Re: [Fwd: More APR fun (was Re: undefined reference to `pthread_mutexattr_init' in apr?)]

[Bruce Korb]

Akim Demaille wrote:
> 
> >>>>> "Bruce" == Bruce Korb <address@hidden> writes:
> 
> Bruce> Hi Alexandre (or someone on autoconf),
> 
> Bruce> Just wondering if this might ring any bells for anyone.  If
> Bruce> someone already knows of the causes and workarounds, advice is
> Bruce> easier than trial-and-error research....
> 
> I'm not sure I really understand the question here.  The first run
> creates the cache which accumulates information during the run of the
> full configure (in particular quite lately configure discovered it
> wanted something like -pthread in CFLAGS).
> 
> Then the second run already knows this stuff, so indeed the second run
> is different: it has the knowledge of the first run.

Why wasn't `-pthread' in CFLAGS after the first run?
What is the right way to accomplish that?

> Finally, the ugly message about pthread is the typical example of
> nested AC_CACHE_CHECK.  Just don't :)

I wouldn't, if I knew how it happened:

> $ egrep AC_CACHE_CHECK configure.in
> $

[[i.e. there was nothing found that matches AC_CACHE_CHECK]]

> Or did I miss the question?

Maybe I didn't ask it very well.  The issue is that the subversion
project needs to build for and link to the pthreads library.  Apparently,
judging by the mail, it will do this if you run configure twice.
I personally did not do that.  Instead, I hand hacked all the relevant
Makefiles.  Either way is wrong.  :-)  It would be good to know how
to fix the Makefile.am or configure.in script so that the right thing
happens from the start (i.e., the ``-pthread'' option is handed to the
compile and the link command uses ``-lpthread''.)  Way over my head.  :-)

So the point: