[m-dev.] Filesystem race condition?

[Zoltan Somogyi]

On Thu, 17 Sep 2015 16:47:38 +1000, Paul Bone <paul at bone.id.au> wrote:
> Has anyone else seen something like this, during ./tools/bootstrap.sh -j8:

Yes. It happens sometimes for multimodule programs in invalid.
There is a comment on line 336 of tests/Mmake.common that
gives two reasons why this may happen with mmake -jN for
multimodule programs, but I don't see how either of those reasons
apply to a program such as children.m, all of whose submodules
are in children.m itself, and not in separate files.

I think the problem is probably caused by either a dependency on a header file
that is not expressed in the makefile fragments we generate, or an attempt
to read in a header file before its creator has finished writing it.

Unfortunately, tracking down these kinds of heisenbugs is often tough
and almost always annoying :-(

Zoltan.