bug#14295: Support copy-file ACLs for Solaris etc.

[Paul Eggert]

> How hard would it to support both?

I don't know, and I'd rather not support --without-acl.  This should
have been an --enable/--disable option in the first place, as it's
enabling or disabling a feature, not building with or without a
package.  There's little point to Emacs departing from existing
GNU practice here.

> This uses EOPNOTSUPP without #ifdef guards; is that universally
> available?

Mostly, except for Windows, but to play it safe we can incorporate the
Gnulib errno module, which defines EOPNOTSUPP if it's not already
defined.

> If I'm right, perhaps it is best to leave the WINDOWSNT parts alone:

Sure, we can do that.

> It sounds wrong to me to condition a call to acl_get_file with a macro
> called HAVE_ACL_SET_FILE.

In practice both are available if either is; in particular, there's no
point calling acl_get_file if you can't invoke acl_set_file later.

> For this to work, acl-errno-valid.c will have to be compiled on
> MS-Windows.  And for that, we will need to solve the EOPNOTSUPP issue
> mentioned above, because Windows doesn't define it.

OK.  Revised patch attached, which tries to address the above.
It relies on Gnulib for EOPNOTSUPP, except for WINDOWSNT where
'configure' isn't working yet, and there it adds a line to
nt/config.nt to deal with this until we get 'configure' working.

acl.txt
Description: Text document