[R] max and pmax of NA and NaN

[Martin Maechler]

>>>>> Michal Burda <michal.burda at centrum.cz>
>>>>>     on Mon, 15 Jan 2018 12:04:13 +0100 writes:

    > Dear R users, is the following OK?

    >> max(NA, NaN)
    > [1] NA
    >> max(NaN, NA)
    > [1] NA
    >> pmax(NaN, NA)
    > [1] NA
    >> pmax(NA, NaN)
    > [1] NaN

    > ...or is it a bug? 

    > Documentation says that NA has a higher priority over NaN.

which documentation ??
[That would be quite a bit misleading I think. So, it should be amended ...]

    > Best regards, Michal Burda


R's help pages are *THE* reference documentation and they have 
(for a long time, I think) had :

?NaN   has in its 3rd 'Note:'

     Computations involving ‘NaN’ will return ‘NaN’ or perhaps ‘NA’:
     which of those two is not guaranteed and may depend on the R
     platform (since compilers may re-order computations).

Similarly,  ?NA  contains, in its 'Details':

     Numerical computations using ‘NA’ will normally result in ‘NA’: a
     possible exception is where ‘NaN’ is also involved, in which case
     either might result (which may depend on the R platform).  ........

-----

Yes, it is a bit unfortunate that this is platform dependent; if
we wanted to make this entirely consistent (as desired in a
perfect world), I'm almost sure R would become slower because
we'd have to do add some explicit "book keeping" / checking
instead of relying on the underlying C library code.

Note that for these reasons, often NaN and NA should not be
differentiated, and that's reason why using  is.na(*)  is
typically sufficient and "best" -- it gives TRUE for both NA and NaN.


Martin Maechler
ETH Zurich