Re: errno variable on the Hurd (was: Re: Postgresql question)

[nisse@lysator.liu.se (Niels Möller)]

Brent Fulgham <brent.fulgham@xpsystems.com> writes:

> > /* errno is a per-thread variable.  */
> > #include <hurd/threadvar.h>
> > #define errno   (*__hurd_errno_location ())
> > #define __set_errno(val) errno = (val)

> Ahh -- this makes  a lot of sense now.  So we should probably file
> bugs upstream against software that makes use of errno without including
> the errno header file in a particular module/source file.  I think
> there's a lot of stuff out there that relies on the belief that it's
> an `extern int`, which is (as we now know) not always the case.

The interesting thing about this errno hack (which is I think is used
on all platforms that support both threads and ANSI-C, not just the
Hurd), is that errno still *is* an extern int, in some sense.

I.e., in

  #include <errno.h>
  extern int errno;

the second line expands to

  extern int (*__hurd_errno_location ());

which is a perfectly valid declaration, although unfortunately not a
prototype, of the function __hurd_errno_location().

You can also take the address of errno and all things should work just
fine (until you try to use the resulting pointer from another thread).

/Niels