CJ now auto-names its input

[MichaelChirico]

Closes #1596,
Closes #2192
Related #2081

• technically a breaking change since I'm sure there's code out there relying on CJ(a, b) being named V1,V2 (there was in our own testing, after all).
• no new tests since this breaks a few old tests, I think the behavior is covered by these existing tests.
• minor unrelated change: if (n < 1L) return( null.data.table() ) moved earlier

Hope is that this name_dots function can be used for fixing this weird autonaming issue as well (but that fix is a bit messier):

data.table(a = 1:5, b = 6:10)[ , if (.BY$a > 2) .(b), by = a]
# vs
data.table(a = 1:5, b = 6:10)[ , .(b), by = a]

[jangorecki]

I would postpone this to 1.12

[codecov]

Codecov Report

Merging #2977 into master will increase coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2977      +/-   ##
==========================================
+ Coverage   90.82%   90.84%   +0.01%     
==========================================
  Files          61       61              
  Lines       11736    11738       +2     
==========================================
+ Hits        10659    10663       +4     
+ Misses       1077     1075       -2

Impacted Files	Coverage Δ
R/setkey.R	83.11% <100%> (-0.08%)	⬇️
R/utils.R	87.09% <100%> (+10.62%)	⬆️
R/data.table.R	91.66% <100%> (+0.05%)	⬆️

---

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 040f534...8583e5b. Read the comment docs.

[mattdowle]

It could be put in now with an option to turn on new behaviour (default off for now). We're getting used to this option shenanigans now, so it's not really much more effort. Then change default to on in 1.12.0 and the 0.5% of folk that break on 1.12.0 can revert until they update. "Do no harm", right?

[MichaelChirico]

thanks for following up, haven't found time to sit down for this these few weeks +1