bug? with boost::find_format_all, boost::regex_finder and custom regex formatter. I'm not sure.

[nic.bertrand@…]

I have a code that has been working for almost 4 years (since boost 1.33) and today I went from boost 1.36 to boost 1.42 and now I have a problem.

I'm calling a custom formatter on a string to format parts of the string that match a REGEX.

For instance, a string like: "abc;def:" will be changed to "abc\2Cdef\3B" if the REGEX contains "([;:])"

boost::find_format_all( mystring, boost::regex_finder( REGEX ), custom_formatter() );

The custom formatter looks like this:

struct custom_formatter() {

template< typename T > std::string operator()( const T & s ) const {

std::string matchStr = s.match_results().str(1);

perform substitutions

return matchStr;

}

}

This worked fine but with boost 1.42 I now have "non initialized" s.match_results() which yield to boost::exception_detail::clone_implINS0_::error_info_injectorISt11logic_errorEEEE - Attempt to access an uninitialzed boost::match_results<> class.

This means that sometimes I am in the functor to format a string but there is no match. It seems I have a new empty match at the end of each string to parse.

Am I doing something wrong? Or is it normal to enter the functor when there is no match and I should check against something?

I can avoid this testing s.begin() == s.end() at the begining of the functor, but is it really necessary or is it a bug

[Steven Watanabe]

This sounds like a bug to me.

[Éric Malenfant <eric.malenfant@…>]

This looks very much like Ticket #2777 :"find_format_all calls formatter with invalid match".

An interesting difference is that the present ticket reports that the problem was absent in Boost 1.36, while ticket #2777 is filed against Boost 1.35.

[nic.bertrand@…]

I've test quickly with boost 1.39 and the bug is not present

[Steven Watanabe]

Duplicates #2777.