Wider tuple design discussion

[Michel Fortin]

On 2011-08-02 13:15:52 +0000, bearophile <bearophileHUGS at lycos.com> said:

> A tuple isn't an object, it's more like a record, it's almost like a 
> more handy POD. So tuples are data structures that you use with code 
> (functions) that they don't contain.

Actually, that's up to debate in my opinion. Andrei's Tuple in phobos 
is a struct, so it's a POD like you say. Walter's tuples in the 
language are just collections of variables.

A language tuple does not have an address in itself, it essentially has 
one address per field. This actually makes tuple packing/unpacking very 
efficient across function calls (because there's no real 
packing/unpacking taking place), but it makes it impossible to take the 
address of a tuple.

We need a way to make those two concepts work together, I think that's 
the hard part.


> Tuples do have some disadvantages. Their fields often are anonymous, 
> this goes against the typical well defined style of D/Java/Ada 
> programs, where you know the name of what you are using. So tuples are 
> better if you use them locally, or if they have only few items. Long 
> tuples become hard to use because you risk forgetting what each field 
> is, so the most common tuples have between 2 and 5 items.

Named tuple elements in a language tuple will be implemented when/if 
someone implement named arguments, something I might do eventually.


-- 
Michel Fortin
michel.fortin at michelf.com
http://michelf.com/