Prevent object serialization in slide error messages

[brookslogan]

When I try outputting an unnamed, length-1 list result from a slide computation, I get an unintelligible, incomplete error message.

library(dplyr)
library(purrr)
library(epiprocess)
library(epipredict)
library(ggplot2)

# Mondays
forecast_dates <- c(as.Date("2020-11-02"))

horizons <- c(7L) # relative to forecast_date

example_forecaster <- function(snapshot_edf, forecast_date) {
  shared_reporting_latency <- forecast_date - max(snapshot_edf$time_value)
  aheads <- horizons + shared_reporting_latency # relative to max time_value
  aheads %>%
    map(function(ahead) {
      snapshot_edf %>%
        arx_forecaster(
          outcome = "case_rate_7d_av", # via JHU-CSSE
          predictors = "percent_cli", # from doctor-visits (claims)
          args_list = arx_args_list(
            ahead = ahead,
            quantile_levels = c(0.25, 0.5, 0.75),
            forecast_date = forecast_date
          ))
    }) %>%
    list()
}

pseudoprospective_forecasts <-
  archive_cases_dv_subset %>%
  epix_slide(
    ref_time_values = forecast_dates,
    before = 365000L, # 1000-year time window --> don't filter out any `time_value`s
    ~ example_forecaster(.x, .ref_time_value),
    as_list_col = TRUE
  ) %>%
  rename(forecast_date = time_value)
})

#> Error: Assertion on 'list(list(structure(list(predictions = structure(list(geo_value = c("ca", "fl", "ny", "tx"), .pred = c(13.9733229746794, 20.1862784685272, 21.6914779432801, 18.7492407321225), .pred_distn = structure(list(structure(list(values = c("25%" = 8.81690382631066, "50%" = 13.9733229746794, "75%" = 19.1297421230482), quantile_levels = c(0.25, 0.5, 0.75)), class = c("dist_quantiles", "dist_default", "vctrs_rcrd", "vctrs_vctr")), structure(list(values = c("25%" = 15.0298593201585, "50%" = 20.1862784685272, 
#> "75%" = 25.342697616896), quantile_levels = c(0.25, 0.5, 0.75)), class = c("dist_quantiles", "dist_default", "vctrs_rcrd", "vctrs_vctr")), structure(list(values = c("25%" = 16.5350587949114, "50%" = 21.6914779432801, "75%" = 26.8478970916489), quantile_levels = c(0.25, 0.5, 0.75)), class = c("dist_quantiles", "dist_default", "vctrs_rcrd", "vctrs_vctr")), structure(list(values = c("25%" = 13.5928215837537, "50%" = 18.7492407321225, "75%" = 23.9056598804913), quantile_levels

[In interactive mode the actual description of what's wrong seems to be cut off? Trying in reprex instead seems to output a whole lot more of the structure; I didn't look to see if it also cuts off the important part.]

The "problem" we're catching here is that an unnamed, length-1 list isn't atomic. (Aside: I think we could probably accept these, and maybe even unnamed lists of other lengths [though it might require some implementation tweaks, not just validation changes]. Named lists are likely a mistake or would need extra processing, though.) This was previously the job of

              if (! (is.atomic(comp_value) || is.data.frame(comp_value))) {
                Abort("The slide computation must return an atomic vector or a data frame.")
              }

but now is

        assert(check_atomic(comp_value, any.missing = TRUE), 
            check_data_frame(comp_value), combine = "or", .var.name = vname(comp_value))

and checkmate's not giving a great message.

[brookslogan]

I'm guessing the culprit here is vname(comp_value). This is just a local variable rather than an argument, so this is probably triggering the [serialization/reverse-parsing] capabilities of deparse within vname.

(function() {
  comp_value = list(letters)
  assert(check_atomic(comp_value, any.missing = TRUE), 
  check_data_frame(comp_value), combine = "or", .var.name = vname(comp_value))
})()
#> Error: Assertion on 'list(c("a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z"))' failed: One of the following must apply:
#>  * check_atomic(list(c("a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l",
#>  * "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z"))): Must be
#>  * of type 'atomic', not 'list'
#>  * check_data_frame(list(c("a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k",
#>  * "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z"))):
#>  * Must be of type 'data.frame', not 'list'.

We can improve this a bit with a custom name

(function() {
  comp_value = list(letters)
  assert(check_atomic(comp_value, any.missing = TRUE), 
  check_data_frame(comp_value), combine = "or", .var.name = "slide computation output")
})()
#> Error: Assertion on 'slide computation output' failed: One of the following must apply:
#>  * check_atomic(slide computation output): Must be of type 'atomic', not 'list'
#>  * check_data_frame(slide computation output): Must be of type 'data.frame', not
#>  * 'list'.

though it still looks a little awkward, and might not look as good if we try to include some context about where the error occurred (see #351 point 2).

@dshemetov any ideas here to make error message output a bit friendlier?

[dshemetov]

So seems like we need to add check_list() in for one of the accepted types here? As for naming, your "slide computation output" seems like a good approach, but I'll see if I have any better ideas when I make a PR for this.

[brookslogan]

Accepting [unnamed] lists [& rejecting named lists] would probably need some more effort verifying/tweaking things to work. If we want to just fix the error message we could focus on that.

[brookslogan]

Sorry, I forgot that you were working on this. Probably going to clobber your work with my eliminate-slide-annoyances pass.

[dshemetov]

I haven't really worked on this, so not a problem.