[OpenAFS] AKLOG build error...

Jeffrey Hutzelman jhutz@cmu.edu
Mon, 01 May 2006 14:05:47 -0400

On Saturday, April 29, 2006 11:25:45 AM -0400 Rodney M Dyer 
<rmdyer@uncc.edu> wrote:

> I think it all boils down to...  If you are going to allow the user to
> set the location of their compiler at all, then you should "obey" that
> throughout, otherwise what is the point?  And more to the point, why does
> a configure script try to "overly agressive" find anything?  Let me
> "configure" where everything is with the "configure" script, eh?

So, I think you are confused.  The vast majority of the configure script 
isn't code we wrote; it comes from autoconf.  On certain platforms, for 
reasons I already explained, we ignore the compiler that is selected for us 
by that standard test, and use a hard-coded one instead.

Yes, this is ugly, and we know it doesn't always play nice with the 
configure tests, and causes problems for a few people.  We didn't design 
the architecture of autoconf, and it doesn't really deal well out of the 
box with the extra constraints involved in building kernel code.  Making 
this work "right" is certainly on the todo list, but at a much lower 
priority than things like actually fixing bugs.  In the meantime, given the 
relatively small number of people affected, how easy the problem is to work 
around, and that the most vocal complainers won't be happy until it is 
actually done right, doing throwaway work on the ugly hack is just more 
effort than it's worth.

I've described how to work around the problem, and you've pointed out 
another way that also works.  Which method is most appropriate for any 
given person probably depends on their particular philosophy about how to 
build software.  For now, I suggest that affected parties pick one and get 
on with life.

-- Jeff