Sidebar re autoconf

[Andrew Tridgell]

> Is one of those the gid_t detection for getgroups? Okay if I remove the
> run-time portion and replace it with what autoconf detects?

yep! that is one of those really ugly things that is crying out for
compile time detection.
 
> Oh, would people prefer a patch which keeps the current system in place or
> one which simply throws the current one out?

throw out the current code! I'd almost managed to forget it was
there. It sure is _ugly_. Note, however, that you can't just check for
sizeof(gid_t). There are systems (Ultrix comes to mind) that are badly
broken and actually get the types totally screwed in
getgroups()/setgroups(). You'll need code like we currently have, but
at compile time as a autoconf test.

While I think of it, one of the tricky ones is the dfree code. The
statfs/statvfs etc calls are very different across different
platforms. We'll need some smart autoconf stuff to fix that. I think I
saw some package that did detect that stuff with autoconf (hmmm, maybe
it was ssh?).

Other ones that come to mind are:

- quotas
- mmap vs. sysV shared memory vs. lock files
- the silly solaris "off by 2 bytes readdir()" bug
- the shmem_ds bug in RH5 Linux
- correct SIGCLD handling

etc etc. It will be great to get some of this stuff cleaned up.

Cheers, Andrew