Update error handling docs

[hadley]

Once r-lib/rlang#1533 and r-lib/rlang#1534 are in a released version of rlang.

[lbui30]

No sure where this would go in your task schedule, but this broke in tidymodels.
tidymodels/recipes#1081

P.S. I am going to go back to having to a roll back of every package in tidyverse due to this bugs similar to this one.

[hadley]

@lbui30 as requested in that issue, I think you need to start by providing a reprex.

[EmilHvitfeldt]

With the help of @lbui30, we were able to locate the problem which happened with a modification of an example we had on tidymodels.org. Thus purrr wasn't at fault.

[lbui30]

Thanks @EmilHvitfeldt is correct. However, something about purr or the tidyverse does make the output of errors still bit confusing to figure out. So, I was recommended by a co-worker to bring this up here as a motivating example of the need to have clearer error handling messages.

Errors were so much easier when they referenced code line numbers and had a common sense order in them. If there is anyway to help the reader identify the lines of code better in the new error handling docs, then I think that would be a great contribution.

[hadley]

@lbui30 we have done our best to make the errors informative and actionable, so if you have a counter example, we'd really appreciate seeing a reprex that illustrates it.

[hadley]

I think this issue is mostly about ?purrr::purrr_error_indexed which we also need to display more prominently.

[hadley]

try_fetch() is now less important, and I suspect we're unlikely to move this error to rlang, so I think we can close this.