[mercury-users] Native garbage collector for Mercury

[Randall Helzerman]

> 
> Basically, I just disagree with the claim that "Mercury's termination
> analysis is already very good".   I might agree if it could prove 90%

Dude, all I can say is your standards are pretty high.  To put this in
perspective, just think how hard it would be to write a program in C++ which
could prove the termination of 50% of the routings in a C++ compiler!

Let me try again to build a case for why termination analysis is important
for distributed/agent computing.  Suppose I've got a deductive database on
my machine which I wish to let the world query.  The queries to this
database will be simple logic programs.  But I don't want crackers to run
malicious programs on my computer.  If I simply stipulate that I will only
accept a database query which I can prove will terminate, I know that 
crackers can't suck too much CPU power.  Yes, this does restrict the sorts
of queries which I will accept, but this is a feature, not a bug.

And this line of thought can be generalized: by using abstract
interpretation, I can enforce all sorts of invarients on incomming queries:
that they don't modify my database, or that they modify only the parts which
they have permissions to, etc.  

In this way we can start thinking of _provably_secure_ networked computing.