Re: [Chicken-users] need regex benchmark

[Alex Shinn]

Tobia Conforto <address@hidden> writes:

> Alex Shinn wrote:
>> Tobia Conforto writes:
>>> The file perltestdata in Edi Weitz's cl-ppcre contains a big PCRE
>>> stress test (15,000 tests) already formatted for lisp.
>>
>> What Felix is looking for, however, is a benchmark.
>
> What's wrong with repeating 1000 times each of those test cases, at
> least the ones that are not supposed to fail, and record the CPU time
> used?

Because there are regexps which you want to be fast, regexps
which you want to just be tractable, and regexps for which
there is really no hope.  The next release of irregex will
include all of these cases, carefully catalogued, measuring
compile time, execution time, and memory usage.  I just have
no idea when I'll get around to this.

In the meantime, if you want to take a representative sample
you might as well take a sample of something people are
actually using.

CL-PPCRE doesn't do a DFA construction, so there are regexps
that irregex can execute instantly that CL-PPCRE couldn't
even execute in a million years.  These are obviously _not_
in the CL-PPCRE test suite, so I'm not too confident of it
being usable for any kind of benchmarking purposes.

-- 
Alex