(j3.2006) More about Note 5.10

[Bill Long]

Van Snyder wrote:

>I couldn't find normative support for Note 5.10 [91:8+1-3].
>
>There are constraints (C725 [163:32] and C727 [164:8]) that prevent
>pointer assignment across images, but nothing (that I could find in
>7.2.1.2) that prevents intrinsic assignment across images if the type is
>one that has pointer components.  I also couldn't find anything in 6(.0)
>about cross-image pointer dereferencing.
>

There is an edit in 07-152 to make such an assignment cause the pointer 
association status to become undefined.

Cheers,
Bill

>
>Perhaps instead of trying (apparently unsuccessfully) to constrain type
>definitions and object declarations so that cross-image pointer
>dereference is impossible, we should either prohibit that, or just say
>its effect is either processor dependent or undefined.  [117:3-4] seems
>like a good place.  Perhaps add "and the reference, the pointer, and the
>target are all on the same image" after "defined".  Then Note 5.10
>should be worded differently.
>
>Maybe add near there "An object of type IMAGE_TEAM or an object that has
>an ultimate component of type IMAGE_TEAM shall not be referenced on a
>different image from the object" somewhere near there.
>
>Then we could delete C446 [63:26] and C526 [91:6].
>
>  
>

-- 
Bill Long                                   longb at cray.com
Fortran Technical Support    &              voice: 651-605-9024
Bioinformatics Software Development         fax:   651-605-9142
Cray Inc., 1340 Mendota Heights Rd., Mendota Heights, MN, 55120