diff(factor) by group causes STRING_ELT error

[MichaelChirico]

Reproducing for me on latest dev. I guess diff(factor) is not meaningful and should be an error, but it's going through and causing an error later down the line.

set.seed(32940)
NN = 7e5
KK = 4e4
TT = 25

DT = data.table(
  id = sample(KK, NN, TRUE),
  tt = sample(TT, NN, TRUE),
  ff = factor(sample(3, NN, TRUE))
)
DT[ , diff(ff), by = id]

Error in rbindlist(l, use.names, fill, idcol) :
STRING_ELT() can only be applied to a 'character vector', not a 'NULL'

traceback() shows it's coming from print:

6: rbindlist(l, use.names, fill, idcol) at data.table.R#2673
5: data.table::.rbind.data.table(...)
4: rbind(deparse.level, ...)
3: rbind(head(x, topn), tail(x, topn)) at print.data.table.R#53
2: print.data.table(x)
1: function (x, ...) 
   UseMethod("print")(x)

[MichaelChirico]

Actually I think this is an R bug:

diff(factor(1:3))

Error in as.character.factor(x) : malformed factor

x = diff(factor(1:3))
dput(x)
# structure(c(1L, 1L), class = "factor")

Warning message:
In str.default(val) : 'object' does not have valid levels()

(i.e. it's only an error to try and print() this broken object)

Whereas

x = factor(1:3)
x[3] - x[1]
# [1] NA

Warning message:
In Ops.factor(x[3], x[1]) : ‘-’ not meaningful for factors

Will try and file a fix with R-devel...