> It looks to me as if a second round of evaluation is being done.
> During the first round the "\"s would be removed leaving "[[0-9]+]".
> The second round would pair the first "[" with the first "]", the
> one before the "+" and would make your character class be digits
> or opening square bracket. The final square bracket would be literal.
> You may have to play with the number of backslashes, probably doubling
> them though in some troff documents it was even necessary to tripling
> them in order to have one preserved when it was needed.
Thank you, your explanation is plausible... I indeed already tried
with some escaping before posting. I'll try harder.
But in any case, doesn't it sound like a bug? Because a correct
regular expression is not matching as it should.