[Help] Targets does not report correct error message when using tidyverse

[Jiefei-Wang]

Help

• I understand and agree to https://books.ropensci.org/targets/help.html.

Description

This is a weird bug. When I have an error in my code(e.g. variable does not exist), tar_make will give a very misleading message. This only happens when I use tidyverse. I created a toy example to reproduce this. The true error is due to selecting a non-existent column in a data.frame using select(y), but the Last error message is irrelevant.

Reproducible example here. Target version 1.8.0

▶ dispatched target test
▶ recorded workspace test
✖ errored target test    
✖ errored pipeline [0.765 seconds]

Error:
! targets::tar_make() error

── Debug target test ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────    tar_meta(test)$error
    tar_workspace(test)

── General debugging ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────    • tar_errored()
    • tar_meta(fields = any_of("error"), complete_only = TRUE)
    • tar_workspace()
    • tar_workspaces()

── How to ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────    • Debug: https://books.ropensci.org/targets/debugging.html
    • Help: https://books.ropensci.org/targets/help.html

── Last error message ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────    
Internal error: `cnd$subscript_type` must be `logical`, `numeric`, or `character`.

── Last error traceback ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────    long_runnint_task()
    select(dt, y)
    select.data.frame(dt, y)
    tidyselect::eval_select(expr(c(...)), data = .data, error_call = error_c...
    eval_select_impl(data, names(data), as_quosure(expr, env), include = inc...
    with_subscript_errors(out <- vars_select_eval(vars, expr, strict = stric...
    try_fetch(expr, vctrs_error_subscript = function(cnd) {     cnd$subscrip...
    withCallingHandlers(expr, condition = function(cnd) {     {         .__h...
    vars_select_eval(vars, expr, strict = strict, data = x, name_spec = name...
    walk_data_tree(expr, data_mask, context_mask)
    eval_c(expr, data_mask, context_mask)
    reduce_sels(node, data_mask, context_mask, init = init)
    walk_data_tree(new, data_mask, context_mask)
    as_indices_sel_impl(out, vars = vars, strict = strict, data = data,     ...
    as_indices_impl(x, vars, call = call, arg = arg, strict = strict)
    chr_as_locations(x, vars, call = call, arg = arg)
    vctrs::vec_as_location(x, n = length(vars), names = vars, call = call,  ...
    (function ()  stop_subscript_oob(i = i, subscript_type = subscript_type,...
    stop_subscript_oob(i = i, subscript_type = subscript_type, names = names...
    stop_subscript(class = "vctrs_error_subscript_oob", i = i, subscript_typ...
    abort(class = c(class, "vctrs_error_subscript"), i = i, ...,      call =...
    signal_abort(cnd, .file)
    signalCondition(cnd)
    (function (cnd)  {     {         .__handler_frame__. <- TRUE         .__...
    handlers[[1L]](cnd)
    cnd_signal(cnd)
    signal_abort(cnd)
    signalCondition(cnd)
    (function (condition)  {     state$error <- build_message(condition)    ...

tar_meta gives correct error message, but in my real project it is not reliable and can be empty

> tar_meta(fields = any_of("error"), complete_only = TRUE) 
# A tibble: 1 × 2
  name  error
  <chr> <chr>
1 test  Cant subset columns that dont exist. Column y doesnt exist.

The progress file in the meta folder is good. it is always reliable in my real project,

# meta/progress
name|type|parent|branches|progress
test|stem|test|0|dispatched
test|stem|test|0|errored

tar_traceback looks good, it is always reliable in my real project

> tar_traceback(test)
 [1] "long_runnint_task()"
 [2] "select(dt, y)"
 [3] "select.data.frame(dt, y)"
 [4] "tidyselect::eval_select(expr(c(...)), data = .data, error_call = error_c..."
 [5] "eval_select_impl(data, names(data), as_quosure(expr, env), include = inc..."  
 [6] "with_subscript_errors(out <- vars_select_eval(vars, expr, strict = stric..."
 [7] "try_fetch(expr, vctrs_error_subscript = function(cnd) {     cnd$subscrip..."
 [8] "withCallingHandlers(expr, condition = function(cnd) {     {         .__h..."
 [9] "vars_select_eval(vars, expr, strict = strict, data = x, name_spec = name..."  
[10] "walk_data_tree(expr, data_mask, context_mask)"
[11] "eval_c(expr, data_mask, context_mask)"
[12] "reduce_sels(node, data_mask, context_mask, init = init)"
[13] "walk_data_tree(new, data_mask, context_mask)"
[14] "as_indices_sel_impl(out, vars = vars, strict = strict, data = data,     ..."
[15] "as_indices_impl(x, vars, call = call, arg = arg, strict = strict)"
[16] "chr_as_locations(x, vars, call = call, arg = arg)"
[17] "vctrs::vec_as_location(x, n = length(vars), names = vars, call = call,  ..."  
[18] "(function ()  stop_subscript_oob(i = i, subscript_type = subscript_type,..."
[19] "stop_subscript_oob(i = i, subscript_type = subscript_type, names = names..."
[20] "stop_subscript(class = \"vctrs_error_subscript_oob\", i = i, subscript_typ..."
[21] "abort(class = c(class, \"vctrs_error_subscript\"), i = i, ...,      call =..."
[22] "signal_abort(cnd, .file)"
[23] "signalCondition(cnd)"
[24] "(function (cnd)  {     {         .__handler_frame__. <- TRUE         .__..."
[25] "handlers[[1L]](cnd)"
[26] "cnd_signal(cnd)"
[27] "signal_abort(cnd)"
[28] "signalCondition(cnd)"                                                         
[29] "(function (condition)  {     state$error <- build_message(condition)    ..."

In my real project, the Last error traceback from tar_make is not correct, I have to use tar_traceback instead, so I think there might be some other tidyverse functions that mess up with the traceback.

[wlandau]

Thanks, the reprex helped. Turns out the error comes from a problematic interaction between rlang and vctrs:

rlang::inform(message = "x", class = "vctrs_error_subscript_oob")
#> Error in `cnd_subscript_type()`:
#> ! Internal error: `cnd$subscript_type` must be `logical`, `numeric`, or `character`.

I wrote a workaround in b8781d2.