Simon Marlow wrote:
The point is, GHC has no such thing as the "overall program memory limit" unless by that you mean the total amount of memory + swap in your machine. You can set a limit with +RTS -M, but there isn't one by default. So what happens when you write a program with a space leak is that it gobbles up all the memory, and turns on the hard disk light for a few minutes until the OS gives up and kills it. Or at least, you hope it kills your program and not something else important you were doing at the time.
Well if having "unbounded" (by default) memory use in the form of heap is OK with most users then I can't see why the same should not be OK for stack. An errant program is just as likely to explode the heap as it is the stack AFAICS.
We used to have a default heap limit, for the reason above, but I was persuaded to remove it because it was too much of a pain to keep having to say +RTS -M<whatever> when your program ran out of memory. So we could do the same with the stack limit - the only concern is whether it will be highly inconvenient when we accidentally write programs with infinite loops, which in my experience seem to occur more than space-leaky programs.
To be honest, in all my years of Haskelling I can't think of a single occasion where I've had a program get stuck in an infinite loop. I've had plenty of stack overflows, and they're reported on the mailing lists pretty regularly, but on all such occasions it's been caused by deep but very definitely finite recursion.
We could also set the limit a lot higher than it currently is. Or, we could try to figure out how much memory is in the machine and set it based on that. Basically, I don't care that much, unless it means I have to do a lot of work to implement it :)
I don't think just keeping the implementation as it is and just changing (or removing) the limit is really an option. Unfortunately, as things are at present, using a lot of stack with a program compiled by ghc really is a "bug" and the limit does provide users with some protection against this. But IMO the bug is in the ghc rts, not the users source code most of the time :-( I think at the minimum, the stack shrinking mod you suggested should be implemented before the limit is removed. Regards -- Adrian Hey