Re: [freeelks-devel] INTEGER.abs

[Alexander K. <kw...@ah...>]

The accompanying bug is in a class invariant:
    sign * abs = item
Let's look at types of the features in class INTEGER_8:
    sign: INTEGER
    abs: INTEGER_8
    item: INTEGER_8
For value -128 we'll have
    sign = -1 (INTEGER)
    abs = -128 (INTEGER_8)
    item = -128 (INTEGER_8)
Then
    sign * abs = 128 (INTEGER)
and
    sign * abs = item
gives false.

If "sign" were of type INTEGER_8, then there were no problems
even with implementation of "abs" that returns negative result  for
minimum integer. Wouldn't it be better to change type of "sign" from
INTEGER to INTEGER_8?

BTW, according to ELKS'95 the type of "abs" should be "like Current"
and not "INTEGER*".

Alexander Kogtenkov