Re: Remarks on SWI 7

Richard O'Keefe:
>On 29/11/2013, at 4:36 AM, Ulrich Neumerkel wrote:
>> Can we agree the Quintus manual differs?
>> 
>> http://quintus.sics.se/isl/quintus/html/quintus/mpg-ref-length.html
>
>We can agree that the manual doesn't say that there is an
>exception if the List argument is malformed, BUT it's arguably
>vague.  The whole of the text talks only about what happens
>when it is a list (possibly of indefinite length).  It's not
>*obvious* what would happen for length(42, 2).

For this reason there is a "Template and Modes" subclause in the
definition of a predicate.  This makes clear what error checking is
intended.  It is

length(?term, ?integer)

Which means: no checking on the first arg, checking for type integer
on the second.  No direct instantiation errors.

http://www.complang.tuwien.ac.at/ulrich/iso-prolog/prologue#length

would the list be checked, it would be

length(?list, ?integer)

I am told, also Quintus 3.5 does not check the list.  So the
documentation is accurate.

>This is actually one of the reasons why a standard scrutinised
>by a legion of nitpickers is so useful.  "Things that go without
>saying" sometimes need saying, so that everyone says the same.

Yes! p.p.3.1 Procedurally, ... is not perfect.

>What I was trying to say was that I have strong sympathy with
>Jan's preference for length/2 BUT if he's right, _everyone_ should
>do the same, and he'd do more good for the world by
> (a) conforming to the standard
> (b) adding a new operation list_length(List, Length) doing what
>     he wants
> (c) then lobbying for either the adoption of list_length/2
>     or a revision to length/2.


Originally, I also favoured checking.  In fact, I wrote that checking
length/2 for SWI.  But then there were also many arguments against it.
Who would go that far as to insist on checking the first argument of
append/3? In any case, that is the very idea of having as many Prolog
people together to resolve such issues.