Re: (*) -> 1

[Tassilo Horn]

Jean Louis <bugs@gnu.support> writes:

>> >> Yes, and I think it's seriously wrong with
>> >> 
>> >>   : (+)
>> >>   -> NIL
>> >> 
>> >> where its docs say
>> >> 
>> >>   Returns the sum of all num arguments. When one of the arguments
>> >>   evaluates to NIL, it is returned immediately.
>> >
>> > For some reason PicoLisp is quite different than other Lisp. I have
>> > asked author about it.
>> >
>> > 15:09 <abu[m]> It is a "feature" that NIL propagates through
>> >                arithmetics
>> 
>> Well, but with (*) and (+), there is no single NIL involved!  And in
>> Elisp (+ nil), where actually a nil is involved, you get an error.
>
> What we can learn from PicoLisp is that there was no use for (*) ➜ 1
> and that programs work, GUI applications and Android/Replicant work,
> and there was no use of (*) ➜ 1 so far.

You are jumping to conclusions.  If someone needed a mathematically
sound product in PicoLisp, they might have defined it as

  (de product @
    (if (not (args))
       1
       (* (next) (apply 'product (rest)))))

[Not sure if that's correct, I've just skimmed the docs.]

>> > 15:09 <jmarciano> How does it help instead of providing identity
>> >                   elements?
>> > 15:09 <abu[m]> (*) especially was not contemplated though, it is a
>> >                pretty useless call
>> > [...]
>> > 15:12 <abu[m]> What is a call like (*) useful for?
>> >
>> > As you see, author also asked naturally why is it useful.
>> 
>> So go and ask why he thinks (apply '+ ()) -> NIL is more useful than
>> 0 given that the sum of the empty set of numbers _is_ 0.
>
> I will ask.

I'm interested in the reply.  I feel it might be just an oversight which
is hard or impossible to fix now.

> But docstring does not speak of empty sets.

Yes, so the docstring is at least incomplete because it doesn't include
the completely valid case where no args are given.

> You introduce "sets" where there is not direct relation to it.

17 is an element of the set of integers, isn't it?

> (+ &rest NUMBERS-OR-MARKERS)
>
> Return sum of any number of arguments, which are numbers or
> markers. Of course I get confused.

Why?  At least when ignoring markers which happen to have an integer
representation which is an implementation detail.

> `apply' can be used with (apply '+ '(a b)) as why would you need in
> apply for addition two arguments?

You don't but you can use it if (a b) is not a literal list but a
variable, i.e., use (+ a b) or (apply #'+ my-list-of-numbers).

>> It's good to signal an error when the expression is wrong as does
>> Elisp with
>> 
>>   (+ nil)
>>   (* 1 2 nil)
>>   (apply #'+ (list 1 nil 19))
>>   (+ 2 "i am not a number")
>
> That is exactly my point, what you see useful there, I see too.

Good! :-)

> Making it less error prone with useless default identity elements
> hides the real event preceding the operation.

Let's agree to disagree then.  In my book, it is useful to have
mathematically sound behavior by default.  If you have a reason to
handle some edge-cases differently in some application (which is totally
possible!), then define your own function which does what you wish.

Bye,
Tassilo