Ethan Merritt - 2026-01-12

The difference will go away if you tell the program to treat a coordinate whose value is found to be NaN the same as a coordinate that is missing from the input data.

set datafile missing NaN

The next question is obviously "why does $2*$1/$1 evaluate to NaN while $2+10 apparently does not?". Good question. When reading in a line of data, the program returns an error flag DF_MISSING to indicate that a required coordinate is missing. The test for that condition is trivial if the coordinate is simply the content of a particular column. If the coordinate is the result of a computation, that is harder to detect. I don't know exactly why in this case the program correctly identifies that one of those expressions depends a missing value but fails to recognize the same thing for the other expression.