Purity

[bearophile]

Bruno Medeiros:

> I think that the concession that pure will be allowed to allocate memory 
> does inescapably remove some of the guarantees that pure functions offer 
> (like that one that the return value depends only on the arguments).
> One possible fix to this would be to say that the allocated memory must 
> be temporary (used only during the execution of the pure function). Thus 
> you would not be able to return any newly-allocated value. But I don't 
> know if this further restriction is desirable or not. I don't remember 
> if this aspect of memory allocation in pure functions was 
> discussed/thought-out extensively or not. (it probably needs to)

I have discussed this a bit with Steven Schveighoffer, see the @transparent attribute, in this "Uh... destructors?" thread:
http://www.digitalmars.com/webnews/newsgroups.php?art_group=digitalmars.D&article_id=130554

Bye,
bearophile