jastrachan | 7 Apr 17:40 2004
Picon

Re: [IDEA] auto-type coercion thoughts


On 7 Apr 2004, at 16:12, John Wilson wrote:
> On 7 Apr 2004, at 15:57, jastrachan@... wrote:
>> On 7 Apr 2004, at 10:41, John Wilson wrote:
>>> So we are doing speculative conversions (which could have 
>>> significant computational overhead) to find the right function to 
>>> call. I think I'd prefer to have an isType() which says if the 
>>> conversion can be performed rather than trying and seeing what 
>>> happens when trying to find a function to call.
>>
>> I don't see how that would change anything other than making it more 
>> complex to implement and open the possibility of saying you can 
>> convert to a type but not actually implementing it (or vice versa)?
>>
>
> toString(), for example, can be very costly and can have side effects. 
> I don't thing it's wise to do a conversion when you may not need the 
> result of that conversion.
>
> There are many ways of breaking the contract (returning an object 
> which is not an instanceof the requested type, for example.

Though typically we'd only try convert a type to one kind of object. If 
we did 2 conversions, then we'd have ambiguity and so we'd be throwing 
an exception anyways. i.e. its only in rare exceptional cases that we'd 
ever try convert an object to 2 types - when we can't actually call the 
method - otherwise we'd only be doing a single conversion.

So typically, if we did do this, we'd only call toType() once in 
typical usage. If we called it twice and returned 2 non-null values 
then it'd be an exception & we wouldn't be able to choose a method.

James
-------
http://radio.weblogs.com/0112098/

Gmane