J3/06-278

Date: 10 August 2006
To: J3
From: Dick Hendrickson
Subject: clause 4 problems
References: J3/06-007

Questions and comments that are more than editorial.

[41:9] Can't a processor provide derived types as an
extension? How is providing a "tensor" type
different from Quad?

[41:30] As with the above. They can also be provided by
processor magic. Add "or by the processor"

[42: note 4.1, last sentence] Are (type bound?) polymorphic
generics resolvable at compile time?

[42:19-21] Should we add UDDTIO to the list? Can it change
the parameters

[50:16] Is this what we want now that we have 18 digit
integers required? Changing it would be a (minor?)
incompatibility with F03. It seems odd to have
(333333333333333333, 1.0) be the same as
(333333300000000000, 1.0)
This will also mess with clause 7

[51:27] Add "elemental," to the list. Back in the good old
days, an elemental function was pure, so it wasn't
needed in the list. Now, we have IMPURE elemental
functions.

/Begin RANT
We have this bad habit of using magic to make
our lists. Rather than say "A, B, C, D, E, or
F", someone notes that "All Es are either D or
F" and bingo, the list becomes "A, B, C, D, or
F. We should stop doing that
/end RANT

[54:32] Must the additional representation methods be dense
like the ones up to 4*(integer size) are. Or,
could a processor allow 1 through 128 and then 143
and 231?

[55:30-31] Is this what we want? For integers surely all
processors overflow for
9999999999999_TwoByteKind.
Ditto for 1.0E666666666666_IEEEsingle. Why
would anyone want to just use the right hand
bits of a typed constant?

[62:22 and 29] I don't understand C447 and C451. Are they
correct?

[62:note 4.32 end] Can UDDTIO also determine the shape?

[65:2] Says pointer components can only be initialized to
disassociated. See [65:30] which seems to say they
can be associated to a real target.

[65:14] Does C459 need to be worded similar to C447?

[65:26-27] Why don't we also allow module variables to be

[72:1] See comment about [130:1]