Re: [Fortran, Patch, PR117643] Implement F_C_STRING()

[Harald Anlauf]

Hi Steve,

Am 18.12.24 um 22:37 schrieb Steve Kargl:

So, it seems that this section of the patch

+         gfc_init_se (&asis_se, se);
+         gfc_conv_expr (&asis_se, asis);
+         gfc_add_block_to_block (&se->pre, &asis_se.pre);
+         tmp = fold_build3_loc (input_location, COND_EXPR, void_type_node,
+                                asis_se.expr, then_branch, else_branch);


needs to be protected by some code that starts with

present = gfc_conv_expr_present (asis->symtree->n.sym);

I tried the following patch on top of yours:

diff --git a/gcc/fortran/trans-intrinsic.cc b/gcc/fortran/trans-intrinsic.cc
index 2370b6dad55..886f999f206 100644
--- a/gcc/fortran/trans-intrinsic.cc
+++ b/gcc/fortran/trans-intrinsic.cc
@@ -10220,6 +10220,16 @@ conv_isocbinding_function (gfc_se *se, gfc_expr
*expr)

          gfc_init_se (&asis_se, se);
          gfc_conv_expr (&asis_se, asis);
+         if (asis->expr_type == EXPR_VARIABLE
+             && asis->symtree->n.sym->attr.dummy
+             && asis->symtree->n.sym->attr.optional)
+           {
+             tree present = gfc_conv_expr_present (asis->symtree->n.sym);
+             asis_se.expr = build3_loc (input_location, COND_EXPR,
+                                        logical_type_node, present,
+                                        asis_se.expr,
+                                        build_int_cst
(logical_type_node, 0));
+           }
          gfc_add_block_to_block (&se->pre, &asis_se.pre);
          tmp = fold_build3_loc (input_location, COND_EXPR,
void_type_node,
                                 asis_se.expr, then_branch, else_branch);

This does fix the case of optional dummy asis for me.

There is another case I found while playing which is rejected:

   print *, f_c_string(c_char_"abc", asis)

I am not entirely sure if this is the right way to encode string
literals that are interoperable, but Intel's ifx accepts it, and
I think we should consider this as legal.  Or what do you think?

Besides the above issues, and several issues with whitespace in
your patch (tabs/blanks), and extending the testcase so that
it covers the optional dummy for 'asis', I a fine with it.

Thanks for the work!

Harald