Re: ggv/gpdf and evince

[Kristian Høgsberg <krh redhat com>]

Luca Ferretti wrote:
> Il giorno gio, 09/06/2005 alle 13.08 -0400, Luis Villa ha scritto:
>
> > I think there seems to be a very strong consensus that ggv/gpdf should
> > be replaced by evince in 2.12. Does anyone object to me going ahead
> > and removing them from the jhbuild moduleset and adding evince?
>
>
> Wait. There are some troubles with poppler+cairo on head :-(
>
> I didn't yet report them, 'cause I'm still performing some tests and
> collecting info on not well behaving PDF files, but it seems there is a
> performance regression in recent CVS builds.

Please, if you come across PDF files that render slowly or incorrectly, 
put them in poppler bugzilla as you go, so we can start working on them.

> The most critical PDF file is this[1] introduction to signal processing
> in Scilab (and maybe all PDFs from the same location): it locks the
> poppler benckmark application at page 2 (!!!) and leaves the evince view
> empty. There are some slowness with some PDFs[2] from Apple website[3]
> too. I _suppose_ it could be a cairo issue, because I remember that all
> works fine in pre cairo-0.5.0 days.

The current slowdown with CVS head comes from the fact that clipping 
didn't actually work until recently.  Right now we have a few issues 
with cairo poppler that will slow things down a lot:

  clipping - this hits a couple of libpixman combinations that don't 
have a fast path;

  images - since most images in PDFs are scaled as they are rendered to 
screen they also escape all the fast paths in libpixman.

and page 2 in the Scilab document hits both of these.  Both issues are 
fixable.

> Please note that I think we should include evince in GNOME, replacing
> gpdf and ggv. I'm only saying that maybe is better disable by now
> poppler's cairo backend: this will produce sometimes a less sexy
> rendering, but it's fast and with no faults.

As Marco points out, we have the possibility to bail out and re-enable 
the splash backend at any time, should we decide that the cairo backend 
is not ready.  In the meantime, I'd like to enable the cairo backend by 
default so it gets a lot of testing.  People are already using poppler 
with cairo and logging bugs about slow PDFs or incorrect rendering, and 
the current cvs head has improved much based on those reports already. 
I'd love to see more of this so we can close the gap between the cairo 
backend and the splash backend and hopefully make cairo itself faster in 
the process.

> As attachment there are my records about poppler benchmarks: I'll report
> them on poppler bugzilla.
>
> [1] ftp://ftp.inria.fr/INRIA/Scilab/documentation/pdf/signal.pdf
> [2] http://images.apple.com/server/pdfs/Workgroup_Manager_TB_v10.4.pdf
> [4] http://www.apple.com/server/resources/

Great, I'll take a look.

cheers,
Kristian