[Sml-implementers] Re: [Smlnj-dev-list] Evolving SML

[Robert H. <rw...@cs...>]

i agree with this view.  the individual fields should not be 
independent "l-values" (to borrow old terminology), and in particular 
should never alias.  treating a mutable record as a ref to a record 
(for which there is a functional update operation for fields) seems to 
be the right thing.  so much so, in fact, that i don't see why it 
should be a primitive notion at all!

bob

On Oct 25, 2005, at 10:31 PM, John Reppy wrote:

> Do you mean the semantics or implementation of equality?  The 
> semantics are
> fairly straightforward: just make equality on records with mutable 
> fields be
> pointer equality.  In fact, I would argue that refs should be viewed as
> syntactic sugar for a record with one mutable field.
>
> If you want to view mutable fields as a derived form (at least 
> formally),
> then I'd argue for a ref of a record (not a record of refs).  The fact
> that individual fields cannot be shared with other data structures is
> important.
>
>     - John
>
> On Oct 25, 2005, at 4:18 PM, Peter Sestoft wrote:
>
>> On Tue, 25 Oct 2005, John Reppy wrote:
>>
>>
>>> Here are a few ideas for short-term (and medium-term) features that
>>> I'd like to see:
>>>
>>
>>
>>>     7. records with mutable fields.
>>>
>>
>> Preferably this should be syntactic sugar for a records with ref
>> fields.  OCaml-style updatable records, while useful, would require a
>> redesign of equality because updates can create cycles...
>>
>> Peter
>>
>>
>