Re: [Bug-apl] Restrictions on localization of system variables?

[Juergen Sauermann]

Hi Kacper,

I believe it is OK to do something useful when otherwise you would have
to raise an error. Most extensions work like that, e.g. dyadic ⎕CR, ⍬.
And we have seen that even ISO isn't always right.

/// Jürgen

On 05/27/2014 06:30 PM, Kacper Gutowski wrote:
> On 2014-05-27 18:06:06, Juergen Sauermann wrote:
> > Note that there are some subtle differences between IBM APL2 and GNU APL
> > when localizing ⎕-vars. In IBM APL2 they are undefined after localizing
> > them.
> > In GNU APL they are pre-initialized with their respective default values.
> > This
> > gives simpler (and therefore faster code) because no check for validity is
> > needed
> > for the variables themselves, but also for the many internal users of them.
> This indeed makes things simpler and I like the idea, but APL2's behavior
> is mandated by ISO 13751 which at the beginning of chapter 12 says:
>
> > Note: When a system-variable-symbol is localised the initial class of
> > its associated system-parameter is nil.  If a primitive operation is
> > invoked that requires a system parameter whose class is currently nil,
> > the primitive operation signals implicit-error.  Therefore, a
> > conforming-program that localises system-variable-symbols should
> > assign them values from their internal-value-set before calling
> > primitive operations that require them.
>
> -k