Le 15/11/2026 à 04:14, Jerry DeLisle a écrit :
See attached patch. For w = z, with w an allocatable TYPE(t) array and
z a CLASS(t) array, reallocation on assignment took the element size
from the type node of z, which is the class container (72 bytes at rank
1), not t. A t larger than that overflowed w, and the new test
segfaults on trunk; a smaller t was over-allocated and w got the
container size as its span. The variable and expr have the same
declared type here (Table 10.8), so the patch takes the element size
from the declared type of z, as is already done when w is CLASS too.

The fix is in gfc_alloc_allocatable_for_assignment in trans-array.cc;
the rest is the new test.

Regression tested on x86_64.

OK for mainline?

Regards,

Jerry

---

fortran: [PR128736] Heap overflow assigning CLASS array to TYPE array

When a CLASS array was assigned to an allocatable array of its declared
type, reallocation on assignment sized each element of the variable as
the class container instead of the declared type.  A type larger than
the container overflowed the new array, and a smaller one was
over-allocated and given the wrong span.  Take the element size from the
declared type of the CLASS expression.

Assisted-by: Claude Opus 5.5

        PR fortran/128736

gcc/fortran/ChangeLog:

        * trans-array.cc (gfc_alloc_allocatable_for_assignment): Take the
        element size of a variable that is not polymorphic from the declared
        type of a polymorphic expression.

gcc/testsuite/ChangeLog:

        * gfortran.dg/pr128736.f90: New test.

OK.

Reply via email to