add `cols` argument to `rbindlist`

[MLopez-Ibanez]

x <- list(list(A=1,B="a",C=c(1,2)), list(A=2,B="b",C=c(3,4), D=0.1))
rbindlist(x, fill=TRUE)[, .(A, B,D)]

It would be very useful to be able to select which names or columns are selected from the input of rbindlist() before the data.table is created. Selecting those columns after it is created may lead to a different behavior, as the example above shows.

I'd like to do:

x <- list(list(A=1,B="a",C=c(1,2)), list(A=2,B="b",C=c(3,4), D=0.1))
rbindlist(x, cols=c("A","B","D"), fill=TRUE)

[ben-schwen]

So if I'm getting this right, you are proposing a select parameter, similar to the one already existing in fread?

In your example your desired behavior is the same as unique(rbindlist(x, fill=TRUE)[, .(A, B,D)]), right?

I'm also not sure how we would implement this, since the recycling already happens before we rbindlist with e.g. dt1 = data.table(A=1,B="a",C=c(1,2)) and dt2 = data.table(A=2,B="b",C=c(3,4), D=0.1)

[MLopez-Ibanez]

So if I'm getting this right, you are proposing a select parameter, similar to the one already existing in fread?

Sure, I don't mind the name.

In your example your desired behavior is the same as unique(rbindlist(x, fill=TRUE)[, .(A, B,D)]), right?

No, it is not the same. The first rbindlist() will return a data.table with 4 rows due to C being removed after the data.table is created. The second one will only have 2 rows due to C being filtered out before.

I'm also not sure how we would implement this, since the recycling already happens before we rbindlist with e.g. dt1 = data.table(A=1,B="a",C=c(1,2)) and dt2 = data.table(A=2,B="b",C=c(3,4), D=0.1)

If the input is a list of data.frames or data.tables, then there may not be much difference between selecting before or after building the output data.table, unless the input tables have lots of columns and we are only interested in a few. However, for lists of lists it does make a huge difference.