Hi,
On Tue, Oct 01, 2019 at 06:56:17PM -0700, Greg Steuck wrote:
> On Mon, Sep 23, 2019 at 2:40 PM Matthias Kilian
> wrote:
> > Instead of adding more madness (like a patch to diff in this case),
> > I compared the bundled gmp sources against the upstream used by
> > devel/gmp (which are identical except for the remoced coeumentation
> > in ghc's version) and hat a look at at the patch to gmp contained
> > in the ghc sources (libraries/integer-gmp/gmp/gmpsrc.patch, which
> > only adds -fPIC for some systems).
>
> I was browsing the old ghc patches and we switched from the ports to
> the bundled gmp some 7 years ago
> https://github.com/openbsd/ports/commit/47aa23534d85b9287ed8fe026c98a13d9e2599e4
Back then, the bundled libgmp was patched for real. Which isn't the case
any longer. So those problems really shouldn't appear again.
> I couldn't get my "large Fibonacci numbers" to crash anything. I'm
> curious if anybody has a
> bottled testcase to try or knows how to modify the one below to stress
> things more.
The problem back in 2012 was that for some people (including bulk
builders), ghc crashed regularly -- for bulk builders as often as
every 2nd or 3rd run. For others (including me) it took many dozens
of builds to run into the crash. So: no, there's no obvious test
case for it.
Ciao,
Kili