http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52774
Bug #: 52774
Summary: A check is needed to prevent deallocation in
realloc-lhs
Classification: Unclassified
Product: gcc
Version: 4.6.3
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: patnel97269-gfort...@yahoo.fr
A check is needed to prevent deallocation and nullification of D.1880.p.data if
it is equal to p1.p.data, after the function call.
This has been reported at http://gcc.gnu.org/ml/fortran/2012-03/msg00141.html
Here is the message of Paul Richard Thomas :
A reduced and tidied testcase that exhibits the
problem is:
program test
implicit none
type po
real(8), dimension(:), allocatable :: p
end type
type(po) :: p1
allocate(p1%p(1))
p1=toto(p1)
print *, p1%p
contains
function toto(dd) result(y)
type(po) ::dd,y
deallocate(dd%p)
end function toto
end program test
The call to toto gives the following code:
D.1880 = p1;
p1 = toto (&p1);
if (D.1880.p.data != 0B)
{
__builtin_free ((void *) D.1880.p.data);
}
D.1880.p.data = 0B;
Thus, if dd is deallocated within toto, without reassigning something
to dd%p, the address of p.data is retained by D.1880 and so it is
doubly freed. This can be tested by adding an assignment to dd%p,
after the deallocation, whereupon everything works as it should. For
example,
program test
implicit none
type po
real(8), dimension(:), allocatable :: p
end type
type(po) :: p1
allocate(p1%p(1))
p1=toto(p1)
print *, p1%p
contains
function toto(dd) result(y)
type(po) ::dd,y
deallocate(dd%p)
dd%p = [1d0, 2d0]
y = dd
end function toto
end program test
