Re: [Maxima-discuss] regex packages, was: Reading data from header

[Volker v. N. <vol...@gm...>]

I can't do better. Compiling pregexp with sbcl I get a factor of about 10,
i.e. nregex is 10 times faster in this situation.

It turns out that there is a significant speed and space advantage for
nregex. On the other hand, as Michel Talon has pointed out earlier in this
thread, nregex is very limited.
'|' is treated as an ordinary character and in case of e.g. "w{3}"
nregex-compile loops endlessly.

I would like to propose the following. We leave nregex in src and pregexp
in share/stringproc and I will modify the Maxima level functions in
share/stringproc in a way that the user can decide which scanner he or she
wants to use. E.g.

regex_match(regex, string) with pregexp as the default,
regex_match(regex, string, 'pregexp) and
regex_match(regex, string, 'nregex).

pregexp also allows to set start and end positions. These might be passed
in as optional arguments.

The documentation then should emphasize the advantages and limitations of
nregex.

Volker


2015-10-26 21:46 GMT+01:00 Leo Butler <l_b...@us...>:

>
> Volker, I doubt you will get such good results with gcl. Nregex is 31
> times faster:
>
> pregex + gcl 2.6.12
>
> (%i5) unicode_add(["GREEK","LATIN","APL","NON-SPACING"]);
> Evaluation took 32.3900 seconds (34.2200 elapsed)
> (%o5)     [[GREEK, 553], [LATIN, 1478], [APL, 73], [NON-SPACING, 0]]
>
> nregex + gcl 2.6.12
>
> (%i4) unicode_add(["GREEK","LATIN","APL","NON-SPACING"]);
> Evaluation took 1.0400 seconds (1.1200 elapsed)
> (%o4)     [[GREEK, 553], [LATIN, 1478], [APL, 73], [NON-SPACING, 0]]
>
> Leo
>
> Volker van Nek <vol...@gm...> writes:
>
> > I have coded and tested pregexp-replacements for functions using nregex
> > in src/commac.lisp and src/cl-info.lisp.
> >
> > This time I used sbcl.
> > (compile_file doesn't work with gcl 2.6.12 in my gcl-Maxima build)
> >
> > I compiled pregexp.lisp and the replacements by compile_file.
> > I.e. I had simply overwritten some functions in an already build Maxima.
> > (For replacing nregex.lisp by pregexp.lisp in maxima.system I would need
> > assistance.)
> >
> >
> > cl-info:
> > regex-sanitize, find-regex-matches
> >
> > As far as I can see, find-regex-matches is used by
> > ??, ? and describe. I checked describe:
> >
> > thru 100 do describe(cint);
> > 7.7080 seconds using 627.061 MB.  <- with nregex
> > 6.8160 seconds using 2378.905 MB. <- with pregexp
> >
> > thru 100 do describe(subvarp);
> > 7.6560 seconds using 631.239 MB.  <- with nregex
> > 7.9520 seconds using 2382.587 MB. <- with pregexp
> >
> > thru 100 do describe(foo);
> > 5.4640 seconds using 439.326 MB.  <- with nregex
> > 5.0360 seconds using 1591.035 MB. <- with pregexp
> >
> > -> same timing results but 4 times more space for pregexp
> >
> >
> > commac:
> > strip-float-zeros, $parse_timedate
> >
> > I don't expect $parse_timedate to be any crucial,
> > I just checked strip-float-zeros:
> >
> > The test I performed was essentially printing floats from 1.0 to n
> > redirected into a string_output_stream.
> > This test calls two regex matching functions:
> > trailing-zeros-regex-f-0 (no match) and
> > trailing-zeros-regex-f-1 (match)
> >
> > test(n) :=
> >   block([old_io, redirection],
> >     old_io : ?\*standard\-output\*,
> >     redirection : make_string_output_stream(),
> >     ?\*standard\-output\* : redirection,
> >     for i:1.0 thru n do print(i),
> >     ?\*standard\-output\* : old_io,
> >     close(redirection) )$
> >
> > test(100000)$
> > 5.3080 seconds using 514.245 MB.  <- with nregex
> > test(100000)$
> > 9.4680 seconds using 1810.400 MB. <- with pregexp
> >
> > -> double time and again 4 times more space for pregexp
> >
> >
> > share/contrib/unicodedata/unicodedata.lisp
> > uses nregex but I did not look very closely here.
> > At first sight a replacement seems to be possible.
> > Maybe Leo Butler wants to reorganize his code by himself if we want to
> > proceed in that direction.
> >
> > The question is: Do we want to proceed in that direction?
> >
> > Volker
> >
> >
> >
> > Am 19.10.2015 um 18:54 schrieb Robert Dodier:
> >> On 2015-10-19, Volker van Nek <vol...@gm...> wrote:
> >>
> >>> Thereafter I will see how I can replace calls to nregex in src. Maybe I
> >>> need help at this point to identify and understand all calls to nregex.
> >>
> >> There are only a few -- grepping the source code, I see they are:
> >>
> >>  * src/commac.lisp: strip trailing zeros from floats; parse time/date
> >>  * src/cl-info.lisp: search for help topic
> >>  * share/contrib/unicodedata/unicodedata.lisp: not sure
> >>
> >> If you can take a look at this stuff, that will be terrific. Thanks a
> >> lot.
> >>
> >> best,
> >>
> >> Robert Dodier
> >>
> >>
> >>
> ------------------------------------------------------------------------------
> >> _______________________________________________
> >> Maxima-discuss mailing list
> >> Max...@li...
> >> https://lists.sourceforge.net/lists/listinfo/maxima-discuss
> >>
> >
> >
> >
> >
> ------------------------------------------------------------------------------
> > _______________________________________________
> > Maxima-discuss mailing list
> > Max...@li...
> > https://lists.sourceforge.net/lists/listinfo/maxima-discuss
>