Re: [Bug-apl] rank operator with axis

[Elias Mårtenson]

Makes sense. The standard specification is so weird that it must be considered broken.

Frankly, I'd love to hear the discussions that preceeded its inclusion in the standard. :-)

Regards,

Elias

On 2 July 2014 23:19, Juergen Sauermann <address@hidden> wrote:

Hi,

I have changed the rank operator with axis, i.e. ⍤[X].

The old implementation tried to improve on the somewhat broken
syntax specified in the ISO standard:

      A ⍤[y] B    ←→    A ⍤ y B
      ⍤[y] B      ←→    ⍤ y B

The new implementation follows NARS which uses ⊂[X] instead of ⊂
when conforming the final result.

Note that the "flat" format like A ⍤ y B can give surprising results, so
you should always put the operator arguments in parentheses like (A ⍤ y) B
for clarity.

The current implementation (axis or not) binds y to B and not to ⍤, so
A ⍤ y B means  A ⍤ (y B) and not (A ⍤ y) B. The "correct" binding of APL2
would read a simple thing as A ⍤ 1 2 3 B as (A ⍤ 1)(2 3 B) which might surprise
less-frequent users of ⍤.

It is unknown how APL2 itself would handle this because they haven't implemented ⍤
(or any other primitive operator with a value instead of a function as the right function
argument).

/// Jürgen