[tex-live] running programs from texlive tlc2

[David Kastrup]

Fabrice Popineau <Fabrice.Popineau at supelec.fr> writes:

>> Still, regardless of the code, enormous differences in processing
>> speed between linux and Win32 are surprising and suggest some
>> serious hardware or configuration problem.
>
>> Are these differences apparent for other documents or only Hebrew?
>
> Sorry, but all this stuff is just unbelievable. I used to do
> profiling on TeX on win32 just to be sure that there were no place
> where it could be made faster. TeX is not system dependent as soon
> as you use native binaries. The only case where you can find a
> significantly slower TeX is when you use djgpp binaries under
> windows. If you use either miktex or web2c natively compiled, I see
> no reason why it should run slower under win32 than TeX under linux
> under the same hardware assumptions. Kernel has nothing to do
> here. Even the C library is not heavily involved in the logs I got
> from profiling.

Well, a few things I could imagine:

a) different executables: one version eTeX-based, one version
   TeX-based, and the latter uses slower code.
b) Heavily I/O bound operations, like writing temporary data out to a
   file on a per-line or per-character base and immediately reading it
   back in again.  Probably avoidable with eTeX's token reparsing
   operations.
c) The same in connection with a defective operation of the file name
   data base.

In all of those possibly OS-dependent cases (except for some
variations of a) I can think of, however, the generated .log file
would become very large, and cutting out a "typical" passage from it
should give a good first pointer about what might be contributing
towards the slowdown (file operations are usually visible in the log).
And even the eTeX-or-not question is answered by the start of the log
file.

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum