setattr crashed when given empty character vector as name or value

[hatal175]

Trying to run setattr(numeric(1), "class", character(0)) or setattr(numeric(1), character(0), "value") crashes because the setattrib c function doesn't check if the value or name string vector actually has length > 0 before accessing it with STRING_ELT.

[franknarf1]

No repro on Win 7:

library(data.table)
setattr(numeric(1), "class", character(0))[]
# [1] 0
setattr(numeric(1), character(0), "value")[]
# Error in setattr(numeric(1), character(0), "value") : 
#   'installTrChar' must be called on a CHARSXP

with

data.table 1.10.5 IN DEVELOPMENT built 2017-09-23 01:51:19 UTC; travis
R version 3.3.3 (2017-03-06)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1

[hatal175]

This managed to crash mine (same for character(0) in the other parameter):
xx = lapply(1:10000, function(x) setattr(numeric(1), "class", character(0)))

R version 3.4.1 (2017-06-30)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 15063)

data.table 1.10.4

[franknarf1]

I guess you should test on the development version before reporting here, per https://github.com/Rdatatable/data.table/wiki/Support

[hatal175]

Trying this with the R-devel version and latest data.table build does not reproduce. But here's the thing, I'm pretty sure this is dependent on what memory the uninitialized data is pointing to. Different builds might do different things. However, the bug I pointed out is still there and it is a simple code flow.

[MarkusBonsch]

I can crash with
xx = lapply(1:10000, function(x) setattr(numeric(1), "class", character(0)))
but not with
xx = lapply(1:10000, function(x) setattr(numeric(1), "bla", character(0)))
and
setattr(numeric(1), character(0), "value")
throws a normal error as expected.

So, the only problem is when using setattr on the class attribute with 0 length value.
This may well have been fixed in the current R-devel, since data.table is just calling the R setAttrib function here. However, I will still fix this issue on data.table side just to be sure.

Cheers,
Markus