expect_error() regexpr matching parent errors

[romainfrancois]

As part of doing the rev dep checks for dplyr 1.0.8 tidyverse/dplyr#6080 there are some packages that start to fail because expect_error() does not match the pattern, because the message is now in a parent condition.

library(testthat)
library(rlang)
#> 
#> Attaching package: 'rlang'
#> The following objects are masked from 'package:testthat':
#> 
#>     is_false, is_null, is_true

f <- function() {
  abort("hello")
}
g <- function() {
  withCallingHandlers(f(), error = function(cnd) {
    abort("bonjour", parent = cnd)
  })
}

expect_error(g(), "Hello")
#> Error: `g()` threw an error with unexpected message.
#> Expected match: "Hello"
#> Actual message: "bonjour"
tryCatch(g(), error= print)
#> <error/rlang_error>
#> Error:
#>   bonjour
#> Caused by error in `f()`:
#>   hello
#> Backtrace:
#>  1. base::tryCatch(g(), error = print)
#>  5. global g()
#>  7. global f()

^{Created on 2021-11-17 by the reprex package (v2.0.1.9000)}

[romainfrancois]

For example, in package tbrf:

test_that("tbr_binom returns error", {
  df <- data.frame(
    date = sample(seq(as.Date('2000-01-01'), as.Date('2005/12/30'), by = "day"), 10),
    value = rbinom(10, 1, 0.25)
  )

  expect_error(df %>% tbr_binom(x = value,
                                   tcolumn = date,
                                   unit = "year",
                                   n = 5,
                                   alpha = 0.1),
                  "unit must be one of years, months, weeks, days, hours, minutes, seconds")
})

now being:

[master] 220.85 MB > df <- data.frame(
+     date = sample(seq(as.Date('2000-01-01'), as.Date('2005/12/30'), by = "day"), 10),
+     value = rbinom(10, 1, 0.25)
+ )
[master] 220.85 MB > df %>% tbr_binom(x = value,
+                  tcolumn = date,
+                  unit = "year",
+                  n = 5,
+                  alpha = 0.1)
Error in `mutate()`:
  Problem while computing `temp = purrr::map(...)`.
Caused by error in `tbr_binom_window()`:
  unit must be one of years, months, weeks, days, hours, minutes, seconds
Run `rlang::last_error()` to see where the error occurred.

[romainfrancois]

Another example in srvyr gergness/srvyr#135

── Failure (test_survey_statistics.r:583:13): survey_STATISTIC functions fail with x being a character ──
`summarise(dstrataGrp, survey_ratio(api99, name))` threw an error with unexpected message.
Expected match: "Character vectors not allowed in survey functions, should be used as a grouping variable."
Actual message: "Problem while computing `..1 = survey_ratio(api99, name)`.\nℹ The error occurred in group 1: awards = No."
Backtrace:
     ▆
  1. ├─testthat::expect_error(...) at test_survey_statistics.r:583:12
  2. │ └─testthat:::quasi_capture(...)
  3. │   ├─testthat .capture(...)
  4. │   │ └─base::withCallingHandlers(...)
  5. │   └─rlang::eval_bare(quo_get_expr(.quo), quo_get_env(.quo))
  6. ├─dplyr::summarise(dstrataGrp, survey_ratio(api99, name))
  7. ├─srvyr:::summarise.grouped_svy(...)
  8. │ ├─dplyr::summarise(.data$variables, !!!.dots, .groups = .groups)
  9. │ └─dplyr:::summarise.grouped_df(.data$variables, !!!.dots, .groups = .groups)
 10. │   └─dplyr:::summarise_cols(.data, dplyr_quosures(...), caller_env = caller_env())
 11. │     ├─base::withCallingHandlers(...)
 12. │     └─dplyr:::map(quosures, summarise_eval_one, mask = mask)
 13. │       └─base::lapply(.x, .f, ...)
 14. │         └─dplyr FUN(X[[i]], ...)
 15. │           └─mask$eval_all_summarise(quo)
 16. └─srvyr::survey_ratio(api99, name)
 17.   └─srvyr:::stop_for_factor(denominator)
 18.     └─base::stop(...)

[lionel-]

Since the parent messages are printed on error, it makes sense to match them. I think we should add a inherit argument that defaults to TRUE. When set, parent classes and messages would be matched.

[romainfrancois]

Also Tplyr :

── Failure (Line 8): all test tables can be built without errors or warnings ─────────────────────
`build(t7)` threw an error with unexpected message.
Expected match: "object 'col_i' not found"
Actual message: "Problem while computing `col_i = fct_expand(...)`."
Backtrace:
  1. testthat::expect_error(build(t7), "object 'col_i' not found")
 21. col_i

[hadley]

@lionel- do you want to do this in #1491?

[lionel-]

@hadley I think it would be better in a separate PR. The logic in #1491 will be conditional whereas matching parent conditions should be unconditional in order to preserve backwards compatibility. I'll send a PR tomorrow.

[hadley]

Ok, makes sense. Thanks!