Bug#581179: oowriter 3.2 loads all linked images on first save (regression)
On Fri, May 21, 2010 at 02:18:47AM +0200, Rene Engelhard wrote:
> Hi,
>
> On Sun, May 16, 2010 at 03:25:36AM +0200, Bernard Jungen wrote:
> > > True. Then please send me a testdoc with files and I'll see.
> >
> > Here they are (very small):
> > http://home.euphonynet.be/bjung/test_quick_save.odt
> > http://home.euphonynet.be/bjung/big.png
>
> OK, so this "takes long to save" story is more a "consumes CPU" one.
>
> I mean, OK, cpu usage goes up to over 80% from the baseline of ~50 I had
> and on the second etc. save it stayed more or less the same around that
> 50, but the actual save in OOo is still quick (or appears to be while
> doing stuff in the background), no?
>
> In the latter case: why is this not "minor" then? :)
With all due respect, I'm really amazed at the lack of common sense displayed
in your response. You're not a newbie at OOo and computer stuff, are you?
First, your machine is probably too fast for this specific test. I don't
(and am not supposed to) know how you measure CPU usage and how many GHz and
cores it has, but here on a single-core 2.2 Ghz K8 (a reasonably fast machine
IMO), the "slow" save takes ~4 seconds, with CPU usage at 100%, and actually
blocks OOo while doing it. "Quick" save takes less than 0.25 second. Significant
difference. A good tester is aware of his hardware and its potential effects
on tests.
Second, the test just intends to prove that images are loaded, while keeping
the test file sizes small. But imagine (or actually try) a document with
10 times more images and whose file sizes are several megabytes each (i.e.
real images that require much more CPU power than blank ones). In that case,
not only does CPU usage dramatically goes up while loading them, but I/O
activity (and the consequent memory thrashing) too. Anyone who already wrote
serious OOo documents containing print-quality images has a good idea of
resource consumption when OOo handles images.
In this special case, all that waste of resources is unnecessary, and may be
a pain depending on machine power and document kind. Is the problem clear now?
Reply to: