s/\n(\n?)(\n*)/$1$1$2/g

Hmmm... so, I just had to work out why that was so much quicker than mine at doing pretty much the same thing...

And then I noticed that rather matching this way around:

  s/(\n)*\n/$1/g;
[download]

I should really check for the "one followed by none or more" rather than "none or more followed by one":

  s/\n(\n)*/$1/g;
[download]

And, lo and behold, this runs over twice as fast as the one before, and much closer to the speed if I0's version (which also meets the original spec much better than mine! :))

Now I'm *really* determined to learn more about how to optimise regexs..

Tony