From: Simon Strandgaard <neoneye@...>
Date: 2004-04-27T22:06:10+09:00
Subject: Re: [ann] regexp engine 0.9

ts <decoux@moulon.inra.fr> wrote:
> >>>>> "S" == Simon Strandgaard <neoneye@adslhome.dk> writes:
> S> Gained 0.66% more, so it now passes 96.66%. Version 0.8 only 
> S> passed 96.025 % of the rubicon tests (1560).
> 
>  Be carefull with this rubicon test.
> 
>  I can say you how it was build : I've retrieved tests from P languages,
>  RX, ... adapted for ruby and run it.
> 
>  Each time that I've found a difference, I've tried to see if I can call it
>  a bug or if it was just a different choice made by ruby.
> 
>  Because I was never able to say that it was a bug, I've adapted the test
>  to make ruby pass all its tests even if sometimes I was surprised by the
>  result :-))

Agree there is a bunch of oddities in the rubicon testdata.
At the moment I have ~50 tests which is kind of odd.

I don't intend to be 100% bug-per-bug compatible. Just want to be compatible
with the good things in GNU.

There is ~20 tests where GNU has some empty sub-captures, where I have
chosen that these sub-captures actually should contain some text.
At least these 20 will never pass ok, so the theoretical limit is 
(1560 - 20) / 1560 => 98.7%

Besides Rubicon, I have my own blackbox tests.. 341 to be precise,
these are somewhat more verbose than the rubicon tests.
Furhter more I have some whitebox tests too.

I use the original rubicon 'regexp.test' file, unmodified. In order to use
it I must do heavy substitution, because I cannot overload $&, $1, $2...

def make_result(registers, repl)
  s = repl.clone
  reg = registers
  reg.map!{|i| i || ""}
  reg.fill("", reg.size..12)
  s.gsub!(/([^\\])\\\#\{\$&\}/, '\1REG0')  # hack to find slash prefixes
  s.gsub!(/([^\\])\\\#\{\$1\}/, '\1REG1')  # hack to find slash prefixes
  s.gsub!(/\#\{\$&\}/, reg[0])
  1.upto(10){|i| s.gsub!(/#\{\$#{i}\}/, reg[i]) }
  s.gsub!(/\\\\/, '\\')
  s.gsub!(/REG0/, '#{$&}')  # hack to replace slash prefixes
  s.gsub!(/REG1/, '#{$1}')  # hack to replace slash prefixes
  s
end  

Regexp as we all love it.. It works :-)


BTW: thanks for the 1560 tests ;-)

--
Simon Strandgaard