Re: a .NET alternative (GJC et al)

[Jeff Sturm <jsturm at one-point dot com>]

On Sun, 29 Jul 2001, Tom Lord wrote:
> 
> 	At least, we're interested if it can improve performance.  
> 
> Please don't gloss over robustness.  Accurate GC is important for
> that, too.

By "robustness" I assume you mean a reasonable upper bound on heap size.

That's a bit of a difficult problem in itself; accurate GC doesn't
guarantee it, nor does conservative GC preclude it.

Type-accurate scanning has been discussed here recently:

http://gcc.gnu.org/ml/gcc/2001-04/msg00626.html

The paper mentioned here demonstrates a worst case of 34% excess bytes for
type-accurate vs. conservative scanning.  The typical is much lower.

With the empirical evidence available I cannot dismiss the importance of
accurate GC, but I do not consider it a high priority either.

>       People have added support for this to gcc in the distant past.
> 
> That's one approach.  Implementing it entirely outside of the compiler
> is another.  Some of the reasons I like the idea of doing it outside
> the compiler are that will be more widely useful, simpler, and less of
> a maintenance burdon on the future.

As mentioned on previous threads, 100% type-accurate scanning seems
impossible with compiler support only given languages that support free
unions (C/C++).

Jeff