Missing warning message when coercing double to integer64

[hcirellu]

In general the warning message when coercing double values that are to high exists. But for some values it does not appear consistently.

as.integer64(c(9223372036854776000))
# integer64
# [1] <NA>

[hcirellu]

Result after fix would be:

as.integer64(c(9223372036854776000))
# integer64
# [1] <NA>
# Warning message:
# In as.integer64.double(c(9223372036854775808)) :
#   NAs introduced by coercion to integer64 range

[truecluster]

Interesting case: i <- as.integer64(d) inside the precision of integer64 but outside the unique integer-representation of doubles: could be solved in three ways

1. return NA and give a warning about NA creation
2. return one of the possible integer64 values for which as.double(i) == d and give a warning like 'returned one of multiple possible integer64 values'
3. return one of the possible integer64 values for which as.double(i) == d and give no warning

I tend to believe that 2. is better than 1., not sure about 3. Options 2. and 3. would be converting to long double, then rounding, then converting to integer64. Should work with 80 bit long doubles (64 bit mantissa) and longer long doubles.

Thoughts?

Kind regards

Jens

[hcirellu]

I prefer the first option. The other two seem to me like gambling (or indeterministic, even though they probably are deterministic), because there are more possible values. If it can't be 'precise' coerced, one has to take a look at the problem and solve it specificly for it's own use case (e.g. rounding, converting to character). And when NA is returned, the problem is detected easily and one is motivated to handle it problem specificly.
With options 2 and 3 there would be the danger, that the 'precision problem' keeps undetected and in the end the final results of an analysis are totaly wrong, becaus a ' best guess' integer64 value is returnd.

Kind regards

Christian