Boost logo

Boost Users :

Subject: [Boost-users] [regex] segfault from construction
From: Andrew Troschinetz (ast_at_[hidden])
Date: 2008-11-25 09:28:52

Hello all,

I've got what I suspect is a relatively unique problem that's
resulting in a segfault whenever I try to construct a boost::regex.

First some machine information:
        Red Hat Enterprise Linux Server release 5.1 (Tikanga)
        Linux 2.6.18-53.el5, SMP, x86_64
        gcc 4.1.2 20070626 (Red Hat 4.1.2-14)
        boost 1.33.1-10

Here's a small example that shows the problem:
        #include <boost/algorithm/string/regex.hpp>
        #include <iostream>
        #include <string>

        using namespace std;

        int main()
                string input ("test test1 test2 test3");
                string pattern ("\\btest\\b");

                erase_regex (input, boost::regex (pattern));

                cout << input << endl; // expecting " test1 test2 test3"
                return 0;

Compile flags (and this is getting to the heart of the problem, I
        -g -m32 -malign-double -O3 -Wall -lboost_regex

Put all the above together in a bag, shake it up, and you get this
        0x08066beb in boost::re_detail::basic_regex_parser<char,
boost::regex_traits<char, boost::cpp_regex_traits<char> >
>::parse_extended (this=0xffde311c)
        at /usr/include/boost/regex/v4/basic_regex_parser.hpp:235
235 switch*this->m_traits.syntax_type(*m_position))
(gdb) print m_position
$1 = 0x0
(gdb) bt
#0 0x08066beb in boost::re_detail::basic_regex_parser<char,
boost::regex_traits<char, boost::cpp_regex_traits<char> >
        at /usr/include/boost/regex/v4/basic_regex_parser.hpp:235
#1 0xf7f9e364 in boost ::re_detail::basic_regex_parser<char,
boost::regex_traits<char, boost::cpp_regex_traits<char> > >::parse ()
        from /usr/lib/
#2 0xf7fa0eba in boost::basic_regex<char, boost::regex_traits<char,
boost::cpp_regex_traits<char> > >::do_assign () from /usr/lib/
#3 0x08050789 in main () at /usr/include/boost/regex/v4/

So the immediate culprit is a dereferencing-null problem.

Interestingly, the problem goes away at -O, -O1, and -O2. Or -O3 but
without -malign-double.

I was wondering if this is a known problem, and if anyone else who's
hand is forced to compile with -malign-double has seen this problem
and been able to work around it?

Andrew Troschinetz
Applied Research Laboratories

Boost-users list run by williamkempf at, kalb at, bjorn.karlsson at, gregod at, wekempf at