This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[tree-ssa] fix fortran debugging
- From: Richard Henderson <rth at twiddle dot net>
- To: gcc-patches at gcc dot gnu dot org
- Date: Fri, 28 Nov 2003 13:17:42 -0800
- Subject: [tree-ssa] fix fortran debugging
For some reason, gimplification of Fortran source is leaving most
of the program variables in a nested BLOCK, like so:
MAIN__()
{
{
int4 x;
}
}
and that nested BLOCK is associated with the outermost BIND_EXPR.
If we discard that BIND_EXPR immediately in lower_function_body,
then we discard basically all of the function's variables.
I didn't bother to find out why this was happening. The code in
lower_function_body just seemed asymmetric.
Bootstrapped and tested on i686-linux.
r~
* gimple-low.c (lower_function_body): Call lower_bind_expr
to handle the outermost BIND_EXPR.
Index: gimple-low.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/Attic/gimple-low.c,v
retrieving revision 1.1.4.13
diff -u -p -r1.1.4.13 gimple-low.c
--- gimple-low.c 16 Nov 2003 23:00:59 -0000 1.1.4.13
+++ gimple-low.c 28 Nov 2003 03:13:47 -0000
@@ -55,11 +55,13 @@ static bool expand_var_p (tree);
/* Lowers the BODY. */
void
-lower_function_body (tree *body)
+lower_function_body (tree *body_p)
{
struct lower_data data;
+ tree bind = *body_p;
+ tree_stmt_iterator i;
- if (TREE_CODE (*body) != BIND_EXPR)
+ if (TREE_CODE (bind) != BIND_EXPR)
abort ();
data.block = DECL_INITIAL (current_function_decl);
@@ -67,9 +69,10 @@ lower_function_body (tree *body)
BLOCK_CHAIN (data.block) = NULL_TREE;
TREE_ASM_WRITTEN (data.block) = 1;
- record_vars (BIND_EXPR_VARS (*body));
- *body = BIND_EXPR_BODY (*body);
- lower_stmt_body (*body, &data);
+ *body_p = alloc_stmt_list ();
+ i = tsi_start (*body_p);
+ tsi_link_after (&i, bind, TSI_NEW_STMT);
+ lower_bind_expr (&i, &data);
if (data.block != DECL_INITIAL (current_function_decl))
abort ();