The code generated to call RANDOM_SEED doesn't take into account the existence
of a possible default integer kind other than 4. For example:
$ cat o.f90
integer :: i, j(8)
call random_seed (size=i)
print *, i
call random_seed (get=j)
print *, j
call random_seed (put=j)
end
$ gfortran o.f90 -fdefault-integer-8 -fdump-tree-original && ./a.out
8
1556651063369256213 1358030301163568053 1969796612819413169
1882033285047892425 0 0
0 0
and the tree dump clearly shows that wrong type arguments are passed to
_gfortran_random_seed, which would lead to worst things on big-endian
platforms:
int8 j[8];
int8 i;
_gfortran_random_seed (&i, 0B, 0B);
I think it would be not to hard to make two different library functions, one
for int4 and the other one for int8. The int8 would then report a SIZE of the
seed that is half the size for int4, and that should do the trick.
Thomas, any opinion on that?
--
Summary: RANDOM_SEED is broken with -fdefault-integer-8
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: fxcoudert at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33054
