[C++] I know NAN != NAN but shouldn't literal(NAN) == literal(NAN)?

[westonpace]

Describe the usage question you have. Please include as many useful details as possible.

I'm pretty sure of this but wanted a sanity test. Also, should literal(quiet_NAN) == literal(signaling_NAN)? My intuition is "no".

Or...more concretely...should these test cases pass?

  // This currently fails
  EXPECT_EQ(literal(std::numeric_limits<double>::quiet_NaN()),
            literal(std::numeric_limits<double>::quiet_NaN()));
  // These currently pass
  EXPECT_NE(literal(std::numeric_limits<double>::quiet_NaN()),
            literal(std::numeric_limits<double>::signaling_NaN()));
  EXPECT_NE(literal(std::numeric_limits<double>::quiet_NaN()),
            literal(std::numeric_limits<float>::quiet_NaN()));

If the answer is "yes, those two expressions are equivalent" then I can fix it in #15180 . I already have a fix but wanted to confirm my understanding.

Component(s)

C++

[westonpace]

Actually, I can't find a good way to test if a NaN is a quiet nan or a signalling NaN so now I'm tempted to say literal(quiet_NaN) == literal(signal_NaN) as well.

[westonpace]

Ah, I can use issignaling which is in <cmath> but has no c++ equivalent.

[bkietz]

I think that we don't need to support signalling nans. Quiet nans may be produced by floating point expressions with invalid arguments, but signalling nans can only be produced explicitly- which is not functionality I think we'd find valuable. For our purposes, I don't think we need to check for signalling nans.

I do think that literal(nan) == literal(nan) should be true- as you say those two expressions are equivalent. I think just to disambiguate the situation the unit test should also assert that equal(literal(nan), literal(nan)) evaluates to false and clarify why that difference exists with a comment

[jorisvandenbossche]

This was closed by #15180