On 03/11/2009 13:10, Simon Marlow wrote:
On 03/11/2009 13:04, Simon Peyton-Jones wrote:
| Can I remind people that T1969 is there for a good reason: when it
| wobbles, it may indicate a performance regression. We should keep the
| boundaries updated so that people don't just ignore it as a matter of
| course.
|
| I realise it's difficult to keep it updated because it has to be
done on
| multiple platforms. But at the very least if your patch makes it fail,
| you should update the values for *your* platform (after checking that
| you haven't really introduced a performance regression).
OK... but then can T1969.hs please contain a comment that explains
what to look for, and what to alter to change the bound? It's all a
bit cryptic.
Look in space_leaks/all.T. I think it's pretty self-explanatory, but
shout if that's not the case for you.
Oh, and it looks like this time we really have regressed. Previously
GHC allocated 430MB on this test, and now it allocates 515MB, on
x86-64/Linux. That deserves looking into at the least.
Cheers,
Simon
_______________________________________________
Cvs-ghc mailing list
Cvs-ghc@haskell.org
http://www.haskell.org/mailman/listinfo/cvs-ghc
