In article <5598ff89-9286-455f-a561-f21df5003c18@f10g2000hsf.googlegroups.com>,
Viatly <postoronnimv77@mail.ru> wrote:
>Mmm, Marc, could you please be a bit more detailed?

Sorry, I was a bit out of time.

Some gcc libraries, specifically the libstdc++, do rely on some hooks in the
libc that do not necessarily exist. Specifically, the libstdc++ will compile
`thread-safe' on a gnu-linux system.

If you compile it on OpenBSD, you will get something that does not work.
Assuming you fix the few issues you found, the configure mechanism still
will not find `proper' `thread-safe' versions of some libc underlying
functions (IO functions), and will give you non-thread safe iostreams.

You can fix that by either implementing the gnu hooks in our libc, or
porting libstdc++ to our environment.

Otherwise, you will end up with a subtly broken C++ system---one that is
not thread-safe at all.

This is the sense of `horribly'. Things will work most of the time, but
without locking mechanisms, things that should be protected from other
threads won't be.

Admittedly, since we only have a functional userland-only thread library
for now, this may not be such a big issue, since pthread does already
abuse signals to such a point that a lot of things don't work in some
fringe cases...