Re: Problem building from 13027:b9a89ca0fb75

[Ben Abbott]

On Aug 30, 2011, at 11:30 AM, John W. Eaton wrote:

> On 29-Aug-2011, Ben Abbott wrote:
> 
> | I problem appears to have been introduced by rev 13024.
> 
> Please remember that the numeric IDs your hg archive may be different
> from mine.  You should use thee hash ID unless we are talking about
> the same archive.  Do you mean revision f3c53fea9fb5?  If so, then
> that changed things so that -no-undefined is passed to libtool by
> default.  But you can disable that by running configure with
> --disable-no-undefined.  Can you try that and see if linking succeeds?
> If so, what is different about the link command?  I'm guessing it will
> be the -Wl,-single_module option, but I'm not sure as I can't seem to
> find any clear documentation about that option on the web.
> 
> If we don't require all the symbols required are available at link
> time for liboctave, does it still work for you when you call a
> function that uses metis?  If so, then we could modify the configure
> script so that -no-undefined is not passed to libtool on OS X
> systems.
> 
> I made -no-undefined the default with the hope that it would make it
> more likely to catch errors on all systems.  But maybe it's not the
> best choice.
> 
> jwe

My IDs are in sync with the ones on savannah. Thus, rev 13024 does refer to 
changeset f3c53fea9fb5.

After looking through this changeset and experimenting with builds, I've come 
to the conclusion that my builds, since moving to MacPorts to support 
dependencies,have not been linking properly to metis libs (I didn't link to 
metis previously).

Thus, I think the build failure is appropriate, and is consistent with your 
goal of making it more likely to catch errors on all systems.

When I added --with-cholmod="-lcholmod -lmetis" to configure, I encountered no 
problem. I'm building again with --disable-no-undefined and no mention of 
-lmetis just to be sure that works as intended, but will switch back to  
--with-cholmod="-lcholmod -lmetis" right after that. I'll report back if I 
encounter an unexpected result using --disable-no-undefined.

Ben