[PATCH] Fix type_info_expression()

[PATCH] Fix type_info_expression()

[Al Viro]

sizeof (typename){initializers}.foo is nice and valid C99 - it's parsed
as sizeof primary.foo <- sizeof postfix.foo <- sizeof postfix <- sizeof unary
<- unary.  Current type_info_expression() stops too early.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 expression.c                         |   12 ++++++++++--
 validation/sizeof-compound-postfix.c |    8 ++++++++
 2 files changed, 18 insertions(+), 2 deletions(-)
 create mode 100644 validation/sizeof-compound-postfix.c

diff --git a/expression.c b/expression.c
index f634b07..124a8ec 100644
--- a/expression.c
+++ b/expression.c
@@ -594,12 +594,14 @@ static struct token *type_info_expression(struct token *token,
 	struct expression **tree, int type)
 {
 	struct expression *expr = alloc_expression(token->pos, type);
+	struct token *p;
 
 	*tree = expr;
 	expr->flags = Int_const_expr; /* XXX: VLA support will need that changed */
 	token = token->next;
 	if (!match_op(token, '(') || !lookup_type(token->next))
 		return unary_expression(token, &expr->cast_expression);
+	p = token;
 	token = typename(token->next, &expr->cast_type, 0);
 
 	if (!match_op(token, ')')) {
@@ -616,8 +618,14 @@ static struct token *type_info_expression(struct token *token,
 	 * C99 ambiguity: the typename might have been the beginning
 	 * of a typed initializer expression..
 	 */
-	if (match_op(token, '{'))
-		token = initializer(&expr->cast_expression, token);
+	if (match_op(token, '{')) {
+		struct expression *cast = alloc_expression(p->pos, EXPR_CAST);
+		cast->cast_type = expr->cast_type;
+		expr->cast_type = NULL;
+		expr->cast_expression = cast;
+		token = initializer(&cast->cast_expression, token);
+		token = postfix_expression(token, &expr->cast_expression, cast);
+	}
 	return token;
 }
 
diff --git a/validation/sizeof-compound-postfix.c b/validation/sizeof-compound-postfix.c
new file mode 100644
index 0000000..3b716fe
--- /dev/null
+++ b/validation/sizeof-compound-postfix.c
@@ -0,0 +1,8 @@
+struct foo {int x, y;};
+static int a(void)
+{
+	return sizeof (struct foo){0,1}.y;
+}
+/*
+ * check-name: Handling of sizeof compound-literal . member
+ */
-- 
1.5.6.6

Re: [PATCH] Fix type_info_expression()

[Christopher Li]

On Sun, Feb 1, 2009 at 11:31 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> sizeof (typename){initializers}.foo is nice and valid C99 - it's parsed
> as sizeof primary.foo <- sizeof postfix.foo <- sizeof postfix <- sizeof unary
> <- unary.  Current type_info_expression() stops too early.

Thanks for the patch. Looks great.

> -       if (match_op(token, '{'))
> -               token = initializer(&expr->cast_expression, token);
> +       if (match_op(token, '{')) {
> +               struct expression *cast = alloc_expression(p->pos, EXPR_CAST);
> +               cast->cast_type = expr->cast_type;
> +               expr->cast_type = NULL;

Setting expr->cast_type = NULL here is not straightly necessary right?
The expr is EXPR_SIZEOF type, the expr->cast_type is not used in
evaluation_sizeof. Of course it doesn't hurt, I just want to get a confirm
that I understand it correct.y.

Chris

Re: [PATCH] Fix type_info_expression()

[Al Viro]

On Thu, Feb 05, 2009 at 10:29:19AM -0800, Christopher Li wrote:

> Setting expr->cast_type = NULL here is not straightly necessary right?
> The expr is EXPR_SIZEOF type, the expr->cast_type is not used in
> evaluation_sizeof. Of course it doesn't hurt, I just want to get a confirm
> that I understand it correct.y.

Take a look at evaluate_type_information().  If ->cast_type is not NULL,
it won't even look at ->cast_expression.  For sizeof (struct foo){0,1} that's
fine (you end with with sizeof(struct foo), essentially), but for
sizeof(struct foo){0,1}.x that'll give you the wrong answer.

IOW, it is necessary.

Re: [PATCH] Fix type_info_expression()

[Christopher Li]

On Thu, Feb 5, 2009 at 11:23 AM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> On Thu, Feb 05, 2009 at 10:29:19AM -0800, Christopher Li wrote:

> Take a look at evaluate_type_information().  If ->cast_type is not NULL,
> it won't even look at ->cast_expression.  For sizeof (struct foo){0,1} that's
> fine (you end with with sizeof(struct foo), essentially), but for
> sizeof(struct foo){0,1}.x that'll give you the wrong answer.
>
> IOW, it is necessary.

That is the subtle detail I am looking for.

I will apply your patch.

Thanks

Chris