data.table(key=) doesn't allow whitespaces #4292

Activity

1. Henrik-P commented on Mar 10, 2020

   

   Henrik-P

   on Mar 10, 2020

   Last edited by Henrik-P

   More actions

   Please note that also by also accepts

   a single character string containing comma separated column names

   ...but here the help text is more informative than for key: it's explicit about the possibility for leading or trailing spaces in column names (although I think that would be a terrible habit):

   where spaces are significant since column names may contain spaces even at the start or end

   In addition, I find the error message in by slightly clearer: because the problematic column name is quoted in the error message, it's easier to spot a space here than in the corresponding error message for key, where the name is unquoted.

   as.data.table(mtcars)[ , .N, by = "vs, am"]
   # Error in eval(bysub, x, parent.frame()) : object ' am' not found
   

   I think that the specification of column names using "a single character string containing comma separated column names" should behave similarly in key and by. That said, I wouldn't miss this option to specify columns if it was deprecated.
2. ColeMiller1 commented on Mar 10, 2020

   

   ColeMiller1

   on Mar 10, 2020

   Contributor

   More actions

   I agree with @Henrik-P. A better way may be to export the .() functionality which would take one more character to do what you are proposing.

   .(A, B)
   "A, B"
   

   FWIW, I am going to propose to deprecate the comma delimited feature in the by arg in another PR. I'll make a separate issue for that discussion.
3. jangorecki commented on Mar 10, 2020

   

   jangorecki

   on Mar 10, 2020

   Last edited by jangorecki

   Member

   More actions

   I would not use .(), then we don't need to substitute. And personally I would deprecate scalar string for multiple columns, both in key and in by, of course very slowly.

   👍React with 👍2Reacted by Sindri and Michael Chirico
4. shrektan commented on Mar 10, 2020

   

   shrektan

   on Mar 10, 2020 · Hidden as resolved

   Author

   show commentMore actions
5. ColeMiller1 commented on Mar 11, 2020

   

   ColeMiller1

   on Mar 11, 2020

   Contributor

   More actions

   @shrektan I do not follow. Not sure if there is a question about a column name containing a comma. But some column names do sometimes have white space.

   @jangorecki using substitute is 5 times slower - it takes 2.6 microseconds. In the long run I still see the col_select being able to be used in situations like these.

   Actual benchmark data

   fx_sub = function(x) {
     all.vars(substitute(x))
     }
   
   bench::mark(
     fx_sub(.(A, B)),
     c('A','B'),
     time_unit = 'ns'
   )
   # A tibble: 2 x 13
   ##nano seconds
   ##  expression        min median
   ## <bch:expr>      <dbl>  <dbl>
   ##1 fx_sub(.(A, B)) 2100.  2600.
   ##2 c("A", "B")      400.   500.
6. MichaelChirico

   added

   wontfix

   WAIWorking as intendedWorking as intended

   on Jul 11, 2025
7. MichaelChirico commented on Jul 11, 2025

   

   MichaelChirico

   on Jul 11, 2025

   Member

   More actions

   I would consider key=<string> support in maintenance mode -- we don't really plan to deprecate it, but neither will we do more feature work like this.

   👍React with 👍1Reacted by Jan Gorecki
8. MichaelChirico

   closed this as completedon Jul 11, 2025