Le 25/08/2026 à 01:38, Jerry D a écrit :
See attached 5 of 5 patch
Regression tested on x86_64
Regards,
Jerry
---
fortran: [PR49802]-5 Absent optional actual argument
passed to a VALUE dummy
The private copy made for a VALUE dummy dereferenced the actual
argument unconditionally, so passing an absent optional actual argument
to an optional VALUE dummy segfaulted. Both the array copy and the
copy made for a CHARACTER dummy of assumed or non-constant length have
to be suppressed when the actual argument is absent.
gfc_conv_subref_array_arg already has the machinery: given the symbol
of the actual argument it suppresses its copy and passes a null pointer
instead, so the array case just has to hand it that symbol rather than
the symbol of the procedure being called. This is what makes the two
pass_optional fixes of the first patch of this series reachable.
conv_dummy_value has no such machinery and guards its CHARACTER copy
directly, passing a null pointer and a zero length when the actual
argument is absent.
PR fortran/49802
gcc/fortran/ChangeLog:
* trans-expr.cc (conv_dummy_value): Suppress the CHARACTER copy
and pass a null pointer and a zero length when the actual argument
is an absent optional one.
(gfc_conv_procedure_call): Give gfc_conv_subref_array_arg the
symbol of the actual argument, so that an absent optional one
suppresses the copy.
gcc/testsuite/ChangeLog:
* gfortran.dg/value_18.f90: New test.
---
OK.
