rbindlist segfault for fill=TRUE and usenames=FALSE

[ben-schwen]

Closes #5444

• fix
• tests
• update news

[jangorecki]

News is not needed as this was in dev only

[ben-schwen]

I changed the behavior of test 2003.5 introduced in #5263 since for me it seems more natural if the data.table gains the column name of the table with the most columns and not a new one.

This is no breaking change since fill=TRUE with usenames=FALSE is only available in current dev version 1.14.3

[codecov]

Codecov Report

Merging #5468 (c7c5ce2) into master (2f67531) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #5468   +/-   ##
=======================================
  Coverage   99.51%   99.51%           
=======================================
  Files          78       80    +2     
  Lines       14756    14772   +16     
=======================================
+ Hits        14684    14700   +16     
  Misses         72       72           

Impacted Files	Coverage Δ
src/fread.c	99.41% <ø> (ø)
src/init.c	100.00% <ø> (ø)
R/notin.R	100.00% <100.00%> (ø)
src/negate.c	100.00% <100.00%> (ø)
src/rbindlist.c	100.00% <100.00%> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[jangorecki]

I think the names of columns in such case shoybe based on the order of occurrence. Unit test for that is pretty simple, doesn't cover for example case of 3 tables. But the results of the current test looks good (names are based on the order of occurrence of columns), better than before.

[MichaelChirico]

I changed the behavior of test 2003.5 introduced in #5263 since for me it seems more natural if the data.table gains the column name of the table with the most columns and not a new one.

This is no breaking change since fill=TRUE with usenames=FALSE is only available in current dev version 1.14.3

now that there's new behavior, I think that does warrant a NEWS edit (either adding to the existing item or getting its own)