[Bug]: fixed `variables()` choices + multi-`datasets()` breaks `merge_srv` after dataset switch

[m7pr]

What happened?

When a picks() spec allows switching datasets() (e.g. ADSL vs ADLB) but variables() uses a fixed character vector of column names that exist on one table only (e.g. PARAM, AVAL on ADLB), switching to the other dataset produces invalid merge code.

Evaluation then returns a qenv.error, and a later join_keys<- call surfaces as no applicable method for 'join_keys<-' ... 'qenv.error', which is misleading because the root failure is earlier (bad select / merge pipeline).

Reproduce with

library(shiny)
library(teal.data)
library(teal.picks)

merge_reprex_data <- function() {
  data <- teal_data()
  data <- within(data, {
    ADSL <- data.frame(
      USUBJID = sprintf("S%03d", 1:8),
      AGE = sample(35:70, 8, replace = TRUE),
      stringsAsFactors = FALSE
    )
    ADLB <- data.frame(
      USUBJID = rep(sprintf("S%03d", 1:8), each = 3),
      PARAM = rep(c("ALT", "AST", "BILI"), 8),
      AVAL = round(rnorm(24, 42, 6), 1),
      stringsAsFactors = FALSE
    )
  })
  join_keys(data) <- join_keys(teal.data::join_key("ADSL", "ADLB", keys = "USUBJID"))
  data
}

merge_reprex_picks_broken <- function() {
  picks(
    datasets(choices = c("ADSL", "ADLB"), selected = "ADLB"),
    variables(
      choices = c("PARAM", "AVAL"),
      selected = c("PARAM", "AVAL"),
      multiple = TRUE
    )
  )
}

merge_reprex_picks_working <- function() {
  picks(
    datasets(choices = c("ADSL", "ADLB"), selected = "ADLB"),
    variables(
      choices = tidyselect::everything(),
      selected = c(1L, 2L),
      multiple = TRUE
    )
  )
}

merge_reprex_make_app <- function(picks, title) {
  cached_data <- merge_reprex_data()
  data_r <- reactive({
    cached_data
  })

  ui <- fluidPage(
    titlePanel(title),
    fluidRow(
      column(4, picks_ui("sel", picks = picks)),
      column(
        8,
        tags$h4("Mapped"),
        verbatimTextOutput("mapped"),
        tags$h4("Merge preview (error text or str of anl)"),
        verbatimTextOutput("merged")
      )
    )
  )

  server <- function(input, output, session) {
    selectors <- list(sel = picks_srv("sel", picks = picks, data = data_r))
    merged <- merge_srv("merge", data = data_r, selectors = selectors)

    output$mapped <- renderPrint({
      yaml::as.yaml(merged$variables())
    })

    output$merged <- renderPrint({
      shiny::req(merged$data())
      td <- merged$data()
      if (inherits(td, "qenv.error")) {
        return(conditionMessage(td))
      }
      str(td[["anl"]], max.level = 1L)
    })
  }

  shinyApp(ui, server)
}

run_merge_reprex_broken <- function() {
  shiny::runApp(
    merge_reprex_make_app(
      merge_reprex_picks_broken(),
      "Broken: variables(PARAM, AVAL) — switch dataset to ADSL"
    ),
    launch.browser = TRUE
  )
}

run_merge_reprex_working <- function() {
  shiny::runApp(
    merge_reprex_make_app(
      merge_reprex_picks_working(),
      "Working: tidyselect::everything(), selected = c(1L, 2L)"
    ),
    launch.browser = TRUE
  )
}
run_merge_reprex_broken()   # opens app: switch dataset ADLB → ADSL → merge preview errors
run_merge_reprex_working()  # opens app: same UI pattern, delayed variables — OK when switching

Relevant log output

No response

Code of Conduct

• I agree to follow this project's Code of Conduct.

Contribution Guidelines

• I agree to follow this project's Contribution Guidelines.

Security Policy

• I agree to follow this project's Security Policy.

[averissimo]

What would be the expected behavior?

[llrs-roche]

imho: Unresolved picks so no further merge code or code evaluation should happen till that is resolved

[m7pr]

Maybe a message that columns do not exist in the second dataset, that is displayed in the UI. So a warning/error?

[gogonzo]

Since picks_srv returns a list of reactiveVal (one per picks) I don't know any solution where error in one pick can be propagated further. We can for example return object of class error either on the "picks-level" or "pick-level".

x:
  <error>

or

x:
  datasets: ADSL
  variables: <error>

In this proposition error will be just an element of the list (handled error) so accessing x()$variables won't throw. We would have to check all pick elements against possible errors inside of tm_ module or in merge_srv.

P.S.
If picks were an environment, then it would be much easier, but not so sure if design allows picks to be env:

env <- new.env()
env$a <- 1
delayedAssign("b", stop("`$variables`: PARAMCD, AVAL don't exist in dataset ADSL"), assign.env = env)
env$b   
# Error: `$variables`: PARAMCD, AVAL don't exist in dataset ADSL

---

Maybe assert/check/test_picks function to check if picks are correct. Then in the modules we could:

selectors <- srv_picks(...)
validate(check_picks(selectors$<pick name>()))
# error in `$variables`: PARAMCD, AVAL don't exist in dataset ADSL

Following checkmate guides we can create a checkPicks covering all validation cases we need. For example

#' @param x (`picks`) 
#' @param null.choices.ok (`logical(1)`) Are `null` choices in any of picks element choices allowed 
#' @param null.selected.ok (`logical(1))` Are `null` choices in any of picks element selected allowed
checkPicks <- function(x, null.choices.ok = FALSE, null.selected.ok = TRUE, ...)

[osenan]

Hi, the example app of this module in progress of migration has this issue. Certainly the UX could be improved, but the app does not break.

The "Row faceting variable" I switched datasets with character vector defined choices and then the similar issue described in here appears.