Re: PATCH: Assorted BP fixes

[Greg McGary <greg at mcgary dot org>]

Ulrich Drepper <drepper@redhat.com> writes:

> Looks OK though some systems are naming core files differently.
> I'm not sure whether we should add this part.

It's harmless because if core doesn't exist, the rename part does
nothing.  On the downside, there's a race condition for parallel
makes, and it's susceptible to renaming a pre-existing stale core file.

If you don't like it as default behavior, can I make it conditional?
I find it very useful to keep cores from overwriting each other.
That's how I discovered that localedata failures were all caused by
the same problem.  I ran a batch gdb on all 78 coredumps I had saved
and grep'ed frame #0.  With this sort of global analysis, I can
prioritize my debugging and go after the bugs that affect the most
tests.

Aside from renaming, there's also the problem that tests run with
their working directory in the source tree.  It's not kosher to write
files in the source tree, but that's where coredumps go.  Is there any
reason why we shoudln't rewrite the test invocation rules to first cd
into $(objpfx)?  All paths in the rule are already absolute.

> If something fails, you'll
> run the tests again and this time without -n for make.

I'm not sure I follow.  Do you mean without `-k', so that it will
stop after the first failure?

Greg

PS: $* stands for whatever % matches, so I can get rid of the
    $(subst -bp,,$*) crap in make-test-out, and go back to plain $*.