[libcxx] r203126 - Fix a couple of -Wabsolute-value warnings in the libc++ tests

[Joerg Sonnenberger]

On Thu, Mar 06, 2014 at 01:10:54PM -0800, Arthur O'Dwyer wrote:
> On Thu, Mar 6, 2014 at 1:03 PM, David Blaikie <dblaikie at gmail.com> wrote:
> > On Thu, Mar 6, 2014 at 12:15 PM, Joerg Sonnenberger <joerg at britannica.bec.de> wrote:
> >> On Thu, Mar 06, 2014 at 04:27:18PM -0000, Marshall Clow wrote:
> >>>
> >>> -    static_assert((std::is_same<decltype(fabsf(0)), float>::value), "");
> >>> -    static_assert((std::is_same<decltype(fabsl(0)), long double>::value), "");
> >>> -    assert(fabs(-1) == 1);
> >>> +    static_assert((std::is_same<decltype(fabsf(0.f)), float>::value), "");
> >>> +    static_assert((std::is_same<decltype(fabsl(0.L)), long double>::value), "");
> >>> +    assert(fabs(-1.f) == 1);
> >>
> >> This looks bogus as in: it shouldn't warn here.
> >
> > What basis did you have in mind? That the constant is in the
> > representable range for the function being called, even though it's
> > not of the right type?
> 
> I guess it makes sense to warn about fabsf(int), because int->float
> conversion can be lossy.
> But surely fabs(int) and fabsl(int) are 100% fine!

There are two cases to distinguish here: constants and variable
arguments. I am fine with the warning for variable arguments, because
that would involve an implicit and potentially undesired float
conversion. Whether or not that is potentially lossy is not relevant for
me. On the other hand, with a constant (literal) argument, clang can
easily check whether the conversion gives the expected identical result.
I don't think it should warn then.

Joerg