On Tue, Sep 24, 2019 at 04:23:34PM -0700, Linus Torvalds wrote:
> On Tue, Sep 24, 2019 at 12:09 AM Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
> >
> > The problem is the sign isn't extended properly when this casts an int
> > to a long. The expr->ctype has to be the original int ctype for the
> > cast_value() call so that the "value = get_longlong(old);" expansion
> > works correctly.
>
> What happens if you just remove the
>
> if (ctype->bit_size == base_type->bit_size) {
> expr->ctype = base_type;
> continue;
> }
>
> part entirely? IOW, leave just
>
> cast_value(expr, base_type, expr, ctype);
> expr->ctype = base_type;
>
> in place unconditionally? I _think- that should be the simpler correct
> fix, but I'll leave it to Luc to think about it more.
Yes, I agree. Dan's patch is the obvious one solving the problem
(expr->ctype adjusted too early) but yours should be equivalent since
bit_size check is also done in cast_value(). I still need to run the
tests tough.
That being said, I think that using cast_value() is error-prone:
expr's ctype should probably be changed there. but this will be in
a separate patch anyway.
> That simpler alternate patch attached,
Thanks.
-- Luc
