Proposed new syntax

[Ian Kelly]

On Thu, Aug 10, 2017 at 11:45 PM, Steve D'Aprano
<steve+python at pearwood.info> wrote:
> On Fri, 11 Aug 2017 08:49 am, Ben Finney wrote:
>
>> The comprehension encourages thinking in sets: an operation that takes a
>> collection as input, and emits a different collection, through one
>> conceptual operation.
>>
>> Adding ‘while’ in there encourages thinking not in terms of a single
>> set-based operation, but an iteration of separate operations. That
>> confuses the model, and I no longer have a coherent model about which to
>> reason what the syntax might mean.
>
>
> Sorry Ben, you've completely lost me.
>
> If you had said that tradition functional style operations such as:
>
> map(func, iterable)
>
> filter(pred, iterable)
>
> "encourages thinking in sets: an operation that takes a collection as input, and
> emits a different collection, through one conceptual operation"
>
> then I would completely agree with you. I agree that is absolutely true:
> traditional functional programming idioms encourage thinking of looping as a
> single conceptual operation. For simplicity, both map() and filter() are often
> implemented as a for loop that operates on one item at a time, in order, but
> conceptually the map and filter could operate in parallel on all items at once.
>
> But that's not the case for list comprehensions and generator expressions (which
> use almost exactly the same syntax). The sequential, one-item-at-a-time nature
> isn't a mere implementation detail, it is an essential part of the semantics of
> the comprehension.
>
> Comprehension syntax makes the sequential loop explicit: the loop is right there
> in the syntax:
>
> [expr for x in iterable]

This is a peculiarity of Python. Here's a list comprehension in
Haskell, which has supported them since version 1.0 in 1990, much
longer than Python:

[x * 2 | x <- L, x * x > 3]

The same thing in Erlang:

[2*X || X <- L, X*X > 3]

C#:

var ns = from x in L
         where x*x > 3
         select x*2;

(Note "from", not "for")

Scheme:

(list-ec (: x 100) (if (> (* x x) 3)) (* x 2))

The grand-daddy of them all, NPL (which actually called them "set
comprehensions" after mathematics):

setofeven(X) <= <:x: x in X & even(x) :>


Nothing about any of these suggests a loop or a requirement for serial
processing. Granted, there are other languages I've not listed here
that use loop syntax like Python.


>> Though ‘for’ is used elsewhere in Python to mean iteration, ‘for’ also
>> has strong connotation in mathematics for set-based operations (“the
>> result is foo for all bar, if baz”). So the same confusion doesn't
>> occur: this is a comprehension which is about set-based thinking, which
>> is supported by all the semantic connotations of the syntax.
>
> I don't understand this. I *think* what you're saying is "I have internalised
> the explicit for-loop and think of it as a filter plus a map and no longer
> think of it as a for-loop", but I'm not sure.

More like, "Python has externalized the list comprehension as a for
loop despite the rich tradition to the contrary."