Bug reported (with longer test case) by Maciej Zwierzycki
at http://gcc.gnu.org/ml/fortran/2009-06/msg00254.html
The following program should produce 1 2 -42 -42
but it produces 1 -42 2 -42
For
a(1,:) = func()
gfortran decides to pass result value by reference - and uses an array
descriptor for this (why?):
func (struct array1_integer(kind=4) & __result)
and
func (&parm.7);
However, as "func(2)" is an explicit-shaped array, it should be contiguous and
thus for "call sub(func)" simply the address is passed, ignoring the stride:
sub ((integer(kind=4)[0:] *) parm.4.data);
At some place the copy-in/copy-out must happen. Actually, I had assumed that
already the call to "func" would cause a temporary be created. That's actually
what happens in case of g95:
SC1 = *func_ ();;
Test case:
integer :: a(2,2)
a = -42
a(1,:) = func()
print *,a
contains
function func()
integer :: func(2)
call sub(func)
end function func
subroutine sub(a)
integer :: a(2)
a = [1,2]
end subroutine
end
--
Summary: Wrong code due to missing copy-in/copy-out stried array
to assumed-size dummy
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: major
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: burnus at gcc dot gnu dot org
OtherBugsDependingO 40443
nThis:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40551
