On 12/02/2014 09:14 AM, Marek Polacek wrote:
On Fri, Nov 28, 2014 at 09:28:03AM -0500, Jason Merrill wrote:
I was thinking even more detailed: one diagnostic for negative count, one
diagnostic for count larger than the precision of the lhs, and then a third
for overflow.
Alright, done.
Thanks. These errors also need to be conditional on (!ctx->quiet), and
if we see one of these conditions we need to set *non_constant_p.
This was a little bit problematic, because I don't really know where
that formula comes from. We use it in c-ubsan.c, so I've just taken
it from there. It was Jakub who came with that formula, so maybe he
remembers its origin. I wrote something more to the comment describing
it, even though that might not be enough.
Looks good.
Jason
