On Thu, 2008-05-29 at 16:01 -0400, Pavel Roskin wrote:
> Hello!
>
> I'm running sparse (the current git version) on this file:
>
> static const int len = 64;
> void foo(void);
> void foo(void)
> {
> int buf[len];
> }
>
> sparse test.c
> test.c:5:10: error: bad constant expression
>
> But if I remove "const", the error message goes away.
It turns out the initializer with "const" is an "implied cast"!
value->type after constant_symbol_value() call in expand_dereference()
is EXPR_VALUE if const is not used and EXPR_IMPLIED_CAST is const is
used. It is EXPR_CAST for this input:
static int len = (int)64;
void foo(void);
void foo(void)
{
int buf[len];
}
That file would trigger the warning too. I believe "value" should be
expanded, at least in some cases. What matters is whether the value
evaluates to a constant at the compile time.
Even if we expand the value, problematic casts are still reported:
static int len = 0x100000000LL;
void foo(void);
void foo(void)
{
int buf[len];
}
warning: cast truncates bits from constant value (100000000 becomes 0)
This patch fixes the bogus errors. The testsuite passes. I ran sparse
on MadWifi where it was detecting those errors, and now it's showing
usable warnings instead.
Please review the patch. Sparse is still almost voodoo science to me.
Once I understand more, I'll write the patch description :-)
diff --git a/expand.c b/expand.c
index 032f0c5..5aa0a1d 100644
--- a/expand.c
+++ b/expand.c
@@ -565,6 +565,7 @@ static struct expression *constant_symbol_value(struct symbol *sym, int offset)
value = sym->initializer;
if (!value)
return NULL;
+ expand_expression(value);
if (value->type == EXPR_INITIALIZER) {
struct expression *entry;
FOR_EACH_PTR(value->expr_list, entry) {
Another question is why assigning a numeric constant to a constant
variable is an "implied cast". It should be possible to use numeric
constants (or anything that can be calculated at the compile time) to
initialize both constant and non-constant variables without any implied
casts as long as the value fits the destination type. Or perhaps the
assignments to non-constant variables should be implied casts.
--
Regards,
Pavel Roskin
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
