Gabriel Dos Reis <g...@integrable-solutions.net> writes:
> On Wed, Apr 25, 2012 at 10:20 AM, Dodji Seketeli <do...@redhat.com> wrote:
>> Gabriel Dos Reis <g...@integrable-solutions.net> writes:
>>
>>> On Wed, Apr 25, 2012 at 9:04 AM, Dodji Seketeli <do...@redhat.com> wrote:
>>>> In gcc/testsuite/gcc.dg/pr30457.c, the first warning was not being
>>>> emitted because the relevant location was inside the var_start macro
>>>> defined in a system header. It can even point to a token for a
>>>> builtin macro there. This patch unwinds to the first token in real
>>>> source code in that case.
>>>
>>> While you are at it, could you also use a non-zero value for the second
>>> argument argument to warning_at?
>>
>> I couldn't find any obvious value for it. I am thinking maybe it would
>> make sense to introduction a new -Wva_start to warn about possible
>> dangerous uses of the va_start macro and use that as the second argument
>> for the relevant warnings emitted by fold_builtin_next_arg. What do you
>> think?
>
> or -Wvarargs?
OK, I have cooked up a patch for this that I will send in a separate
thread shortly.
>>
>> In any case, this topic seems logically unrelated to the purpose of
>> enable -ftrack-macro-expansion by default, so IMHO it would be better to
>> do this in a separate self contain patch. Among other things, this
>> would ease possible future back-ports. Would you agree?
>
> OK.
While testing the separate patch, I realized that this one was missing
adjusting the location in another spot. So I have updated this patch
accordingly. The patch that adds -Wvarargs will come on top of it, and
will add some needed regression tests for the whole.
Tested on x86_64-unknown-linux-gnu against trunk. Bootstrap is still
running ...
* builtins.c (fold_builtin_next_arg): Unwinds to the first
location in real source code.
---
gcc/builtins.c | 23 +++++++++++++++++++----
1 files changed, 19 insertions(+), 4 deletions(-)
diff --git a/gcc/builtins.c b/gcc/builtins.c
index b47f218..5ddc47b 100644
--- a/gcc/builtins.c
+++ b/gcc/builtins.c
@@ -12095,6 +12095,13 @@ fold_builtin_next_arg (tree exp, bool va_start_p)
tree fntype = TREE_TYPE (current_function_decl);
int nargs = call_expr_nargs (exp);
tree arg;
+ /* There is good chance the current input_location points inside the
+ definition of the va_start macro (perhaps on the token for
+ builtin) in a system header, so warnings will not be emitted.
+ Use the location in real source code. */
+ source_location current_location =
+ linemap_unwind_to_first_non_reserved_loc (line_table, input_location,
+ NULL);
if (!stdarg_p (fntype))
{
@@ -12119,7 +12126,9 @@ fold_builtin_next_arg (tree exp, bool va_start_p)
{
/* Evidently an out of date version of <stdarg.h>; can't validate
va_start's second argument, but can still work as intended. */
- warning (0, "%<__builtin_next_arg%> called without an argument");
+ warning_at (current_location,
+ 0,
+ "%<__builtin_next_arg%> called without an argument");
return true;
}
else if (nargs > 1)
@@ -12154,7 +12163,9 @@ fold_builtin_next_arg (tree exp, bool va_start_p)
argument. We just warn and set the arg to be the last
argument so that we will get wrong-code because of
it. */
- warning (0, "second parameter of %<va_start%> not last named
argument");
+ warning_at (current_location,
+ 0,
+ "second parameter of %<va_start%> not last named
argument");
}
/* Undefined by C99 7.15.1.4p4 (va_start):
@@ -12164,8 +12175,12 @@ fold_builtin_next_arg (tree exp, bool va_start_p)
the default argument promotions, the behavior is undefined."
*/
else if (DECL_REGISTER (arg))
- warning (0, "undefined behaviour when second parameter of "
- "%<va_start%> is declared with %<register%> storage");
+ {
+ warning_at (current_location,
+ 0,
+ "undefined behaviour when second parameter of "
+ "%<va_start%> is declared with %<register%> storage");
+ }
/* We want to verify the second parameter just once before the tree
optimizers are run and then avoid keeping it in the tree,
--
Dodji
