i column not found in rolling join with by

Start by running the reproducible DT[i,j,by,roll] examples in the issue, especially regular[irregular, ...]. Compare the failing mean(Y) case with the working .N grouping and the alternative non-equi join. Done means the rolling join can aggregate Y with by=grid without materializing joined, producing the expected group counts and means.

hi all! I'm not an expert in rolling joins, but I thought it should be possible to use arguments by and roll at the same time to do rolling join and summarization efficiently in one step. But I observed a "not found" error when I tried that.
Here is a complete example:

library(data.table)
X.vec = c(0.05, 0.15, 0.75, 0.95)
Y.vec = 10*X.vec
(irregular = data.table(X=X.vec, Y=Y.vec, key="X"))
grid.space = 0.2
offset = grid.space/2
grid = seq(offset, 1-offset, by=grid.space)
(regular <- data.table(grid, X=grid, min=grid-offset, max=grid+offset, key="X"))
(joined = regular[irregular, roll="nearest"])
joined[, .(N=.N, mean.Y=mean(Y)), by=grid]
irregular[regular, .(grid, N=.N, mean.Y=mean(Y)), on=.(X<max, X>min), by=.EACHI, nomatch=0L]
regular[irregular, .(N=.N), roll="nearest", by=grid]
regular[irregular, .(N=.N, mean.Y=mean(Y)), roll="nearest", by=grid]

The goal is a rolling join between regular and irregular, with matching each irregular$X to the closest regular$grid value, then computing the mean of irregular$Y for each value of regular$grid.

> (irregular = data.table(X=X.vec, Y=Y.vec, key="X"))
      X   Y
1: 0.05 0.5
2: 0.15 1.5
3: 0.75 7.5
4: 0.95 9.5
> (regular <- data.table(grid, X=grid, min=grid-offset, max=grid+offset, key="X"))
   grid   X min max
1:  0.1 0.1 0.0 0.2
2:  0.3 0.3 0.2 0.4
3:  0.5 0.5 0.4 0.6
4:  0.7 0.7 0.6 0.8
5:  0.9 0.9 0.8 1.0

One way to do that is by first materializing the joined table, then summarizing it:

> (joined = regular[irregular, roll="nearest"])
   grid    X min max   Y
1:  0.1 0.05 0.0 0.2 0.5
2:  0.1 0.15 0.0 0.2 1.5
3:  0.7 0.75 0.6 0.8 7.5
4:  0.9 0.95 0.8 1.0 9.5
> joined[, .(N=.N, mean.Y=mean(Y)), by=grid]
   grid N mean.Y
1:  0.1 2    1.0
2:  0.7 1    7.5
3:  0.9 1    9.5

That works fine, but I thought there would be some more efficient way to do that (which could avoid allocating memory for the intermediate joined table). So I tried adding by to the same call as roll but I got an error:

> regular[irregular, .(N=.N, mean.Y=mean(Y)), roll="nearest", by=grid]
Error in `[.data.table`(regular, irregular, .(N = .N, mean.Y = mean(Y)),  : 
  object 'Y' not found

Is this a bug? In other words, when doing DT[i,j,by,roll], shouldn't it be possible to use columns of i in j?
One thing that suggests that it should be possible is the code below, which shows that you can count the number of items in each by group:

> regular[irregular, .(N=.N), roll="nearest", by=grid]
   grid N
1:  0.1 2
2:  0.7 1
3:  0.9 1

Finally, another way of computing the result that I want, in one step (without the intermediate table) would be:

> irregular[regular, .(grid, N=.N, mean.Y=mean(Y)), on=.(X<max, X>min), by=.EACHI, nomatch=0L]
     X   X grid N mean.Y
1: 0.2 0.0  0.1 2    1.0
2: 0.8 0.6  0.7 1    7.5
3: 1.0 0.8  0.9 1    9.5