Re: configure: error: cannot compute sizeof (long double), 77

["Martin J. Bligh" <mbligh at google dot com>]

Mike Frysinger wrote:

> On Thursday 26 January 2006 10:24, Martin J. Bligh wrote:
>  
>
> > Mike Frysinger wrote:
> >    
> >
> > > On Wednesday 25 January 2006 21:12, Martin Bligh wrote:
> > >      
> > >
> > > > I took a *very* brief look at config.guess. Isn't it fundamentally
> > > > flawed for what you want in that it's trying to work out the
> > > > architecture of the running kernel, and what you want to know is the
> > > > architecture of userspace?
> > > >        
> > > it does what it does and that's that
> > >      
> > Umm. Not sure what that's meant to mean.
> >    
>
> it means that what config.guess does currently is pretty much all it's 
> supposed to do and it does it pretty well
>
>  
>
> > It would seem that either what   
> > it's trying to determine isn't what you want to know, or there's a bug
> > in it. Or are you saying that the answer we wanted in this circumstance
> > really was x86_64, and the bug is in configure / crosstool, etc ?
> >    
>
> i'm saying config.guess should have shown i686-... and it prob didnt because 
> of the env it was running in
>
>  
OK, what I'm trying to get at is "what does the answer i686 represent".
Is it the CPU architecture, the userspace, the kernel, what ? They're 
definitely not equivalent. If you're using uname -r, etc you're going to 
get kernel architecture, I believe ... but what exactly is config.guess 
trying to specify here?

> > > this is why you can override it with --build/--host/--target configure
> > > options
> > >      
> > What *exactly* is the thing we're trying to determine here? Presumably
> > the "host" architecture? And does that mean the physical chip, or the
> > userspace you're running in, or what kind of executables the currently
> > running kernel will support?
> >    
>
> if we were concerned with just the host architecture, we wouldnt have tuplets 
> that contain architecture/operating system/file format/etc...
>
> the name determines pretty much all of these things
>  

Right. But the host architecture is what we seem to be struggling with 
here. So which facet of architecture are we (1) wanting to find out and 
(2) is config.guess giving us?

> > "You can override it" seems like a poor option to me - crosstool is
> > intimidating for new users, and it would be nice if it either gave a
> > clear error or just worked out of the box
> >    
>
> i was talking about configure in general, not crosstool
>  

So ... I'm still not clear exactly why it's blowing up here. On a CPU 
and kernel architecture that supports both ia32 binaries and x86_64 
binaries, it should be able to work with either, surely?

Or are we asking for x86_64 host support without some vital component in 
the build chain being correct ... eg do we have to have a 64 bit 
binutils or soemthing that we dont have?

M.

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sourceware.org