 UTC
*** Bug 649102 has been marked as a duplicate of this bug. ***
Comment 24 Gary Ching-Pang Lin 2011-05-03 09:52:14 UTC
Created attachment 187112 [details]
duplicated items

I found some items in Applications are duplicated after I updated gnome-shell to 3.0.1 (see the attachment). After reverting the commit, the menu was normal again.
Comment 25 Colin Walters 2011-05-03 12:53:45 UTC
(In reply to comment #24)
> Created an attachment (id=187112) [details]
> duplicated items
> 
> I found some items in Applications are duplicated after I updated gnome-shell
> to 3.0.1 (see the attachment). After reverting the commit, the menu was normal
> again.

Hi Gary,

This is fixed by:
http://git.gnome.org/browse/gnome-shell/commit/?h=gnome-3-0&id=569008b08417d50b899266ba7c3eba4aee77c032
Comment 26 Gary Ching-Pang Lin 2011-05-04 02:45:45 UTC
Oh! Thanks! It works for me :)
Comment 27 Rui Matos 2011-06-05 11:12:50 UTC
*** Bug 651879 has been marked as a duplicate of this bug. ***
Comment 28 Rui Matos 2011-11-10 13:55:26 UTC
*** Bug 663754 has been marked as a duplicate of this bug. ***
Comment 29 flamma 2011-11-10 15:16:23 UTC
My request (663754) has been marked as duplicated of this. I don't think is quite the same. I asked the application search to be asyncronous, so typing don't halt while searching for the first one or two letters. Maybe only looking half a second after your last typed char. I think that would better the user experience.
Comment 30 Milan Bouchet-Valat 2011-11-10 15:34:07 UTC
(In reply to comment #29)
> My request (663754) has been marked as duplicated of this. I don't think is
> quite the same. I asked the application search to be asyncronous, so typing
> don't halt while searching for the first one or two letters. Maybe only looking
> half a second after your last typed char. I think that would better the user
> experience.
Colin said above he wanted to keep this bug open to deal with issues related to performance. No need to have several bugs open about that: you share the same goal of making the search responsive, and developers will choose the best solution to do that.
Comment 31 Milan Bouchet-Valat 2011-11-11 11:24:31 UTC
*** Bug 663841 has been marked as a duplicate of this bug. ***
Comment 32 Saulo Toledo 2011-11-11 20:01:21 UTC
I think this can be a solution:

1) At first time Gnome is opened, a file in a tree format is writen with the application load. If the file ("cache") content store a fast data structure, like a B-Tree, for example, will be fast to be loaded and updated.

2) A background proccess running to check the .desktop files change and update the cache file.

I think it's an easy way, and the slow it's only at first time Gnome is used (KDE does something like this too). After this, the load and the update will be better than read the files one by one.
Comment 33 Jasper St. Pierre (not reading bugmail) 2011-11-11 20:08:35 UTC
I believe gmenu already keeps its own cache around. The slow thing is loading the pixbufs and actors.
Comment 34 Rui Matos 2011-11-19 19:12:04 UTC
*** Bug 664359 has been marked as a duplicate of this bug. ***
Comment 35 Milan Bouchet-Valat 2012-02-13 15:13:38 UTC
*** Bug 669992 has been marked as a duplicate of this bug. ***
Comment 36 Christian Kirbach 2012-03-31 20:09:01 UTC
I can confirm that the delay is definitely not caused by I/O waiting.

Restarting the gnome shell and then opening the applications tab exhibits the noticeable delay again, but only for the first time opening.
Comment 37 Milan Bouchet-Valat 2012-04-29 15:04:31 UTC
*** Bug 675073 has been marked as a duplicate of this bug. ***
Comment 38 Saulo Toledo 2012-05-04 22:41:21 UTC
In my opinion, this problem is annoying.
The response time of the applications lists exceed any accepted limit for software  developments.

Programmers don't worry about this questions, but users, even if they do not know, reject Gnome each time they open the application list.

Isn't me who affirm this, any software usability text talk something about the accepted response time.

I'm posting this because I don't see importance to this problem since 2011-11-11. This problem come with Gnome since the first release! New users form their opinions at first time they use. How many potential users we possible have lost? How many time Gnome 3 is out?

My suggestion: if we can't solve the load problem, put a load there! The load will give users the imediate response, and this will be better than click and don't know if is working or not. The effect of this load at users will be significant.

I have no experience for now to submit a patch to Gnome, then, unfortunately, my contribution end here, where starts, but I will be here until see this problem fixed.
Comment 39 Star Brilliant 2012-05-06 14:17:22 UTC
I don't think waiting for 10 or 11 seconds is annoying,
however, there should be a label and a spinning wheel showing 'loading...' to tell the user to be patient.

From the comments above, I found that drawing the icon costs the most time.
Why not load the app screen with a 'loading' icon instead and draw the real icon in background a few seconds later?

The most import thing is to be user-friendly, I think.
Comment 40 Allan Day 2012-08-23 11:36:45 UTC
*** Bug 613814 has been marked as a duplicate of this bug. ***
Comment 41 Florian Müllner 2012-09-15 19:36:42 UTC
*** Bug 684113 has been marked as a duplicate of this bug. ***
Comment 42 ferrazrafael 2012-09-15 21:42:48 UTC
I think that wait 10 or 11 seconds is annoying, no other system takes that long, the user will complain about that.
Comment 43 Jasper St. Pierre (not reading bugmail) 2012-09-15 21:44:44 UTC
Right now I believe this is caused by SVG, which should be fixed when we get librsvg off the main thread, which will be fixed when pango is thread-safe.
Comment 44 Milan Bouchet-Valat 2012-09-16 09:40:19 UTC
*** Bug 684112 has been marked as a duplicate of this bug. ***
Comment 45 ferrazrafael 2012-09-16 18:28:53 UTC
Jasper, this would fix it even for slow cpus? Like netbooks or older hardware?
Comment 46 Jasper St. Pierre (not reading bugmail) 2012-09-16 18:34:30 UTC
As far as I'm aware SVG is the major bottleneck of the overview.
Comment 47 Bastien Nocera 2012-10-05 09:41:43 UTC
Created attachment 225866 [details] [review]
Force loading high-res bitmaps for non-recolored icons

A PoC hack to show that we could load high-res bitmaps concurrently
instead of blocking on loading SVGs consecutively.

A final version of the patch would pass down the fact that we're loading
menu icons and we'd force using bitmaps there.
Comment 48 Jasper St. Pierre (not reading bugmail) 2012-10-05 15:39:57 UTC
Review of attachment 225866 [details] [review]:

::: src/st/st-texture-cache.c
@@ +943,3 @@
 
+  flags = GTK_ICON_LOOKUP_USE_BUILTIN;
+  if (colors)

I think this is supposed to be !colors?
Comment 49 Bastien Nocera 2012-10-05 15:51:53 UTC
(In reply to comment #48)
> I think this is supposed to be !colors?

Yes. Can you tell it's completely untested? :)
Comment 50 Bastien Nocera 2012-10-11 09:54:55 UTC
Created attachment 226243 [details] [review]
Force loading high-res bitmaps for non-recolored icons

A PoC hack to show that we could load high-res bitmaps concurrently
instead of blocking on loading SVGs consecutively.

A final version of the patch would pass down the fact that we're loading
menu icons and we'd force using bitmaps there.
Comment 51 Jasper St. Pierre (not reading bugmail) 2012-10-11 12:21:05 UTC
Did you test this to ensure it improved performance?
Comment 52 Bastien Nocera 2012-10-11 12:56:05 UTC
(In reply to comment #51)
> Did you test this to ensure it improved performance?

Still not.
Comment 53 Milan Bouchet-Valat 2012-12-16 17:23:35 UTC
*** Bug 690298 has been marked as a duplicate of this bug. ***
Comment 54 Rui Matos 2012-12-30 23:03:04 UTC
*** Bug 690894 has been marked as a duplicate of this bug. ***
Comment 55 entropy 2012-12-30 23:21:04 UTC
I have an Intel 1,66MHz CPU. It can't be a CPU issue. Previous GNOME 3 versions allowed me to search for apps with easiness. Now when I look for an app I have to wait for a while...and sometimes the display freezes and I can't do anything but button-shutdown it. It's surely a gnome-shell bug :)
Comment 56 Christian Kirbach 2013-01-09 10:57:03 UTC
entropy, your issue is not what this problem report is all about.

Testing (Ubuntu) GNOME 3.6 on my old 1.6ghz single-core laptop, I can say the user experience has improved overall when compared to GNOME 3.0 / 3.2. The bottleneck now is opening the activities overview for the first time when there is still some delay you can notice on old hardware. However the delay used to be at least twice of what it is today (judging without a stopwatch).
Comment 57 entropy 2013-01-09 11:20:31 UTC
Christian,

Someone marked my bug as a duplicate of this bug.
Comment 58 ferrazrafael 2013-03-15 16:48:53 UTC
testing on GNOME 3.6 on fedora 18 I can see a improvement, but for a such basic feature it could be faster. Considering the hardware Im running GNOME.

Intel Core i5 2.4~3.0Ghz, 4gb DDR3, dedicated nvidia GPU.
Comment 59 GNOME Infrastructure Team 2021-07-05 14:09:11 UTC
GNOME is going to shut down bugzilla.gnome.org in favor of  gitlab.gnome.org.
As part of that, we are mass-closing older open tickets in bugzilla.gnome.org
which have not seen updates for a longer time (resources are unfortunately
quite limited so not every ticket can get handled).

If you can still reproduce the situation described in this ticket in a recent
and supported software version, then please follow
  https://wiki.gnome.org/GettingInTouch/BugReportingGuidelines
and create a new ticket at
  https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/

Thank you for your understanding and your help.