Fixed bug where setattr crashed on character(0) value.

[MarkusBonsch]

Closes #2386 .
setattr(1, "class", character(0)) used to cause a segmentation violation. Now, it throws a proper error.

[codecov-io]

Codecov Report

Merging #2628 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2628      +/-   ##
==========================================
+ Coverage   93.03%   93.03%   +<.01%     
==========================================
  Files          61       61              
  Lines       12115    12116       +1     
==========================================
+ Hits        11271    11272       +1     
  Misses        844      844

Impacted Files	Coverage Δ
R/data.table.R	97.61% <100%> (ø)	⬆️
src/wrappers.c	95.08% <100%> (+0.08%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 01ef8e3...f2b0497. Read the comment docs.

[mattdowle]

Great find. I guess this will fail on any object, not just 1; that 1 was just to illustrate? If so, the side-issue then with 1L and possibly 1 too is that they could be R's internal length-1 globals. Then setattr() could potentially add an attribute to the global value of 1L. Similarly to how data.table could change the global value of TRUE to FALSE several years ago in R-devel before release that was caught early on via tests against R-devel . Will use this PR as a nudge to look into length-1 globals status in R-devel/altrep.

[MarkusBonsch]

@mattdowle Wow, I didn't even know these things existed :). You are completely right, 1 was just to illustrate. I replaced with data.table(x=1:10) in the tests in order to avoid side-effects.