Inconsistencies with NA and other types of 'missingness'

[ateucher]

I noticed that different classes of NA were treated differently on a round trip:

library(V8)
ct <- v8()

## Logical and character of length one return as NULL
ct$assign("na", NA)
ct$get("na")
#> NULL

ct$assign("na_char", NA_character_)
ct$get("na_char")
#> NULL

## Real, integer, and complex return as "NA" string
ct$assign("na_real", NA_real_)
ct$get("na_real")
#> [1] "NA"

ct$assign("na_int", NA_integer_)
ct$get("na_int")
#> [1] "NA"

ct$assign("na_complex", NA_complex_)
ct$get("na_complex")
#> [1] "NA"

## When vector is greater than length 1, get vector of NAs back:
ct$assign("nas", c(NA, NA))
ct$get("nas")
#> [1] NA NA

I then looked at other 'missing' types: NaN and NULL:

## NaN returns as "NaN" string
ct$assign("nan", NaN)
ct$get("nan")
#> [1] "NaN"

## NULL returns as an empty list()
ct$assign("null_", NULL)
ct$get("null_")
#> named list()

I think the inconsistencies in the NA types is a bit of a problem, but not sure what you think about the NaN and NULL situations?

[jeroen]

Yeah the problem is that we currently use json to serialize the data, which does not support NA types. Maybe with the new V8 we can try a binary serialization format.

[ateucher]

Yup, that makes sense for the NaN vs NULL vs NA cases... but I'm still surprised at the differences between NA_character_ and NA_real_ (for example), where the former is comes back as NULL, and the latter comes back as "NA" character string...

EDIT: Actually, this is perhaps better discussed in jsonlite:

# length one
jsonlite::toJSON(NA_character_)
#> [null]
jsonlite::toJSON(NA_real_)
#> ["NA"]

# greater than length one
jsonlite::toJSON(rep(NA_character_, 2))
#> [null,null]
jsonlite::toJSON(rep(NA_real_, 2))
#> ["NA","NA"]

# And coming back the other way:
jsonlite::fromJSON('null')
#> NULL
jsonlite::fromJSON('"NA"')
#> [1] "NA"

jsonlite::fromJSON('[null,null]')
#> [1] NA NA
jsonlite::fromJSON('["NA","NA"]')
#> [1] NA NA

^{Created on 2019-02-12 by the reprex package (v0.2.1)}