Re: Make 3.81 and 3.82 break on parallel build

[Paul Smith]

On Wed, 2011-11-16 at 17:35 -0600, David Hagood wrote:
> On Wed, 2011-11-16 at 17:02 -0500, Paul Smith wrote:
> > On Wed, 2011-11-16 at 15:40 -0600, address@hidden wrote:
> > > > Try running the parallel version with -d (redirect the output because
> > > > it's voluminous) and see what make says about trying to build
> > > > somedir/libfoo.so: what does it say about the somedir/libbar.so
> > > > prerequisite?
> > > That it does not exist, so it fails. But if run not-parallel, it
> > > correctly identifies it must be built first.
> > 
> > ?? You mean _make_ says the file doesn't exist and so make fails?  Or
> > the linker says it doesn't exist and the linker fails?
> The linker is run, and fails. However, the -d -p output says that Make
> is fully aware the prereq does not exist, but for some reason decides
> that doesn't matter. The output ALSO clearly shows that make knows the
> prereq must BE made. It's like the scheduling of what rules to run next
> gets out sync.

Hm.  Strange.

The only time I've seen anything like that happen is due to unexpected
interactions between make's "rebuild included makefiles" operation
failing and normal builds.

You can try getting the latest version of GNU make out of CVS and
building it (you'll need autotools and other stuff installed: see the
README.cvs file once you get a copy) to see if that works any better.

Other than that, I don't think I can help much more without a
reproducible test case.  In particular I wonder how you've configured
your auto-generated prerequisites to work.