[Bug c++/50015] New: std::initializer_list<std::string> members are not properly initialized

["danny.dyk@tu-dortmund.de" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50015

           Summary: std::initializer_list<std::string> members are not
                    properly initialized
           Product: gcc
           Version: 4.6.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: danny.dyk@tu-dortmund.de


Created attachment 24940
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24940
Output of 'g++ -std=c++0x -save-temps -o test-case test-case.cc'

When creating a std::initializer_list<std::string> as part of a struct, the
actual
object is not properly initialized. Consider the following test case:

-- BEGIN --
#include <initializer_list>
#include <iostream>
#include <string>

struct T
{
    std::initializer_list<std::string> s;
};

T foo
{
    {"foo", "bar", "baz"}
};

int main(int, char**)
{
    for (auto t = foo.s.begin(), t_end = foo.s.end() ; t != t_end ; ++t)
        std::cout << *t << std::endl;

    return 0;
}
-- END --

After compiling this test case using

  g++ -Wall -pedantic -std=c++0x -o test-case test-case.cc
  [no errors, no warnings]

the execution of the resulting binary via

  ./test-case

either raises a segmentation fault or produces garbled output, e.g.,

  foo1`particularÿÿÿÿbar1ÿÿ

Looks very much like terminating '\0's are not in place.

This bug was first seen when using gcc-4.4.5 and is reproducable under
gcc-4.5.0
up to gcc-4.6.1. The particular version of gcc-4.6.1 was configured as

 configure \
    --prefix=/het/packages/gcc-${VERSION} \
    --program-prefix=het- \
    --program-suffix=-${VERSION} \
    --enable-languages=c,c++ \
    --disable-multilib

and built on an amd64 system.

I will attach the output of

  g++ -std=c++0x -save-temps -o test-case test-case.cc
  g++ -std=c++0x -fdump-tree-gimple -o test-case test-case.cc

to this bug report.

Thank you to Alexander Faeroey for help in reducing the segfaulting code to
a manageable test case.