Re: Threads, glibc, and all that

[hjl at lucon dot org (H.J. Lu)]

> 
> hjl@lucon.org (H.J. Lu) writes:
> 
> > That glibc bug only shows up when -ansi is used on that test case.
> > Ulrich has fixed it in glibc 2.1 and upcoming glibc 2.0.7. I don't
> > think -ansi is used a lot on C++ code. The worst case is libstdc++
> > includes a proper signal.h and install it in ...../include/g++ if
> > necessary. Ulrich, do you want me to make a patch to libstdc++?
> 
> Installing a separate copy of the header in a g++ directory is no
> good.  This would only mean the with updates of the C library still
> the old header is used in C++ programs.
> 
> Instead I'd suggest that
> a) either disable -ansi (or undefine the test sysmbols) for now
> or
> b) add a comment in the README (with a patch for the libc header) and
>    refer to glibc 2.0.7 (2.1).
> 

Ulrich, how about putting this in ...../include/g++? We can check
to see if it is necessary before we install it. I don't think it
will hurt anything. We can remove it during next installation when
we determine it is no longer needed. A patch to libstdc++ shoudn't
be hard.


H.J.
----
#include_next <signal.h>

# if !defined __sigset_t_defined && defined __need_sigset_t
typedef __sigset_t sigset_t;
#  define __sigset_t_defined	1
# endif /* `sigset_t' not defined and <signal.h> or need `sigset_t'.  */
# undef __need_sigset_t