http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52102
--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-02-03
13:37:39 UTC ---
Related issue with MOLD=, here one gets with "x = 5" default initialization and
allocate (two%a(8), mold=t(4))
print '(*(i2))', two%a(:)%x
the result:
0 0 0 0 0 0 0 5
That is: only the last element gets the value.
>From the dump:
two.a._data.offset = -1;
...
struct t t.2;
t.2.x = 5;
D.1886 = t.2;
__vtab_MAIN___T._copy (&D.1886,
&(*(struct t[0:] * restrict) two.a._data.data)[two.a._data.offset +
8]);
Note the constant offset of "8", which matches the ubound. Note further that
the _copy construct is *not* in a scalarizing loop but only called once for the
last element.
* * *
The following looks odd:
gfc_trans_allocate (gfc_code * code)
{
...
if (code->expr3 && !code->expr3->mold)
{
...
/* Do a polymorphic deep copy. */
...
actual->next->expr = gfc_copy_expr (al->expr);
dataref = actual->next->expr->ref;
if (dataref->u.c.component->as)
The latter makes no sense in this case as the base object is not the class
container. We need to access the last component ref on the ref list.
