Re: [q-lang-users] H/M-based typing for Q

[Andrew B. <an...@vo...>]

Perhaps a word of clarification:

>> My intuition about this is that most of the code that gets written
>> is monomorphic.  (Is that a word?) =20

> (It is.)

Aah, good.  The little bit of Latin I've figured out is not completely =
useless.

> The trouble is not so much with monomorphic variables as with =
monomorphic
> collections.  ML, for example, insists on having monomorphic lists so
> that when you remove an object from a list you know its type.  That is
> very annoying.

My point was not that this is good, but rather that the larger majority =
of functions only ever get one type passed to them, or in the Q =
parlance, the majority of expressions use the same types from evaluation =
to evaluation.  Going back to my initial statement, my (probably =
provably incorrect) belief, based on an incomplete understanding of the =
problem space, is that it is possible to automatically determine the =
types in these cases, and not require the programmer to spell out all =
that extra noise.  In the cases where the programmer does something =
polymorphic, then the compiler's task gets a bit harder, and perhaps it =
will have to emit a different reduction for each input type or =
something.  In the cases where the programmer is doing something =
complicated with delayed execution, it is not hard to imagine that the =
compiler would have to fall all the way back to calling an interpreter =
in the run-time library. =20

-andrew