Shiny selectInput depending on validated reactive does not pass on validation error

Question

I'm trying to use validate in R shiny to prevent outputs from showing when they shouldn't be. When a validated reactive is used in creating an input inside a uiOutput and the validation throws an error, this error does not get passed on to a depending reactive or output.

Reading https://shiny.rstudio.com/articles/validation.html and using reactlog has me thinking that the issue lies in the generated input not depending on the validated reactive. The article states:

Shiny will: [...] pass a validation error to any reactive expression or observer object that depends on it

I'm unsure what an input inside a uiOutput really is, but I suspect it's neither a reactive expression nor an observer.

Consider the following example:

library(shiny)
library(data.table)

cars <- data.table(mtcars, keep.rownames = T )
setnames(cars, "rn", "name")

ui <- fluidPage(
  selectInput("cyl", "Cyl", c(4, 12)),
  uiOutput("uiOutCars"),
  h4("Filtered Table"),
  tableOutput("filteredTable")
)

server <- function(input, output, session) {

  availableCars <- reactive({
    choices <- cars[cyl == input$cyl, name]
    validate(need(try(length(choices) > 0),
                  "No cars with this cyl!"))
    choices

  })


  output$uiOutCars <- renderUI({
    selectInput(inputId = "selectCar", 
                label = "Car", 
                choices = availableCars())
  })


  output$filteredTable <- renderTable({
    cars[name == input$selectCar]
  })


}

shinyApp(ui, server)

Initial state:

When changing the cyl input from 4 to 12 (no cars with 12 cylinders exist in cars), the selectCar input is not shown. The validation message No cars with this cyl! is:

My expectation was that the filteredTable also stops showing, because input$selectCar should not have a proper value. However, it seems to retain its last value when availableCars() did not yet throw the error.

Is it possible to pass on the validation error "through" an input generated by a uiOutput?

EDIT

@TimTeaFan adds that neither does it work when using updateSelectInput (thanks for checking!):

library(shiny)
library(data.table)

cars <- data.table(mtcars, keep.rownames = T )
setnames(cars, "rn", "name")

ui <- fluidPage(
    selectInput("cyl", "Cyl", c(4, 
    selectInput("selectCar", "Car", 
                cars[cyl == 4, name]), # differs from original example here
    h4("Filtered Table"),
    tableOutput("filteredTable")
)

server <- function(input, output, session) {

    availableCars <- reactive({
        choices <- cars[cyl == input$cyl, name]
        validate(need(length(choices) > 0,
                      "No cars with this cyl!"))
        choices

    })

    # differs from original example here
    observeEvent(input$cyl, {
        updateSelectInput(session, "selectCar",
                          choices = availableCars())
    })


    output$filteredTable <- renderTable({
        cars[name == input$selectCar]
    })


}

shinyApp(ui, server)

Though, if you start the app with selecting `cyl=12`, then the table doesn't load, as expected, until you switch the selection. So the input (and table) reacts once, and will continue to react until `availableCars()` takes the value of the validation error message. So the input is sort of reactive, but not like other reactive expressions... — phalteman, Aug 01 '19 at 23:51
That seems to be the case, yes. Do you have an idea if I'm using an anti-pattern here by trying to validate the table this way? — bendae, Aug 02 '19 at 07:30
To be honest, I don't know. I suspect that `validate()` doesn't really change the value of its dependencies but instead just passes an error message and stops downstream calculations, which is why `input$selectCar` isn't updating. But I don't know enough to dig around to confirm that's true. I'll provide a workaround below, though I recognize that's not necessarily what you're really after. — phalteman, Aug 02 '19 at 16:05

score 2 · Answer 1 · answered Aug 02 '19 at 16:11

You can get the behaviour you're looking for by replacing your renderTable() with the following:

  output$filteredTable <- renderTable({
    req(availableCars())
    cars[name %in% input$selectCar]
  })

As noted in the comments, my suspicion is that when the validation error gets passed on, it stops any downstream calculations and therefore input$selectCar doesn't get updated and the table continues to show the last selection. Using req() gets the app to check for the current value of availableCars() which then passes the validation message and updates the table.

I know this doesn't answer your question exactly, but hopefully it provides a useful workaround.

Thanks for the workaround, it works as described. For simple cases it should be good enough. It seems to me that it has some drawbacks for more complicated ones, though. If I have more than one reactive/output depending on `input$selectCar`, it will get harder and harder to update all the `req` once you introduce changes to what `input$selectCar` is depending on (e.g. rename `availableCars`, etc). I'll keep the question open in case someone chimes in. — bendae, Aug 05 '19 at 09:55

score 2 · Accepted Answer · answered Aug 13 '19 at 14:12

For completeness, per my comment on the other answer, strictly speaking you cannot have a validation message propagate to an input (in uiOutput or otherwise). As quoted from the article, validation propagates to reactive expressions and observers. In the language of reactivity (see this article), these are reactive conductors and reactive endpoints, respectively. This means they can take on a reactive dependency. Reactive conductors can also be used to create a reactive dependency. input is a reactive source. Reactive sources can be used to create a reactive dependency, but strictly speaking can't take on a reactive dependency (technically renderUI is actually making output$uiOutCars have the reactive dependency not input$selectCar)

All this being said, here is another workaround. It creates a pseudo-input which is a reactive expression (i.e. will have validation messages propagate through).

library(shiny)
library(data.table)

cars <- data.table(mtcars, keep.rownames = T )
setnames(cars, "rn", "name")

ui <- fluidPage(
  selectInput("cyl", "Cyl", c(4, 12)),
  uiOutput("uiOutCars"),
  h4("Filtered Table"),
  tableOutput("filteredTable")
)

server <- function(input, output, session) {

  availableCars <- reactive({
    choices <- cars[cyl == input$cyl, name]
    validate(need(try(length(choices) > 0),
                  "No cars with this cyl!"))
    choices

  })


  output$uiOutCars <- renderUI({
    selectInput(inputId = "selectCar", 
                label = "Car", 
                choices = availableCars(),
                selected = "Merc 230")
  })

  pseudoSelectCar <- reactive(
    {

      availableCars()
      # include other dependencies here
      input$selectCar

    })

  output$filteredTable <- renderTable({
    req(pseudoSelectCar()) # prevent warning on initialization
    cars[name == pseudoSelectCar()]
  })


}

The idea is that as the expression for the the pseudo input is re-evaluated, if any of the dependencies fail validation, they will short-circuit the "input" evaluation as well. Otherwise, the pseudo input is just equal to the input value.

The downside is you do have to explicitly list all your dependencies for input$selectCar in the definition of pseudoSelectCar. The upside is you just have to do this in one place and you can use pseudoSelectCar everywhere else in your code where you would have otherwise used input$selectCar

Thanks for this, great explanation and the most practical workaround yet for more complicated applications. From your description of inputs as reactive sources it seems to me that the answer is **no**, one can't pass on the validation error "through" an `input` generated by a `uiOutput`, which is a shame. I'll keep the question open for another couple of days in case someone chimes in, after which I'll accept your answer. — bendae, Aug 16 '19 at 13:54

score 1 · Answer 3 · answered Aug 09 '19 at 15:22

Do you need to have the error message repeated twice or could you move the h4 and tableOutput into the renderUI?

library(shiny)
library(data.table)

cars <- data.table(mtcars, keep.rownames = T )
setnames(cars, "rn", "name")

ui <- fluidPage(
  selectInput("cyl", "Cyl", c(4, 12)),
  uiOutput("uiOutCars")

)

server <- function(input, output, session) {

  availableCars <- reactive({
    choices <- cars[cyl == input$cyl, name]
    validate(need(try(length(choices) > 0),
                  "No cars with this cyl!"))
    choices

  })


  output$uiOutCars <- renderUI({
    tagList(
      selectInput(inputId = "selectCar", 
                  label = "Car", 
                  choices = availableCars()),
      h4("Filtered Table"),
      tableOutput("filteredTable")
    )
  })


  output$filteredTable <- renderTable({
    cars[name == input$selectCar]
  })


}

shinyApp(ui, server)

Thanks for this workaround! It works nicely for this simple case. However, in more complicated ones I fear it falls short, e.g. if I have a module that returns `reactive(input$selectCar)` and the table is part of another module with this selection as an input. — bendae, Aug 12 '19 at 08:22
@bendae so going back to your original question about passing a validation message to an input, from my assessment the answer is no. I'm adding a different answer with hopefully a slightly more general workaround. — Marcus, Aug 13 '19 at 13:50

Shiny selectInput depending on validated reactive does not pass on validation error

EDIT

3 Answers3

Linked