[cairo] text measurement in perf?

[Federico Mena Quintero]

On Tue, 2007-05-29 at 19:35 -0400, Behdad Esfahbod wrote:

> > So, is the actual problem that the per-scaled-font 256-element cache is
> > too small for CJK?
> 
> I believe so, but I've not tested it.

BTW, is the old pango-profile module still useful?

- It has a nice corpus, if a bit quirky.  We should probably strip the
"%s" and other weird substrings to avoid getting meaningless / bidi
runs.

- It doesn't really take timings which you can track over time (you
could time(1) an invocation for a single language, though...)

Also, shouldn't any kind of caching be done in Pango rather than in
Cairo or below?  Pango is the layer that has the most information about
how data is being used; AFAIK Cairo only has 
"gimme_metrics(font, glyph_no)".

>   Also worth noting is that it may
> not be a real problem for real scenarios.

Hehe, but it's one hell of a regresson :)  What would a good test be?
Fill a treeview with CJK text and see if it scrolls nicely?

One useful thing when profiling GtkFileChooser was to hack Pango to
write a file with all the strings that it is asked to process.  That way
I had real data, which I could feed later into a benchmark easily ("how
much time does it take to Pangoify all the strings that the file chooser
needs?").

  Federico