Re: [Gcl-devel] Latest ansi-test fixes

[Paul F. Dietz]

Camm Maguire wrote:

> Are you sure?  Surely an array which can only be filled with a certain
> type of element is a subtype of an array which can be filled with
> anything, no?

No.  The spec says (on the page for SIMPLE-ARRAY):

    The types simple-vector, simple-string, and simple-bit-vector
    are disjoint subtypes of type simple-array.

SIMPLE-VECTOR is a subtype of (VECTOR T);
SIMPLE-BIT-VECTOR is a subtype of (VECTOR BIT);
SIMPLE-STRING is a subtype of the union of (VECTOR c) for c
from some collection of subtypes of CHARACTER (just which subsets
is implementation dependent; the collection must contain CHARACTER
and BASE-CHAR, at least.)

Consider also the difference between these and their supertype,
(VECTOR *).


>
  In any case, if I eliminate the relationships you
> propose, I then get the following inconsistencies:
>
> Test CL-TEST::TYPES-9 failed
> Form: (CL-TEST::TYPES-9-BODY)
> Expected value: NIL
> Actual value: ((SIMPLE-BASE-STRING SIMPLE-STRING ARRAY)
>                (SIMPLE-BASE-STRING SIMPLE-STRING SIMPLE-ARRAY)
>                (SIMPLE-BASE-STRING SIMPLE-STRING VECTOR)
>                (BASE-STRING STRING ARRAY) (BASE-STRING STRING VECTOR)
>                (SIMPLE-BASE-STRING STRING ARRAY)
>                (SIMPLE-BASE-STRING STRING VECTOR)).

The output of this test is a set of triples of types (T1 T2 T3) such
that (SUBTYPEP T1 T2) ==> t t, (SUBTYPE T2 T3) ==> t t, but
(SUBTYPEP T1 T3) ==> nil t (i.e., failure of transitivity).

All those relationships should be there -- strings are vectors and
arrays, and simple-strings are simple-arrays (but not simple-vectors).


        Paul