NICE-ESG-Libs Digest        Mon, 26 Feb 96       Volume 2 : Issue  21 

Today's Topics:
                         Class ARRAY proposal


  NICE Eiffel Standards Group -- Library Committee Mailing List

  To post to list:
     NICE-ESG-Libs@atlanta.twr.com
  To send mail to the Chairman of the committee:
     NICE-ESG-Libs-chair@atlanta.twr.com
  Administrative matters (sign up, unsubscribe, mail problems, etc):
     NICE-ESG-Libs-request@atlanta.twr.com

 
Date: Mon, 26 Feb 1996 18:10:25 -0800 (PST)
From: ericb@eiffel.com (Eric Bezault)
Subject: Class ARRAY proposal
To: NICE-ESG-Libs@atlanta.twr.com (NICE ESG LIBS)

Dear Library Committee members,

As the new delegate for ISE to the NICE Library Committee,
I reviewed once again Jim McKim's proposal for class ARRAY
and noticed some minor problems. Here they are:

o The precondition of feature `size' says:

        valid_indices: min_index <= max_index

  which prevents us from "sizing" to an empty array. Since it is allowed
  to create an empty array, I suggest that it should also be possible
  to size it to empty as well. So I propose the following precondition
  in replacement to the former one:
 
                valid_indices: min_index <= max_index + 1
 
  Note that it is the same as in `make'.

  [After a private discussion, Jim was concerned about the contract
  (especially the precondition) of feature `accommodate' as well. But
  I think the specification of `accommodate' is semantically correct
  and doesn't need to be changed.]

o The following precondition
 
                type_identity: same_type (other)
 
  from GENERAL is missing in `copy'. Jim told me that it was not intentional.
  So I propose this precondition to be added to feature `copy'.

o In the RB-3 amendment (and also in the original proposal and Bertrand's
  suggestion), one can read as one of the postconditions of `is_equal':
 
        definition: -- Result = (lower = other.lower) and (upper = other.upper)
                and for_all i, lower..upper (item (i)  = other.item (i))
 
  However the check for item equality is not valid when the bounds are not
  the same. In my view, that postcondition should be:
 
        definition: -- Result = (lower = other.lower) and (upper = other.upper)
              and then for_all i, lower..upper (item (i)  = other.item (i))
 
  Note the "and then" instead of "and" before "for_all".

o And last (but not least), since there is no agreement yet for what the
  signature of `to_c' (or `to_pointer') should be, I suggest that it would
  be removed from the standard. First, there is no reason to put something
  which is not standard in the Standard. It had been agreed at the time of
  ELKS '95 announcement that `to_c' should not be part of the Standard
  for the same reason (See NICE-ESG-Libs Digest V1 #230, V1 #231, V1 #232
  and v1 #241).
  Also, I don't think Jim's proposal can be adopted as-is as the standard
  if one can read:

          to_c: POINTER
                  -- Address of actual sequence of values, 
                  -- for passing to external (non-Eiffel) routines.
 
          CONCERN: ISE would like the type of this feature be ANY. Again,
            the current proposal takes no stand on this issue.
 
  My concern here is the same as the one about `resize' which had two
  interpretations (now split in two routines `size' and `accommodate')
  or `make_from_array' whose header comment is ambiguous.


I already notified our unofficial "Amiable Array Advocate" (Jim wants
to be called AAA ;-) about the above remarks and if Christine asked me
to do so I will propose amendments for them.

PS. Have a look at the Web page I'm putting together. Comments are welcome.
http://eiffel.com/private/nice/library


Best regards,


-- 
Eric Bezault 
Interactive Software Engineering Inc.
270 Storke Road, Suite #7                      / Tel +1 (805) 685-1006 \
Goleta, California 93117, USA                  \ Fax +1 (805) 685-6869 /