[bitc-dev] Location semantics

[Jonathan S. Shapiro]

On Wed, 2006-05-31 at 17:56 +0100, Sam Mason wrote:
> Sam Mason wrote:
> >The restriction may get annoying for parameters though, so I would
> >be tempted to loosen the requirement to the type of return values
> >only.
> 
> I think I was a bit hasty with that comment.  There are situations
> where it could be good if it did this, but I think the safe-by-default
> rule sould apply here.  There are going to be many more situations
> where this behaviour is just plain confusing and a default of no
> inferred locations is much better!

Sam:

There have been further discussions here about location, and we dropped
some of the restrictions.

The problem with your earlier question is this:

Suppose you have a structure of *value* type in a variable 'a'. It has
some field 'b'. The goal is to update the field.

The problem is that this really *isn't* a REF type situation, and we
need to be *very* careful to make sure that REFs do not escape in the
wrong ways.

shap