More information error when `quarto_run()` fails

[hadley]

I think you can take the same approach as pkgdown does with rmarkdown, i.e. if it fails and it's in quiet mode, you can still print the appropriate part of the stderr that callr has captured.

(It doesn't matter too much here, but in general when you're wrapping errors, you should always use withCallingHandlers not tryCatch)

  path <- withCallingHandlers(
    callr::r_safe(rmarkdown_render_with_seed, args = args, show = !quiet),
    error = function(cnd) wrap_rmarkdown_error(cnd, input_file, call)
  )

wrap_rmarkdown_error <- function(cnd, input, call = caller_env()) {
  lines <- strsplit(gsub("^\r?\n", "", cnd$stderr), "\r?\n")[[1]]
  lines <- lines[nchar(lines) > 0]
  # Feeding random text back into cli, so have to escape
  lines <- gsub("{", "{{", lines, fixed = TRUE)
  lines <- gsub("}", "}}", lines, fixed = TRUE)

  cli::cli_abort(
    c(
      "!" = "Failed to render {.path {input}}.",
      set_names(lines, "x")
    ),
    parent = cnd$parent %||% cnd,
    trace = cnd$parent$trace,
    call = call
  )
}

[cderv]

(It doesn't matter too much here, but in general when you're wrapping errors, you should always use withCallingHandlers not tryCatch)

Thanks for the suggestion. I did switch to withCallingHandler now

I think you can take the same approach as pkgdown does with rmarkdown, i.e. if it fails and it's in quiet mode, you can still print the appropriate part of the stderr that callr has captured.

Do you have some usage where you found lack of information in error ? Is it always when quiet = TRUE ?

I think I may need to change the way processx::run() is used. 🤔 I see in pkgdown you are using callr::r_safe() with show = !quiet - so I understand that even if quiet = TRUE, you'll still have error information from rmarkdown::render() background code, because callr will handle the "show it or not" part.

Here, when quiet = TRUE in quarto_run(), this is setting the --quiet flag in Quarto. And it happens that Quarto will not print any message to stderr at all.

❯ quarto render not-existing --quiet

❯ echo $LASTEXITCODE
1

❯ quarto render not-existing
ERROR: No valid input files passed to render

Stack trace:
    at Command.actionHandler (file:///C:/Users/chris/Documents/DEV_R/quarto-cli/src/command/render/cmd.ts:288:13)
    at eventLoopTick (ext:core/01_core.js:178:7)
    at async Command.execute (https://deno.land/x/cliffy@v1.0.0-rc.3/command/command.ts:1948:7)
    at async Command.parseCommand (https://deno.land/x/cliffy@v1.0.0-rc.3/command/command.ts:1780:14)
    at async quarto (file:///C:/Users/chris/Documents/DEV_R/quarto-cli/src/quarto.ts:193:5)
    at async file:///C:/Users/chris/Documents/DEV_R/quarto-cli/src/quarto.ts:225:5
    at async file:///C:/Users/chris/Documents/DEV_R/quarto-cli/src/core/main.ts:45:14
    at async mainRunner (file:///C:/Users/chris/Documents/DEV_R/quarto-cli/src/core/main.ts:47:5)
    at async file:///C:/Users/chris/Documents/DEV_R/quarto-cli/src/quarto.ts:215:3

❯ echo $LASTEXITCODE
1

Exit code is correctly a failure (1 and not 0), but nothing is shown.

So in the current implementation, I don't think I can show more. I just added already a helper message to warn user about this.

quarto::quarto_render("not-existing", quiet = TRUE)
#> Error in `quarto::quarto_render()`:
#> ✖ Error running quarto cli.
#> ℹ Rerun with `quiet = FALSE` to see the full error message.
#> Caused by error:
#> ! System command 'quarto.exe' failed

So I am thinking that if we want to be more verbose, --quiet should not be passed to Quarto, but rather it should be processx::run(echo = !quiet) 🤔

Am I correct ?

Did you have another case where it does not involve --quiet ?

Thank you

[hadley]

I have forgotten the motivating case, but your analysis sounds correct to me.