[sugar] [PATCH] screenshots hurt

[Erik Garrison]

On Thu, Sep 04, 2008 at 10:42:32AM +0200, Tomeu Vizoso wrote:
> On Thu, Sep 4, 2008 at 7:29 AM, Erik Garrison <erik at laptop.org> wrote:
> > ....
> >
> > The attached patch to Sugar (most recent git head, 48dd24c6e) does
> > simply that.  Andrés Ambrois and I have tested the changes in question
> > on a machine running 656 and on a machine running 2263 (the most recent
> > joyride I have available-- code in question is identical to the current
> > sugar git head).  We noted a considerable improvement in the speed and
> > consistency of activity switching performance.
> >
> > These qualitative results are consistent with prior user interface
> > profiling tests by Riccardo Lucchese, which have established that
> > screenshot functionality occupies just shy of 50% of CPU time during
> > activity switching [1].
> >
> > ...
> 
> The patch looks good in that it does what intends and doesn't seem to
> have any unintended effect. About the convenience of totally
> deactivating screenshots in the next release, I think it's a decision
> that needs to be taken on non-technical grounds. Are we going to
> better serve our users' interests by doing so? Is the improvement of
> perceived performance worth the lost functionality? Greg, can you help
> on this?
> 

I agree that this is not a technical issue in that the effects are
purely important from a user-experiential point of view, and there is no
specific system failure which the patch addresses.

Removing the screenshots and the nondeterministic latencies which their
creation brings to window manager navigation is the most immediate and
effective measure that I have identified to improve the performance of
Sugar.

I'm really only interested in the responsiveness (in speed and
consistency) of the system.  When people say that Sugar or the XO is
'slow', they are not talking about in-activity performance.  They are
expressing that the process of navigating the user interface, moving
between screens, and starting and stopping activities is inconvenient
because it is so slow and inconsistent.

With the screenshots turned off, it becomes very clear that the general
minimalism of most activities gives the system a huge performance boost
relative to 'heavier' UIs such as Gnome (which I have tested on the XO
via debxo).

> Note that I understand this as a temporary measure, for the next
> release I think we should properly address this issue. I have just
> added two tickets on this regard:
 
I do not understand this as a temporary measure as it is completely
unclear why we are taking screenshots at every window manager navigation
event.  What problem are we trying to solve in doing so?  The resultant
screenshots are barely even legible when they appear in the joural.

My opinion is that this should be a 8.2.0 release target.  What do we
need to do to initate the introduction of this patch?  Technically I
cannot identify anything holding this change back from distribution with
the early G1G1 machines.  If need be we can replace the activity screen
previews in the journal with activity icons.

> https://dev.laptop.org/ticket/8303
> https://dev.laptop.org/ticket/8304

Thank you for the tickets.  Note also http://dev.laptop.org/ticket/4365

That said I am still quite unsure of what motivates this work.  On a
resource-constrained system such as the XO, why are we willing to
sacrifice so much responsiveness for so little feature gain?  I am not
capable of working on system performance within the current design
paradigm until I understand the motivational logic of this tradeoff and
can use it to guide further work.

Erik