Ralf Hinze writes:
> Dear Alastair,
>
> while playing around with the stack size the following error croped
> up:
>
> Prelude> foldl (\n _ -> n + 1) 0 [1..1000]
> 1000
> Prelude> foldl (\n _ -> n + 1) 0 [1..10000]
>
> ERROR: Control stack overflow
> Prelude> foldl (\n _ -> n + 1) 0 [1..100000]
> Segmentation fault (core dumped)
We ought to be catching the relevant signal (shouldn't we, Mark?)
I'll go have a look at why we're not.
> BTW it would be truely wonderful if Hugs provided a way to increase the
> stack size using a command line option.
The merged Hugs-GHC system will address this kind of problem in two ways:
1) The stack will be stored on the heap.
(We have to do this to provide true concurrency in Concurrent.forkIO)
2) The heap need not be a single contiguous block of memory - so we'll
have much more flexibility in how we respond to heap/stack overflows.
eg We can keep mallocing more heap space till we reach some hard limit,
till we reach a machine limit or until the VM system starts swapping pages
too much.
We also plan to put bytecodes in the heap (so one less wired-in limitation)
and to use a multigenerational collector (so the GC won't spend all its
time moving the code objects around in memory).
Alastair
