nomatch=0 fires incorrect (?) warning

[kodonnell]

Consider the following

> dt <- data.table(a=1:4, b=LETTERS[1:4])
> setkey(dt, a, b)
> dt[.(1:2, LETTERS[1:3]), nomatch=NA]
   a b
1: 1 A
2: 2 B
3: 1 C
Warning message:
In as.data.table.list(i) :
  Item 1 is of size 2 but maximum size is 3 (recycled leaving a remainder of 1 items)

This is a reasonable warning - as we can see, the first column has been recycled. However:

> dt[.(1:2, LETTERS[1:3]), nomatch=0]
   a b
1: 1 A
2: 2 B
Warning message:
In as.data.table.list(i) :
  Item 1 is of size 2 but maximum size is 3 (recycled leaving a remainder of 1 items)

I don't think this is reasonable - since we've specified nomatch=0 there should be no issue. This is a toy example, but the use case is to return all rows which contain a value in some candidate vector (here 1:2 and LETTERS[1:3]) i.e. a %in% 1:2 & b %in% LETTERS[1:3].

As a separate aside dt[.(1:2, LETTERS[1:4]), nomatch=0] doesn't throw an error, because (I guess) it deems that recycling is expected when lengths are multiples (as per normal R behaviour). However, it means the appearance of this bug depends on the length of the candidate vectors I provide (specifically, whether the smaller is a divisor of the other), which is not intuitive for this use case: if I know column a only contains 1:2 then I should be able to specify 1:2 or 1:12345 as my candidate vector, and get the same result. This isn't the case here.

Again - it makes sense with nomatch=NA, but not nomatch=0. I'm not sure how easy it is to fix this ... I suspect it's just internal reordering of code so that nomatch=0 gets implemented before passing to as.data.table.list.

[franknarf1]

This is a toy example, but the use case is to return all rows which contain a value in some candidate vector (here 1:2 and LETTERS[1:3]) i.e. a %in% 1:2 & b %in% LETTERS[1:3].

No, that is not what the join is meant to do. A join x[i] will take each row of i and look up corresponding rows in x. While it is possible to pass i as a list instead of a table, this is just a syntactical convenience and doesn't imply doing a different join than you'd get if you actually did

m = as.data.table(list(1:2, LETTERS[1:3]))
dt[m, nomatch=0]

Currently, there is no join/index/"bmerge" functionality for dt[a %in% 1:2 & b %in% LETTERS[1:3]], but Arun has filed it: #1453 (if I understand correctly).

[kodonnell]

A join x[i] will take each row of i and look up corresponding rows in x

But in the case of nomatch=0L, the resulting table should not involve any recycling, and hence the warning message should not appear. Correct?

[jangorecki]

Warning comes not from nomatch=0 but from i argument being evaluated. If you just copy content of i and replace .() into data.table() you will get the same warning.

> library(data.table)
data.table 1.11.5 IN DEVELOPMENT built 2018-09-10 04:40:14 UTC; jan  Latest news: r-datatable.com
> dt <- data.table(a=1:4, b=LETTERS[1:4])
> setkey(dt, a, b)
> data.table(1:2, LETTERS[1:3])
   V1 V2
1:  1  A
2:  2  B
3:  1  C
Warning message:
In data.table(1:2, LETTERS[1:3]) :
  Item 1 is of size 2 but maximum size is 3 (recycled leaving remainder of 1 items)

Thus recycling does not occur in resulting table but in input table that you want to join to. Closing then.