Le 12/10/2026 à 17:20, Paul Richard Thomas a écrit :
Hello Mikael,
Thanks for taking a look at this. I have incorporated your suggestion
in the attached, revamped patch.
Slightly more substantial modification has resulted for a review by
Claude, which pointed out that co-dimension variables were still
giving double free signals and already deallocated variables,
deallocated again with stat =, were segaulting. These defects are
fixed.
The patch passes regression testing on FC44/x86_64. OK for mainline?
Cheers
Paul
Hello Paul; just one nit...
diff --git a/gcc/fortran/trans-stmt.cc b/gcc/fortran/trans-stmt.cc
index 19817180acd..6c3e9cdbb59 100644
--- a/gcc/fortran/trans-stmt.cc
+++ b/gcc/fortran/trans-stmt.cc
@@ -7968,6 +7968,21 @@ gfc_trans_deallocate (gfc_code *code)
se.expr, pdt_rank);
}
+ if (tmp)
+ {
+ tree null_cond, zero;
+ null_cond = se.expr;
+ if (GFC_DESCRIPTOR_TYPE_P (TREE_TYPE (se.expr)))
+ null_cond = gfc_conv_descriptor_data_get (null_cond);
+ zero = build_int_cst (TREE_TYPE (null_cond), 0);
+ null_cond = fold_build2_loc (input_location, NE_EXPR,
+ logical_type_node,
+ null_cond, zero);
+ tmp = build3_loc (input_location, COND_EXPR, void_type_node,
+ null_cond, tmp,
+ build_empty_stmt (input_location));
Indentation is wrong here; OK with that fixed.
+ }
+
if (tmp)
gfc_add_expr_to_block (&se.pre, tmp);
