[OpenAFS] Re: Failure to build OpenAFS 1.6.1a

[Andrew Deason]

On Tue, 09 Oct 2012 10:23:00 -0700
Russ Allbery <rra@stanford.edu> wrote:

> > We may want to try probing /usr/heimdal/bin/krb5-config in the krb5
> > configure logic.
> 
> /usr/bin/krb5-config.heimdal is how Debian spells this.  I've been
> avoiding doing either, though, because usually MIT is available as the
> system library and that krb5-config will be found first, so I don't
> think adding /usr/heimdal/bin to the search path will actually help
> anything.

But what about when heimdal is installed and MIT is not? (Which is what
happened for the OP, originally.) In the case when
/usr/heimdal/bin/krb5-config is the only krb5-config on the system...
how does this not help? I meant possibly trying it last, if we didn't
find anything else. Currently we don't find any krb5-config.

> > We also probably need to do better about detecting the proper header
> > vs detecting libraries, apparently... (I haven't looked at this in
> > detail, maybe this is better in master or something)
> 
> I'm able to build my Kerberos packages with both Heimdal and MIT
> installed on the system on different paths, so I know that the macros
> work and produce the right flags.

I'm not sure if I'm being clear. In the situation I saw broke, only
heimdal was installed, which installed /usr/lib64/libkrb5.*, and
/usr/heimdal/include/krb5.h (et al). Since we did not find a
krb5-config, we tried to probe by finding functions in -lkrb5 (et al)
directly, which succeeded, but we did not add any -I paths to CPPFLAGS
et al.

So, when a file goes to #include<krb5.h>, that fails, since we have no
krb5.h in our include path. (I have config and build logs if I'm not
making sense.)

So, what I was trying to say: it seems like maybe we (or "Russ", or
"rra-c-util", etc) should be probing for krb5.h in addition? Or maybe a
newer version does this; I haven't looked around yet.

-- 
Andrew Deason
adeason@sinenomine.net