regex_replace bug

[nobody]

boost::tregex e("(.*)");
strResult = boost::regex_replace("a", e, "$1b");

strResult now equals "abb", but should be "ab".

My e-mail: kdn01003@student.mdh.se

[John Maddock]

Logged In: YES 
user_id=14804

This is by design, and is compatible with the way Perl does
things: two matches are found, the first matches the "a",
the second matches the zero length string immediately after
the "a".  Therefore two replacements are performed, and
hense the output.  Use .+ instead of .* if you don't want to
find zero length matches.

John Maddock

[coffman@…]

I have found this behavior still exists even when using posix basic or extended regexes. While this behavior may be correct for perl regexes it doesn't seem to be correct for posix basic or extended regexes.

std::string s("a"); const boost::regex e("(.*)", boost::regex::extended); std::string result = boost::regex_replace(s, e, "$&b");

The result is "abb"

If I use gnu's sed utility (GNU sed version 4.1.2) instead I get "ab"

echo a | sed -r "s/(.*)/\1b/g"

and gnu's sed utility claims to use posix extended regex expressions when called with the -r option.

[John Maddock]

(In [79339]) Fix regex_iterator and regex_replace to do what grep/sed do for multiple repeats when matching in POSIX mode. Fixes #589.

[John Maddock]

This is now fixed in Trunk provided match_posix is passed to regex_replace or regex_iterator.