https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65470
Bug ID: 65470
Summary: regex_search corrupts matches when haystack is
destroyed
Product: gcc
Version: 4.9.2
Status: UNCONFIRMED
Severity: major
Priority: P3
Component: libstdc++
Assignee: unassigned at gcc dot gnu.org
Reporter: aral at gmx dot de
Created attachment 35063
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35063&action=edit
minimal bug example - compile with g++ -std=c++11 regexbug.cpp -o regexbug
Tested on g++ (Debian 4.9.2-10) 4.9.2.
regex_search with matches apparently depends on the haystack (in both the const
basic_string and const *charT versions) to remain intact. The matches object
returned seems to point to locations in the original haystack. When the
haystack is destroyed, the matches are corrupted.
This leads to very unpleasant results when using regex_search with the
temporary strings returned by .string() or .c_str() methods of many objects
(e.g. boost::filesystem::path.filename() ), as those strings are destroyed at
the end of the line containing the regex_search.
Fix recommendation:
1) if this behavior is intended for performance, add a REALLY BIG FLASHING RED
WARNING MESSAGE to the function documentation on cplusplus.com (and anywhere)
that the haystack MUST NOT be a temporary string, and MUST be kept until after
the matches have been evaluated.
2) to fix, modify the (sub)matches class so that it creates a copy of each
match and manages that copy destruction itself
To reproduce (also submitted as attachment):
-----
#include <regex>
#include <iostream>
using namespace std;
int main( void )
{
cmatch match; // store the matches here,
this object seems to depend on haystack after search
char *haystack = strdup ("BUG DEMO");
regex_search( haystack, match, regex(".*") ); // perform regex
search in the haystack, always matches, not checking match.size() for brevity
cout << "correct match: "<< match[0] << endl; // document the
correct match
delete haystack; // destroy the haystack
cout << "corrupt match: "<< match[0] << endl; // document the
correct match
return 0;
}
-----
(compile with g++ -std=c++11 regexbug.cpp -o regexbug)
