Need an easier way to use dynamically determined symbols

[renkun-ken]

Suppose I have a data.table created as follows:

dt <- data.table(x1 = 1:10, x2 = 10:1, x3 = 1:10)

I need to do calculations with dynamically determined symbols within j like

s1 <- "x2"
s2 <- "x3"

Two approaches can do the work:

> dt[, get(s1) * get(s2)]
 [1] 10 18 24 28 30 30 28 24 18 10
> dt[, .SD[[1]] * .SD[[2]], .SDcols = c(s1, s2)]
 [1] 10 18 24 28 30 30 28 24 18 10

But if the data is very big and by= is used, the performance can significantly decay. Also the first approach using get() has scoping problem if s1 or s2 are themselves columns of dt.

Is there any possibility that makes it easier to use dynamically determined symbol without such significant performance decay and scoping problem?

For example, something like

dt[, ..s1 * ..s2]

which is inspired by the ..x notation introduced lately.

[tdeenes]

eval(as.name(s1)) is more efficient. But still, it would be great to generalize the .. notation so that it could replace the eval(as.name()) workaround.

[franknarf1]

I think the .. prefix would make sense, maybe a dupe of #633

[jangorecki]

AFAIK using language objects is most efficient way, and kind of base R way of handling the task. Two examples:
https://stackoverflow.com/a/37408321
https://stackoverflow.com/a/37008966

[jangorecki]

to be addressed by #4304

dt[, s1 * s2, env=list(s1=s1,s2=s2), verbose=TRUE]
#Argument 'j'  after substitute: x2 * x3
#Detected that j uses these columns: x2,x3 
# [1] 10 18 24 28 30 30 28 24 18 10

closing as duplicate of #2655