Re: support for multithread-safe libraries

[Bruno Haible]

Paul Eggert wrote:
> I have one minor quibble: the name "lock".
> That's a generic name, and "lock.h" is likely to collide with file
> names already in use.  I just checked, and OpenBSD has a <sys/lock.h>;
> I wouldn't be surprised if someone else (unwisely) grabbed <lock.h>.

FreeBSD is going away from <sys/lock.h>, they renamed theirs to <sys/lockmgr.h>.
Cygwin has a dummy <sys/lock.h> that provides nothing useful. So far I
don't see anyone using a <lock.h> file; POSIX certainly won't add one
either. A few packages, far away from us, have a "lock.h"; they won't
collide with gnulib.

"lock.h" has the big benefit that it describes what it is, and resembles
glibc's <bits/libc-lock.h>.

> I have access to a Solaris 8 sparc host.  Would it be worth my time to
> see what the problem is on that host?  (I don't know what test-lock
> is, though.)

It's not worth it. The problem is that the scheduler in Solaris 7/8
is not "fair". It is fixed in Solaris 9.

The testing package is in http://www.haible.de/bruno/gnu/gnulib-lock.tar.gz .
If someone with access to FreeBSD 5 could test it - would be welcome!

> That was an impressive array of hosts you tested on, by the way.  How
> do you get access to such a wide variety?

Compilefarm.sourceforge.net is an interesting resource.

Bruno