Assigning NA_real_ to multiple rows

[MichaelChirico]

It appears recycling is failing for NA_real_ here:

DT = data.table(a = c(NA, NA))
DT[ , a := NA_real_]

However this works:

DT = data.table(a = c(NA, NA))
DT[ , a := rep(NA_real_, 2L)]

[ben-schwen]

I guess it actually recycles it, but seems to use memrecycle and memrecycle does not warn about coercing NA to other values.

By supplying the full values you fully overwrite the column. set behaves here the same.

dt = data.table(a = c(TRUE, TRUE))
dt[, a := NA_real_]
dt
#         a
#    <lgcl>
# 1:     NA
# 2:     NA

dt = data.table(a = c(TRUE, TRUE))
dt[, a := 0L]
dt
#         a
#    <lgcl>
# 1:  FALSE
# 2:  FALSE

[Mukulyadav2004]

Hi @MichaelChirico @ben-schwen
I tried removing !ISNAN(val) from CHECK_RANGE condition which prevents NA_real_ from triggering warnings.

However, I discovered that the vector assignment path (rep(NA_real_, 2L)) uses BODY macro instead of CHECK_RANGE, so it never warns at all. This also creates an inconsistency-

DT[, a := NA_real_] giving warning but
DT[, a := rep(NA_real_, 2L)]doesn't

So from my understanding, single value assignment goes through CHECK_RANGE which have warning logic but vector assignment which goes through BODY doesn't have warnings