[tex-live] Patch for "texdoc -s" to use the ls-R database

[Frank Küster]

"George N. White III" <gnwiii at gmail.com> wrote:

> There are still performance problems.  Try to find the latin modern
> docs:
>
> "texdoc -s lmodern" doesn't help, but "texdoc -s lm" takes much too
> long because the last kpathsea search is run for every lm*.tfm file!
>
> What is needed is a way to select only the shortest prefixes from
> "kpsewhich --expand-path='$TEXDOCS'" so that find isn't searching
> the same places over and over:

Here's a new suggestion.  It uses a new switch for the new behavior (-S)
and does the following to reduce the time needed:

I define a NODOC_PATTERN which contains the most abundant extensions of
probably-not-documentation files:

NODOC_PATTERN="\.tfm$|\.afm$|\.enc$|\.pfb$|\.pfa$|\.pfm$|\.vf|\.fd$|\.ttf$|\.htf$|\.mf$|\.otf$|\.[[:digit:]]*pk$"

there's still a for loop over all files found in ls-R, but it kpsewhich
is only called if this pattern does not match.  I think the exact
pattern should finally be made customizable with an option.  I also
check whether a filename appears more than once in the grepped list, and
call kpsewhere instead of kpsewhich in this case.

With the old implementation (texdoc -s lm), I get 


real	1m14.285s
user	0m0.364s
sys	0m1.336s

Using the new one immediately afterwards, I get

real	0m18.237s
user	0m13.577s
sys	0m4.056s

(after a reboot, I even got only 

real	0m16.457s
user	0m10.781s
sys	0m3.312s

)

This is still not lightning fast, but it's a considerable
improvement. Of course, it would be best to grep only through the doc
parts of ls-R, but I don't have any idea how to get that with shell and
simple tools.  Or I reimplement texdoc in Perl - maybe that's even an
improvement for the windows people?

Regards, Frank

-- 
Dr. Frank Küster
Single Molecule Spectroscopy, Protein Folding @ Inst. f. Biochemie, Univ. Zürich
Debian Developer (teTeX/TeXLive)