test() doesn't distinguish plain NA_real_, NaN

[MichaelChirico]

# no error
data.table:::test(1, NA_real_, NaN)
# but it's not totally NA-NaN-unaware
data.table:::test(1, list(NA_real_), list(NaN))
# Test 1 ran without errors but failed check that x equals y:
# > x = list(NA_real_) 
# First 1 of 1 (type 'list'): 
# [[1]]
# [1] NA

# > y = list(NaN) 
# First 1 of 1 (type 'list'): 
# [[1]]
# [1] NaN

[MichaelChirico]

@jangorecki continuing discussion (#7864 (comment)) here.

Further context is #7856, #7854.

Maybe the right solution is instead to make list(NA_real_) vs. list(NaN) resolve to TRUE as it does in other contexts? But IIRC that was intentional for some tests in froll context where the goal was to be more "correct" about NA vs. NaN. WDYT?

[jangorecki]

I think we are correct but that behavior is not portable across different platforms. IMO we should not worry about that aspect as long as R core suggests to ignore it.

[MichaelChirico]

Is there a simple behavioral test we could use to skip on platforms where we think the behavior is wrong/worse than ours? I'm still hung up on the inconsistent behavior:

test(0,      NA,       NaN)  # pass
test(0, list(NA), list(NaN)) # fail