Re: Gnat GPL 2010 available soon (conditional expressions)

[Phil Clayton <phil.clayton@lineone.net>]

On Jul 6, 8:36 am, "Dmitry A. Kazakov" <mail...@dmitry-kazakov.de>
wrote:
> On Mon, 5 Jul 2010 15:47:41 -0700 (PDT), Phil Clayton wrote:
> > On Jul 5, 2:12 pm, "Dmitry A. Kazakov" <mail...@dmitry-kazakov.de>
> > wrote:
> >> On Sun, 4 Jul 2010 18:27:04 -0700 (PDT), Phil Clayton wrote:
> >>> In my view, condition expressions are a step in the right direction...
> >>> but I'm the sort of person who wants to write an array aggregate like
>
> >>>   (for I in 1 .. 5 => (for J in 1 .. 7 => (if I = J then 1.0 else
> >>> 0.0)))
>
> >>   (for I in 1..5 => (for J in 1..7 => Float (Boolean'Pos (I = J))))
>
> > Yep, I should have seen that one coming straight back... and taken
> > steps to avoid it :)
>
> BTW, is your construct 1) a macro expansion (as I remember PL/1
> preprocessor also had %for statement alike) and how far are you ready to go
> allowing "for" macros? Loops unrolling? Or else 2) it is an attempt to name
> the components of an array element index.

Definitely the latter...


> In the latter case
>
>    (I, J : others => <an expression involving I,J>)
>    (I : 1..5 => <an expression involving I>)
>    (Odd : 1 | 3 | 5 | 7 => <an expression involving Odd>, others => 0.0)
>    etc
>
> looks to me more consistent with existing Ada, e.g. when Error : others =>
> in exception handlers.

Interesting.  So, considering the difference between

  1.  X : T
  2.  for X in T

My take on Ada is that [1] just declares a new variable X of subtype T
(which may be initialized depending on context) whilst [2] is used
when declaring a new variable X of subtype T and then giving X the
values of T in some specified order, i.e. introducing some iteration
scheme.  If that is a fair assessment of things, then perhaps

  (for I in 1 .. 5 => F(I))

is more in keeping with existing Ada?  (If I'm talking absolute
drivel, please let me know...)


The concept of an order is important because the function F could have
side effects in the above example (though I would not advocate that as
good practice).  However, we would be able to write

  (for I in reverse A'Range => A(I))

to represent the reverse of the array A is perhaps a little confusing.


Also, it really only makes sense to have one variable holding the
array index.  For example, if we could write

  (I, J : 1 .. 5 => ...)

surely I and J would always be equal: there are only 5 elements in the
array.  Having

  (for I in 1 .. 5 => ...)

syntactically rules out multiple variables.

Phil