ping
On 11/7/26 10:48 AM, Jerry D wrote:
On 11/6/26 3:12 PM, Jerry D wrote:
The following attached patch regression tested on x86_64 and arm.
OK for mainline?
Regards,
Jerry
I did some additional checking while working on another PR and discovered I
missed a dependency. Attached here is a v2 of the patch which now includes its
own CLASS copy helper and no longer depends on PR51284 or PR128017-3 which I am
working on.
Regression tested on x86_64.
Ok for mainline?
Jerry
---
[PATCH] fortran: [PR128478] Copy a parenthesized scalar CLASS actual argument
A parenthesized scalar CLASS variable passed as an actual argument was
not copied, so the dummy aliased the variable, and passing it to a TYPE
dummy with allocatable components segfaulted, because the components of
the CLASS container were copied and freed as if it were the derived
type. Make the value of a parenthesized scalar CLASS variable a copy in
gfc_conv_expr_op, which also serves the ASSOCIATE selector, keep the
free of the copy until after the call, and leave its components to the
copy's own finalization.
Assisted-by: Claude Opus 5.5
PR fortran/128478
gcc/fortran/ChangeLog:
* trans-expr.cc (gfc_conv_expr_op): Copy a parenthesized scalar
CLASS variable.
(gfc_conv_procedure_call): Do not copy or free the allocatable
components of the copy of a parenthesized scalar CLASS variable.
(gfc_conv_expr_reference): Keep the free of that copy in se->post.
* trans-stmt.cc (trans_associate_var): Remove the copy of a
parenthesized CLASS selector, now done by gfc_conv_expr_op.
gcc/testsuite/ChangeLog:
* gfortran.dg/asan/paren_class_copy_1.f90: New test.
---