[MLton] Re: [Haskell-cafe] fastest Fibonacci numbers in the West
Jesper Louis Andersen
Fri, 28 Jan 2005 02:49:43 +0100
> I guess that if we could estimate the amount of stack needed then we could
> see if it would fit.
I think this is wrong. Resource limits of any kind, data/memory/stack,
is there to make the system administrator happy. Programs should not
mess with them, but users should. I think a proper error report is
good, but I think mangling the resource limits is to push it too far.
My conclusion would be something like: If the stack is changed, then
why not move the rlimit for data also, when the GC needs more memory
and a garbage reclaim does not work.
While I would be happy to never see an SML program segfaulting, it is
a bit too academic assuming an inifinite amount of resources on the
host we are executing on, I think. At least for a particular
implementation on UNIX ;)