Confusing behavior with DT[, min(var):max(var)]

[franknarf1]

With this data...

library(data.table)

DT = data.table(
    id = c(1L, 1L, 1L, 2L, 2L, 2L, 3L, 3L, 3L, 4L, 4L, 4L, 5L, 5L, 5L, 6L, 6L, 6L), 
    t = c(4L, 9L, 14L, 1L, 4L, 9L, 5L, 14L, 18L, 1L, 3L, 12L, 3L, 5L, 7L, 4L, 10L, 13L)
)


    id  t
 1:  1  4
 2:  1  9
 3:  1 14
 4:  2  1
 5:  2  4
 6:  2  9
 7:  3  5
 8:  3 14
 9:  3 18
10:  4  1
11:  4  3
12:  4 12
13:  5  3
14:  5  5
15:  5  7
16:  6  4
17:  6 10
18:  6 13

I expected DT[, min(t):max(t)] to be the same as with(DT, min(t):max(t)) -- just a vector of integers counting over that interval. I got something quite different:

>     DT[, min(id):max(id)]
    id
 1:  1
 2:  1
 3:  1
 4:  2
 5:  2
 6:  2
 7:  3
 8:  3
 9:  3
10:  4
11:  4
12:  4
13:  5
14:  5
15:  5
16:  6
17:  6
18:  6
>     DT[, seq(min(id), max(id))]
[1] 1 2 3 4 5 6
>     
> 
>     DT[, min(t):max(t)]
     t
 1:  4
 2:  9
 3: 14
 4:  1
 5:  4
 6:  9
 7:  5
 8: 14
 9: 18
10:  1
11:  3
12: 12
13:  3
14:  5
15:  7
16:  4
17: 10
18: 13
>     DT[, seq(min(t), max(t))]
 [1]  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17

Maybe this is a side-effect of the move to make DT[, j] behave more like DF[, j]? I didn't check to see if it also happened in older versions.

[renkun-ken]

This behavior looks quite inconsistent with data.table's evaluation rules.

[MichaelChirico]

Seems like a bug to me.

In particular, note:

DT[, dput(min(id):max(id))]
# 1:6
# [1] 1 2 3 4 5 6

Indeed it's something funky with scoping rules, as this fixes things to work as expected:

DT[, (min(id):max(id))]
# [1] 1 2 3 4 5 6
DT[, (min(t):max(t))]
# [1]  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18

But it certainly seems like an unnatural workaround...

[MichaelChirico]

Here's the culprit:

#root here is :
root = if (is.call(jsub)) as.character(jsub[[1L]])[1L] else ""
if (root == ":" || ...) with=FALSE

The logic isn't robust enough to allow for a:b as a seq(a, b) construct -- it's automatically interpreted as cola:colb.

I'm not sure if there's a simple way to fix the logic to accommodate this and still set with = FALSE automatically?

And anyway, I still have no idea why/how the output is what it is, given that DT[ , 1:6] is an error.

[franknarf1]

@MichaelChirico I'm guessing it's somehow becoming DT[, id:id]. I can't figure out why though...

This produces the same result:

DT[, sum(t):paste(t)]

and

DT[, sum(t):paste(t):max(t):min(t)]

[renkun-ken]

What about only symbol:symbol be interpreted as column selection, and others (e.g. call:call) should be regarded as a typical j?