I was just bitten by an extremely nasty bug, caused in part by my own foolishness, but also by a subtlety of Boost Regex that probably should be documented explicitly.

http://boost.org/libs/regex/doc/format_syntax.html gives a list of "The following Perl like expressions" and says that $N "Expands to the text that matched sub-expression N."

HOWEVER, Boost Regex is sufficiently general that $10 will match the tenth subexpression, $25 the twenty fifth, and so forth. Perl is not that general - it recognizes only $1 through $9 here.

I believed that Boost behaved like Perl. I concatenated "$1" with a random hexadecimal string, occasionally producing "$10", "$15", etc. before the letter hexits appeared. This led to many hours of misery.

The documentation should note Boost Regex's generality. The generality is a good thing - $1 can be separated from later digits with empty () parentheses - but it needs to be made clear.

Will do.

Does this also affect the regex standardization proposal? It'd be bad if some implementers thought they only had to support $1 - $9.

Yep, the proposal uses the ECMA standard by reference - and that allows for $n or $nn to be recognised as refering to sub-expressions, so you access $99 but $100 is really ${10}0. John.