Unix-only code (was Re: svn commit: r10060 - trunk/subversion/libsvn_fs)

On Thu, 2004-06-24 at 07:30, John Peacock wrote:
> Doesn't make it the right method, though. At some point, I plan on spending
> some significant time getting Subversion working under VMS (I know, can't be
> helped) and the more of these "If it's not WIN32 it _must_ be Unix" bodges I
> have to dodge, the more difficult it is going to be.

Sorry. As of r10065 the new one will be in the same place as most of
the others, at least.

> Why not add a stanza to configure to probe for the feature? This is always
> the best method...

That seems a little silly; as far as I know, configure only runs on
systems which have this functionality.

I also don't know what a configure test for the sgid bit would look
like. (Remember that the host system isn't necessarily the same as the
target system.)

brane wrote:
> Why not check for APR_HAVE_UNISTD_H instead?

Because the functionality has little to do with unistd.h.

What we could do is wrap each #include in a HAVE_FOO_H and then test for
S_ISGID. I don't think that would really improve our portability
situation much, since it would still leave open the hypothetical case
where the system has S_ISGID but always fails when you try to set it on
a directory. (Solaris has traditionally broken all sorts of autoconf
tests by providing functions and having them always return "not
implemented" error codes.)

> Should be fairly safe then. Besides, those includes should be done
> through apr_want.h.

Except those headers aren't in apr_want.h, and we can't add code to APR
and use it in Subversion immediately. (And if we were going to add code
to APR, wouldn't we just add code to the file-attribute interface to
allow setting the sgid bit?)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Jun 24 19:01:09 2004