targets::tar_dir({
targets::tar_script({
library(targets)
deps <- rlang::syms(c("x", "y"))
list(
tar_target(x, 1),
tar_target(y, 2),
eval(
bquote(
tar_target(
name,
list(..(deps)) # splice syntax
),
splice = TRUE
)
)
)
})
targets::tar_visnetwork()
})Since I have learnt about the rlang package, I have been fascinated by the power of metaprogramming in R. The first time I learnt such a programming pattern might date back to when I found the eval() function in writing MATLAB codes. However, this type of programming is really a disaster in MATLAB, for it only accepts a string of code. It is R shows me the interesting part of the world of metaprogramming.
At that time, I knew it is called non-standard evaluation (NSE) in R, and it is just used to save typing so that users do not need to type quotes, and what’s more useful, I can access the data columns without typing the data variable names. Then after reading advanced R, it reveals a new path about metaprogramming in R, although some strange terms also make me confused, e.g., quasiquotation, unquotation, and so on.
What makes me really write codes with metaprogramming paradigm is when I began to use targets package. And one day, I found “static branching” supported by the package, which requires certain metaprogramming. And the author of that package also created targetopia, really requiring some metaprogramming skills to create so-called targets factory functions. So I then resorted to rlang package, trying to have a deeper understanding of metaprogramming in R.
The first days with rlang package were really happy and funny. But the first frustration came when I found expr() is really annoying becasue I treated it as equivalent of substitute(), only also supports !! and !!! injection. After a real long time, I finally knew it is equivalent to bquote(). At the end of the day, I just begin to love base R, especially for metaprogramming. For example, bquote() and substitute() are more intuitive to me.
But today I got to know that bquote() is more power than I thought. It also supports splicing!! Amazing, although only added after R 4.0.0. I write a simple use case for me, which is really special for usage in targets package. As I sometimes nested name for several targets into a list, and I want to make sure targets understand that the dependency is the elements of the list, not the list itself. So the splicing feature of bquote() is really useful for me. Here is a simple example: