[mercury-users] Native garbage collector for Mercury

[Thomas Charles CONWAY]

Randall Helzerman, you write:
> Quick question:  Will the mercury's planned native garbage collector
> support hard real time operation?

In a word: no.
BUT
We have talked about using variations on the current termination analysis
to find upper bounds on time or amount of memory allocated, and so on.
Here is an interesting thought:
If you can determine an upper bound U on the size of the terms produced
by a computation, then assuming infinite memory during the computation,
a GC can be performed after the computation in time proportional to U:
using the same technique as solutions/2 uses with heap-reclaimation on
failure.

Of interest also, is a flavour of granularity analysis by Andy King et al
where they work out if a computation is bigger than some fixed grain size
by abstractly interpreting it, counting the cost of the computation until
they reach the grain size, or the abstract interpretation terminates.
Applied to realtime constraints, the situation is more or less reversed:
rather than looking for parts of a computation bigger than the grain size,
you're using the technique to prove that a computation is smaller than
some realtime constraint.

Thomas
-- 
Thomas Conway <conway at cs.mu.oz.au>
Nail here [] for new monitor.  )O+