[PATCH] Refactor attribute arguments

[Eli Friedman]

On Mon, Jul 22, 2013 at 2:37 PM, Aaron Ballman <aaron at aaronballman.com> wrote:
> My testing showed it to be the case -- when using the index returned
> from checkFunctionOrMethodArgumentIndex, it was always off by one or
> two depending on context.  Specifically:
>
> const char* g2(const char*) __attribute__((format_arg(2)));
>
> checkFunctionOrMethodIndex would return index 0 because it cares about
> the const char * -- adding 1 to it when passing in the argument to
> addAttr (to make it one based instead of zero based) was not
> sufficient; I would get crashes in other parts of the system for that
> test case.  Having it pass in two is consistent with the previous
> behavior.

Oh, right... checkFunctionOrMethodArgumentIndex is performing the
conversion, but Sema for format attributes expects the unconverted
version.  We should probably clean that up at some point, but that
shouldn't block this patch.

-Eli