[bitc-dev] Re: Closure Conversion

[Jonathan S. Shapiro]

On Fri, 2005-07-08 at 16:34 -0400, Swaroop Sridhar wrote:
> Cases like:
> 
> (define F (lambda (x) ((lambda (y) x) 2)))
> 
> are easy. The inner-lambda's closure can be safely stack allocated.
> 
> But consider:
>    (define G (lambda (f) (let ((p 3)) (f (lambda (x) p)))))
> 
> the inner lambda is apparently not escaping downward. But it is not 
> clear whether its closure can be stack allocated or not.
> 

> i) Escaping will come to mean escaping upwards or downwards, and most 
> lambdas will have their closures heap allocated

I think that the escape analysis indeed is not trivial, but I don't have
a simple answer for you at the moment. I agree that we need to think
about this.

Ask MarkM for Jonathan Rees's email address and ask him about this --
this a richly explored area and we probably don't need to do any
invention.

shap