How should non-reactive functions handle errors to be compatible with shiny?

validate_non_reactives.R

## How should we write non-reactive (business logic) functions so that they 
## are maximally compatible with shiny/reactive processing?

## Here's a function which performs business logic. It may be used at the 
## console or it may be used in a shiny app.
## A shiny app will probably provide the right type of input so the 
## input validations are perhaps not as critical, but a typical design 
## is to stop() if a critical condition is not met during calculations.
var_top_cars <- function(d, minval) {
  stopifnot("mpg not found" = utils::hasName(d, "mpg"))
  
  d <- dplyr::filter(d, mpg >= minval)  
  stopifnot("no rows remaining, can't calculate hp variance" = nrow(d) > 0)
  
  var(d$hp)
}

## this can be used at the console
var_top_cars(mtcars, 10) # works
var_top_cars(mtcars, 30) # works
var_top_cars(mtcars, 40) # error, can't calculate
var_top_cars(iris, 40) # error, bad input

library(shiny)

## this ui is common to all examples below
ui <- fluidPage(
  titlePanel("hp Variance"),
  sidebarLayout(
    sidebarPanel(
      sliderInput("min_mpg", "Minimum MPG:", min = 0, max = 50, value = 25)
    ),
    mainPanel(
      textOutput("variance")
    )
  )
)

## this app works because renderText() catches the error
server_simple <- function(input, output) {
  output$variance <- renderText({
    paste0("variance of top cars' hp = ", req(var_top_cars(mtcars, req(input$min_mpg))))
  })
}
runApp(list(ui = ui, server = server_simple))

## this app *crashes* when the slider goes too high 
## because the stop() occurs within reactive processing
## despite being in non-reactive code
server_complex <- function(input, output) {
  
  out <- reactiveValues(v = NA)
  
  res <- reactive({
    req(var_top_cars(mtcars, req(input$min_mpg)))
  })
  
  observeEvent(req(input$min_mpg), {
    out$v <- req(res())
  })
  
  output$variance <- renderText({
    paste0("variance of top cars' hp = ", out$v)
  })
}
runApp(list(ui = ui, server = server_complex))

## this non-reactive function has some shiny protection
## validate throws a classed error and shiny knows what to do with it
var_top_cars_shiny <- function(d, minval) {
  validate(need(utils::hasName(d, "mpg"), "mpg not found"))
  
  d <- dplyr::filter(d, mpg >= minval)  
  validate(need(nrow(d) > 0, "no rows remaining, can't calculate hp variance"))
  
  var(d$hp)
}

## but it works in a non-reactive context just fine
var_top_cars_shiny(mtcars, 10) # works
var_top_cars_shiny(mtcars, 30) # works
var_top_cars_shiny(mtcars, 40) # error, can't calculate
var_top_cars_shiny(iris, 40) # error, bad input

## this app still works, as expected
server_simple_validate <- function(input, output) {
  output$variance <- renderText({
    paste0("variance of top cars' hp = ", req(var_top_cars_shiny(mtcars, req(input$min_mpg))))
  })
}
runApp(list(ui = ui, server = server_simple_validate))

## now this app _does_ work (doesn't crash)
## admittedly, the output does not update if 
## the slider goes too high
server_complex_validate <- function(input, output) {
  
  out <- reactiveValues(v = NA)
  
  res <- reactive({
    req(var_top_cars_shiny(mtcars, req(input$min_mpg)))
  })
  
  observeEvent(req(input$min_mpg), {
    out$v <- req(res())
  })
  
  output$variance <- renderText({
    paste0("variance of top cars' hp = ", out$v)
  })
}
runApp(list(ui = ui, server = server_complex_validate))

## another option would be to return(NULL) rather than stop
var_top_cars_ret <- function(d, minval) {
  if (!utils::hasName(d, "mpg")) {
    warning("mpg not found")
    return(NULL)
  }
  
  d <- dplyr::filter(d, mpg >= minval)  
  if (!nrow(d) > 0) {
    warning("no rows remaining, can't calculate hp variance")
    return(NULL)
  }
  
  var(d$hp)
}

## this of course works in a non-reactive context 
var_top_cars_ret(mtcars, 10) # works
var_top_cars_ret(mtcars, 30) # works
var_top_cars_ret(mtcars, 40) # returns NULL
var_top_cars_ret(iris, 40) # returns NULL

## this app works, but the value may be empty
## and the warning is at the developer's console
server_simple_null <- function(input, output) {
  output$variance <- renderText({
    paste0("variance of top cars' hp = ", req(var_top_cars_ret(mtcars, req(input$min_mpg))))
  })
}
runApp(list(ui = ui, server = server_simple_null))

## this app _does_ work (doesn't crash)
## again, the output does not update if 
## the slider goes too high
server_complex_null <- function(input, output) {
  
  out <- reactiveValues(v = NA)
  
  res <- reactive({
    req(var_top_cars_ret(mtcars, req(input$min_mpg)))
  })
  
  observeEvent(req(input$min_mpg), {
    out$v <- req(res())
  })
  
  output$variance <- renderText({
    paste0("variance of top cars' hp = ", out$v)
  })
}
runApp(list(ui = ui, server = server_complex_null))