[J3] J3 Fortran interp letter ballot #40 - due 19-December-2024

Mark LeAir mleair at nvidia.com
Wed Nov 27 06:44:10 UTC 2024


Yes  No   Number    Title

-Y-  ---  F23/003  Conflicting rules for COMMON block names
-Y-  ---  F23/004  OUT_OF_RANGE and ROUND argument
-Y-  ---  F23/005  Defined assignment/operators and dynamic type
-Y-  ---  F23/006  Underflow in IEEE_SCALB
-Y-  ---  F23/008  Real argument I in IEEE_SCALB
-Y-  ---  F23/009  Coarray subobject of component
-Y-  ---  F23/010  MOVE_ALLOC with coarray arguments
-Y-  ---  F23/011  NULL and procedure pointers
-Y-  ---  F23/012  Coarray correspondence in DEALLOCATE
-Y-  ---  F23/013  BOZ literals in interoperable enumerators
---  -N-  F23/014  Parent component naming
-Y-  ---  F23/015  Coindexed objects in structure constructors
-Y-  ---  F23/016  Segments associated with allocation
-Y-  ---  F23/017  CFI_establish nonalloc nonpointer null base address
-Y-  ---  F23/018  Correspondence of unallocated coarrays

Nvidia disagrees with the proposed answer to Interp F23/014. The answer conflicts with the behavior of 6 out of 8 Fortran compilers that we tested. A consensus already exists among implementors regarding this interp. Compilers that process a renaming of a derived parent type through a use statement should also assume the new name when referencing the parent component.

The proposed answer leaves the original name of the parent component intact, which can obscure its identity for implementors and users of type extensions, particularly when the original parent type name is renamed in a module that is hidden inside a library or a module file.

Currently, 6 of the 8 tested compilers process the following example without issue. Under the proposed answer, however, these compilers would now need to report an error at the commented source line below. This change could become a significant inconvenience for some users, especially those who may not know the original names of hidden parent types in advance.

module m1
  type foo
  end type
end
module m2
  use m1, derived_type => foo
end
use m2, dt => derived_type
type, extends(dt) :: mytype
  integer :: foo ! this should not be an error, but now it would be with 6 of the 8 tested compilers
end type
end


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.j3-fortran.org/pipermail/j3/attachments/20241127/b6eec5a5/attachment.htm>


More information about the J3 mailing list