Losing imported module in functions when adding functions to a class using R6$set

[dereckmezquita]

I’ve run into a problem while trying to combine box with R6 classes.

The issue seems to be tied to how environments are managed when a function from a box module relying on imported objects like constants gets attached to an R6 class. When I call that function from the R6 object, I hit an "object not found" error".

The Problem

Here’s the setup I’m working with:

• I’ve got constants defined in a box module constants.R.
• Another module validators.R imports those constants and uses them in a function.
• An R6 class MyClass.R grabs that function and attaches it as a private method using $set().
  • In my real code I have hundres of lines of code for validator functions which is why I made them a separate module and used $set() to add them to the class, this allows me to use self$ and private$ in the function bodys without having to pass it as an argument.

When I call the method on an R6 object, I get this error:

Error in private$.validate_color(color) : object 'constants' not found

Minimal Reproducible Example

File: constants.R

#' @export
VALID_COLORS <- list(
  RED = "red",
  GREEN = "green",
  BLUE = "blue"
)

File: validators.R

box::use(./constants)

#' @export
.validate_color <- function(color) {
    if (!color %in% constants$VALID_COLORS) {
        stop("Invalid colour: ", color)
    }
}

File: MyClass.R

box::use(R6, ./validators)

#' @export
MyClass <- R6$R6Class(
    "MyClass",
    public = list(
        set_color = function(color) {
            private$.validate_color(color)
            private$.color <- color
        }
    ),
    private = list(
        .color = NULL
    )
)

MyClass$set("private", ".validate_color", validators$.validate_color)

File: main.R

box::use(./MyClass)

obj <- MyClass$MyClass$new()
obj$set_color("red")

When I run Rscript main.R, I get:

Error in private$.validate_color(color) : object 'constants' not found
Calls: <Anonymous> -> <Anonymous> -> %in%
Execution halted

What I Think Might Happening?

After poking around online and reading up on how box and R6 handle environments (like the box docs and R6 vignettes), here’s my take:

• Box Modules: When box::use(./constants) runs in validators.R, it imports VALID_COLORS into the module’s environment. The function .validate_color can access it as constants$VALID_COLORS because it’s defined in that module’s scope.
• R6 Environment Switch: When I attach .validate_color to the R6 class with MyClass$set(), the function gets copied over to the R6 object’s private environment. In R, functions carry their defining environment (via environment()), but R6 seems to reassign it to the object’s private environment? That new environment doesn’t include the box module’s imports, so constants is nowhere to be found when the function runs.

I think I just have to change the way I do imports, I did get a set up similar to this working earlier but I can remember how, I am searching through my commits to try to find it - the truth is I don't really understand how this is working/not working.

R version 4.5.0 (2025-04-11)
Platform: aarch64-apple-darwin20
Running under: macOS Sequoia 15.4.1

Matrix products: default
BLAS:   /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRblas.0.dylib 
LAPACK: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.12.1

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

time zone: America/Chicago
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
 [1] crayon_1.5.3        vctrs_0.6.5         dereck_2.0.1       
 [4] cli_3.6.5           rlang_1.1.6         DBI_1.2.3          
 [7] stringi_1.8.7       generics_0.1.4      jsonlite_2.0.0     
[10] data.table_1.17.0   glue_1.8.0          listenv_0.9.1      
[13] future.apply_1.11.3 scales_1.4.0        grid_4.5.0         
[16] tibble_3.2.1        MASS_7.3-65         lifecycle_1.0.4    
[19] box_1.2.0           stringr_1.5.1       compiler_4.5.0     
[22] codetools_0.2-20    fs_1.6.6            RColorBrewer_1.1-3 
[25] timechange_0.3.0    Rcpp_1.0.14         pkgconfig_2.0.3    
[28] future_1.49.0       farver_2.1.2        digest_0.6.37      
[31] R6_2.6.1            pillar_1.10.2       parallelly_1.44.0  
[34] parallel_4.5.0      magrittr_2.0.3      tools_4.5.0        
[37] gtable_0.3.6        stenographer_1.0.2  globals_0.18.0     
[40] lubridate_1.9.4     ggplot2_3.5.2  

[dereckmezquita]

This is interesting, I very quickly stumbled upon the solution but I don't understand why it works. I had to import constants into where the class is and the functions are attached with $set():

# ./MyClass.R

box::use(R6, ./validators, ./constants)

#' @export
MyClass <- R6$R6Class(
    "MyClass",
    public = list(
        set_color = function(color) {
            private$.validate_color(color)
            private$.color <- color
        }
    ),
    private = list(
        .color = NULL
    )
)

MyClass$set("private", ".validate_color", validators$.validate_color)

[klmr]

R6 seems to reassign it to the object’s private environment?

That’s exactly right. In my opinion this is a seriously questionable design decision of R6, but here we are.

My suggestion would be not to attach named functions to R6 classes. Instead, you can dispatch via an anonymous function, e.g.:

MyClass$set("private", ".validate_color", \(c) validators$.validate_color(c))

That said, your solution works as well. The reason it works is that importing ./constants makes the name constants available in the module environment where MyClass is defined. With that, MyClass$.validate_color will be able to find that name, since the module environment is an enclosing (/parent) environment of the MyClass environment.

[dereckmezquita]

That makes sense. I wonder why R6 made that choice and what the limitations are.

I like your use of an anonymous function but I will stick with the import in my specific case here being that I have about 20-30 helper functions to attach to a class - it is significantly less boiler plate to just import the module.