On mån, 2010-11-15 at 23:34 -0500, Tom Lane wrote:
> It's clear to me that we are very far from having a handle on what
> it'll really take to run parallel builds safely, and I am therefore
> now of the opinion that we ought to revert the patch. Hypothetical
> gains in parallelism are useless if we can't actually use parallel
> building reliably. We are currently worse off than before in terms of
> time to build the system.
We don't have to revert it, we just have to insert .NOTPARALLEL targets
into some places that are not properly "parallelized", thus effectively
restoring the behavior of the old for loop. I have attached a patch
that gets make -j 100+ working for me. Other platforms might need more
things, perhaps.
Btw., my original notes for this development were labeled "make make -k
work properly". So I would really like to keep that. It just turned
out that parallel make could benefit from the same changes, and it's a
better marketing name. ;-)