This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: problems with the recent snapshot (971207) (Solved)
On Mon, Dec 08, 1997 at 05:05:56PM -0800, H.J. Lu wrote:
> > /tmp/cca294841.o:/home/phenning/EGCS/Build/build/libraries/libstdc++/testsuite/../../../../egcs-971207/libstdc++/testsuite/../std/bastring.h:148: undefined reference to `ostream & operator<<<char, string_char_traits<char>, __default_alloc_template<false, 0> >(ostream &, basic_string<char, string_char_traits<char>, __default_alloc_template<false, 0> > const &)'
> > [plus a lot of other similar errors]
> >
> > so, tstring.cc compilation and tstring.cc -O compilation fail unexpectedly.
> >
> > What else can I provide to help resolve this?
>
> Same here on i586-linux-gnu. But it was ok on i586-linux-glibc1.
> Since egcs 971207 miscompiles things on x86, it is not that strange.
>
The patch I submitted for bastring is the source of these problems. If
you look at the error messages carefully you'll notice
undefined reference to `ostream & operator<<<char, string_char_traits<char>,
__default_alloc_template<false, 0>
^^^^^
>(ostream &, basic_string<char, string_char_traits<char>,
__default_alloc_template<false, 0> > const &)'
^^^^^
In the library you find
basic_string<char, string_char_traits<char>,
__default_alloc_template<true, 0> >
^^^^^
The issue is the -D_PTHREAD during the compilation of sinst.cc.
This causes the thread safe allocator to be used by default under Linux
glibc2. However, _PTHREAD is NOT defined when tstring.cc is compiled and
it attempts to use the NON-thread safe allocator which was not instantiated
in libstdc++.
I mentioned this issue with the patch but did not have a solution.
What version(s) should go into the library ? SGI uses _PTHREAD in the stl
and egcs defines it for linux in libstdc++/config/linux.mt. How should it be
defined for general programs ?
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
jodyg@idt.net | By definition all answers are replies. Unfortunately,
| not all replies are answers. -- JMS