Re: Disclose fails with mixed ranks

[Dr . Jürgen Sauermann]

Hi,

if we look at the rank operator in ISO APL (which is not present in IBM 
APL 2) then they
define an operation "conform" which reads:

*conform: *The shape of a result is the frame of the argument (relative 
to the cells to
which the function applies) catenated with the shape of the individual 
results produced
by applying the function to the individual cells. If the results do not 
agree in shape
they are brought to a common shape as follows:
If the ranks differ, they are brought to a common maximum rank by 
reshaping each
individual result to introduce leading unit lengths.
If the individual shapes differ (after being brought to a common rank), 
each is brought
to a common shape by using take on each individual result, using as the 
argument
to take the shape which is the maximum over the shapes.

Even though ISO defined this operation only for conforming the argument 
of the rank operator,
it seems reasonable to me to use it also for disclose as opposed to 
using a different operation
to conform the argument of disclose. In particular since the 
"deficiency" of the arguments (i.e.
differing ranks of the items) is exactly the same in both cases.

Best Regards,
Jürgen


On 9/24/21 7:30 PM, Jay Foad wrote:
> Apart from compatibility with Rank (which didn't exist in Dyalog 
> before 14.0), why do you think adding leading ones to the shape is 
> "obvious"? Why not trailing ones, for example?
>
> Jay.
>
> On Fri, 24 Sep 2021, 15:48 Elias Mårtenson, <lokedhs@gmail.com 
> <mailto:lokedhs@gmail.com>> wrote:
>
>     Fair enough. That said, I think dzaima/APL implements it. As does
>     KAP now.
>
>     It's also a pretty obvious implementation, and the behaviour
>     orthogonal, instead of having a special case for scalars.
>
>     Regards,
>     Elias
>
>     On Fri, 24 Sept 2021 at 22:35, Jay Foad <jay.foad@gmail.com
>     <mailto:jay.foad@gmail.com>> wrote:
>
>         Elias,
>
>         This was an extension to Mix in Dyalog 14.0
>         (http://docs.dyalog.com/14.0/Dyalog%20APL%20Release%20Notes.pdf
>         <http://docs.dyalog.com/14.0/Dyalog%20APL%20Release%20Notes.pdf>
>         p22)
>         inspired by the way the Rank operator assembles results with
>         differing
>         ranks. I'm not aware of any other APL2-like APLs implementing this
>         extension.
>
>         Jay.
>
>         On Wed, 22 Sept 2021 at 14:20, Elias Mårtenson
>         <lokedhs@gmail.com <mailto:lokedhs@gmail.com>> wrote:
>         >
>         > The following expression should work, if I'm reading the ISO
>         spec correctly:
>         >
>         > ⊃ 3 (1 2) (2 2 ⍴ 0 1 2 3)
>         >
>         > It does work on Dyalog, and gives the expected result.
>         >
>         > Regards,
>         > Elias