Recreate reactiveValues in shiny without reactiveness (but same scoping behaviour)

Question

I have a shiny app split into multiple modules. To pass data between the modules I use reactiveVal(ues). These can be passed to functions and modified from within the other functions.

I would like to create a similar object, but without all the reactive capabilities of the built-in reactiveValues(). I want to pass quite large dataframes/tibbles around and want to manually control if and when datatables/calculations are done.

In this example, I would like to create a object instead of list(), that can pass data around the scopes in the same way as reactiveValues():

library(shiny)

## Module
module_ui <- function(id) {
    ns <- NS(id)
    actionButton(inputId = ns("increase"), label = "Increase")
}

module_server <- function(id, vals, not_working) {
    moduleServer(id, function(input, output, session) {
        observeEvent(input$increase, {
            vals$a <- vals$a + 1
            not_working$a <- not_working$a + 1
        })
    })
}

## Main
ui <- fluidPage(
    h1("ReactiveValues Test"),
    textOutput(outputId = "out1"),
    textOutput(outputId = "out2"),
    actionButton(inputId = "update", label = "Update Shown Values"),

    module_ui("modid")
)

server <- function(input, output, session) {
    vals <- reactiveValues(a = 10)
    not_working <- list(a = 10)

    observeEvent(input$update, {
        output$out1 <- renderText(paste("The value of vals$a is:", vals$a))
        output$out2 <- renderText(paste("The value of not_working$a is:", not_working$a))
    })

    module_server("modid", vals, not_working)
}

shinyApp(ui, server)

In the example above

the problem with reactiveValues is that after the inital "Update Shown Values" is pressed, the vals$a value is updated instantly when "increase"-button is pressed and does not require the manual "Update Shown Values". Because I have large amounts of data and do some calculation I don't want it to trigger automatically, and I don't want to be forced to only read and modify the value inside reactive environments.
the problem with list is that the object gets copied into the module_server-call and this the value is not updated in the main function scope.

I have tried to read the source code of reactives.R but it is too advanced R for me to understand what is happening.

I have tried to find a way to pass by reference, but not found one yet. But I think passing a list() by reference might work. In python passing a dictionary would behave in the way I want. Where key/value-pairs can be modified inside function scopes.

How could I go about created an object which scoping behaves as reactiveValues, but without all of the reactiveness-functionality? Are there any good resources describing what is happening in the reactives.R source code? (or not that code specifically but the language-feature used)

This sounds like it might be an [XY problem](https://en.wikipedia.org/wiki/XY_problem) to me. Can you explain why `isolate`ing the code you don't want to trigger updates won't work for you? — Limey, Jul 04 '23 at 17:49
I think it is mostly because I don't fully understand it. In the code above, my assumption was that `observeEvent(input$update, {})` would only trigger from `input$update`, but the text `[...] of vals$a is: 10` updates instantly. Guessing `renderText` gets an expr and this is what causes the re-render when the value changes? I want to control when things update, especially big tables, and then I want the default behaviour to be "isolate". I thought it was easier to create an object with only the parts I want, instead of working against the reactive-system and typing `isolate` everywhere. — Elias, Jul 05 '23 at 09:03
It's your choice. But I do honestly think that investing the time in *understanding* will pay benefits in both the long and medium term. Also, I notice that your module server does not have a return value and nothing in your main server reactis to changes in the module. That seems strange, but I don't full grasp what you are trying to do. — Limey, Jul 05 '23 at 09:50

Elias · Accepted Answer · 2023-07-05T09:04:59.827

After some further digging and experimentation, I found that the underlying object that was used was a fastmap. If we wrap it in some S3 object notation we can use it as a replacement for reactiveValues() without the reactiveness.

This code is mostly copied from reactives.R. All credit goes there.

# nolint start: object_name_linter.
dataHolder <- function(...) {
  args <- rlang::list2(...)
  if ((length(args) > 0) && (is.null(names(args)) || any(names(args) == "")))
    rlang::abort("All arguments passed to dataHolder() must be named.")

  values <- structure(
    list(
      impl = fastmap::fastmap()
    ),
    class = "dataholder"
  )

  lapply(names(args),
         \(name) {
           .subset2(values, "impl")$set(name, args[[name]])
         })

  values
}

checkName <- function(x) {
  if (!is.character(x) || length(x) != 1) {
    rlang::abort("Must use single string to index into dataholder.")
  }
}

print.dataholder <- function(x, ...) {
  cat("<DataHolder>", "\n")
  cat("  Values:   ", paste0(.subset2(x, "impl")$keys(sort = TRUE), collapse = ", "), "\n")
}

is.dataholder <- function(x) inherits(x, "dataholder")

`$.dataholder` <- function(x, name) {
  checkName(name)
  .subset2(x, "impl")$get(name)
}

`[[.dataholder` <- `$.dataholder`

`$<-.dataholder` <- function(x, name, value) {
  checkName(name)
  .subset2(x, "impl")$set(name, value)
  x
}

`[[<-.dataholder` <- `$<-.dataholder`

`[.dataholder` <- function(values, name) {
  rlang::abort("Can't index dataholder with `[`.")
}

`[<-.dataholder` <- function(values, name, value) {
  rlang::abort("Can't index dataholder with `[`.")
}
# nolint end

Recreate reactiveValues in shiny without reactiveness (but same scoping behaviour)

1 Answers1