[Texmacs-dev] Improved facilities for benchmarking

[Joris van der Hoeven]

In timer.hpp/timer.cpp on the CVS you will find some simpler
routines for benchmarking parts of TeXmacs. The scheme routines
bench-print and bench-print-all can be used for interactive
benchmarking. Some further results for loading and article of mine:

Timing ] Task 'decode pk' took 516 ms (67 invocations)
Timing ] Task 'decode tfm' took 2 ms (66 invocations)
Timing ] Task 'load file' took 283 ms (441 invocations)
Timing ] Task 'read directory' took 2 ms (209 invocations)
Timing ] Task 'stat' took 100 ms (3001 invocations)

It seems that an incremental implementation of pk font decoding
in load_pk.cpp (similar to what we already do for TrueType fonts)
would allow for a gain of the order of half a second (on a 2.4GHz
computer), since most characters are not actually used. If someone
wants to do that, here is a useful reference:

        http://tex.loria.fr/general/dvistd0.dvi

I still do not understand where the main bottleneck lies though,
but this can now be investigated in more detail. The article takes
about a second to be typeset (update buffer), but about 12 seconds
to be loaded for the first time...