[R] Discussion: should we use {rlang} for condition handling?

[asfimport]

Currently we have a mixture of base::stop() and rlang::abort() in the codebase. Should we make a deliberate choice to use one or the other?
If we decided to switch to {}rlang{}, we could:

• change the codebase as we go along, switching from base functions ({}stop(), warning(), message(){} to their rlang equivalents abort(), warn(), inform()

• make an addendum to STYLE.md on condition handling

  We could cover additional aspects of condition handling such as the wording of messages - for example, by using key words such as must or {}should{}.

  Please discuss direction and why. :)

  Note: the arrow R package already depends on rlang so any outcome will not result in additional dependencies.

  Update: given @paleolimbot's suggestion of the abort(glue)) pattern I think we should include in scope adding a dependency on and using {}glue{}. 

Reporter: Dragoș Moldovan-Grünfeld / @dragosmg

Related issues:

• [R] More special handling for known errors in arrow_eval (is related to)

_{Note: This issue was originally created as ARROW-15359. Please see the migration documentation for further details.}

[asfimport]

Dragoș Moldovan-Grünfeld / @dragosmg:
Advantages (from my point of view):

• rlang allows for classed conditions, which would enable finer grain condition handling
• consistency from a user's perspective
• a closer formatting of condition messages to the tidyverse ones
• adopting (guidelines around) key words (such as {}must{}, {}should{}) would reduce cognitive load for developers and improve messaging consistency

[asfimport]

Dewey Dunnington / @paleolimbot:
Thanks for bringing this up! I've found that using rlang::abort() with glue::glue() has made the error messages in the R package I maintain much better since it makes it easy to provide useful output. As you've pointed out, rlang is already a namespace dependency and so is glue (through vctrs). In addition to your proposed update, I think we should add that abort(glue() (without the rlang:: or glue:: should be the preferred way to signal an error if additional information is required (as opposed to paste(), paste0(), or sprintf()).

A note that this is how errors are signalled everywhere in the tidyverse where those dependencies are allowed (e.g., dplyr, ggplot2).

[github-actions]

This issue has been marked as stale because it has had no activity in the past 365 days. Please remove the stale label or comment below, or this issue will be closed in 14 days. If this improvement is still desired but has no current owner, please add the 'Status: needs champion' label.