https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121805
kargls at comcast dot net changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |kargls at comcast dot net
--- Comment #3 from kargls at comcast dot net ---
Here's a slightly shorter testcase. It shows what I think is an off-by-one
issue.
program pointer_array_to_struct
implicit none
type str
integer data
integer(1) a
end type
type(str), target :: e(8)
integer i, array_sum
integer, pointer :: dd(:)
integer idx(8)
idx = [(i,i=1,8)]
e%data = idx
dd=>e%data
array_sum = sum(idx(dd))
if (array_sum /= 36) stop 1
end program
If I compiler with -fbounds-check -fdump-tree-original. One sees that idx is
correctly filled, but the array indexing is messed up. Here's the relevant
part with long lines wrapped.
{
integer(kind=4) val.2;
integer(kind=8) D.4692;
val.2 = 0;
D.4692 = dd.dim[0].ubound - dd.dim[0].lbound;
{
integer(kind=8) S.3;
S.3 = 0;
while (1)
{
if (S.3 > D.4692) goto L.2;
{
integer(kind=4) D.4694;
integer(kind=8) D.4695;
D.4694 = *(integer(kind=4) *) (dd.data + (sizetype)
((dd.dim[0].stride *
NON_LVALUE_EXPR <S.3>) * 4));
D.4695 = (integer(kind=8)) D.4694;
I think the above should be D.4695 = (integer(kind=8)) D.4694 + 1.
if (D.4695 <= 0)
{
_gfortran_runtime_error_at (&"At line 19 of file
gg.f90"[1]{lb: 1 sz: 1},
&"Index \'%ld\' of dimension 1 of array \'idx\' outside of
expected
range (%ld:%ld)"[1]{lb: 1 sz: 1}, D.4695, 1, 8);
}
if (D.4695 > 8)
{
_gfortran_runtime_error_at (&"At line 19 of file
gg.f90"[1]{lb: 1 sz: 1},
&"Index \'%ld\' of dimension 1 of array \'idx\' outside of
expected
range (%ld:%ld)"[1]{lb: 1 sz: 1}, D.4695, 1, 8);
}
val.2 = idx[D.4695 + -1] + val.2;
Note, we are correctly subtracting 1 to get zero-index C semantics.
}
S.3 = S.3 + 1;
}
L.2:;
}
array_sum = val.2;
}
