knitr capture reports every chunk error as rlang_error (knitr wraps errors via rlang::entrace)

Open
#1,129 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
55/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
r, rust
Domain
backend, testing

Research direction

Start with crates/quarto-core/src/engine/knitr/resources/rmd/capture.R:150 and the failing test crates/quarto-core/src/engine/knitr/capture_tests.rs:170; trace how the captured condition supplies its class and error details. Run the reproduction command with R, knitr, rmarkdown, jsonlite, and rlang installed. Done means captured errors retain the original class, with tests covering ordinary and custom errors; the issue also leaves the desired error text and R-enabled CI coverage for discussion.

Written by the indexing model from the issue text.

Description

bug testing

Summary

When rlang is installed, knitr runs every chunk error through rlang::entrace(). That function wraps the original error in a new condition of class rlang_error. Our knitr capture code (crates/quarto-core/src/engine/knitr/resources/rmd/capture.R:150) records ename = class(p)[[1]], so every captured error is reported as rlang_error. The user's actual class, such as simpleError or a custom class, is lost.

Since rlang is installed in almost every real R setup, this affects most users.

Reproduction

With R, knitr, rmarkdown, jsonlite and rlang installed:

cargo nextest run -p quarto-core --lib -E 'test(knitr::capture_tests::capture_classifies_pieces)'
panicked at crates/quarto-core/src/engine/knitr/capture_tests.rs:170:13:
assertion `left == right` failed
  left: "rlang_error"
 right: "simpleError"

Fails with both:

  • R 4.6.0, knitr 1.51, rlang 1.2.0, evaluate 1.0.5
  • R 4.6.0, knitr 1.52, rlang 1.3.0, evaluate 1.0.5 (latest CRAN as of 2026-11-09)

Cause

knitr:::knit_handlers appends rlang_entrace_handler (list(error = function(e) rlang::entrace(e))) to the chunk's calling_handlers whenever knitr:::has_rlang is TRUE. Reproduced with evaluate alone:

h <- evaluate::new_output_handler(
  calling_handlers = list(error = function(e) rlang::entrace(e)))
e <- evaluate::evaluate('stop("boom")', output_handler = h)[[2]]
class(e)        # "rlang_error" "error" "condition"
class(e$error)  # "simpleError" "error" "condition"

A custom error class is also replaced: with stop(structure(class = c("myError", "error", "condition"), ...)), class(e) is rlang_error and class(e$error) is myError.

Suggested fix

In capture.R, when p inherits from rlang_error and has an $error field, read ename from class(p$error)[[1]]. Otherwise fall back to class(p)[[1]]. Then decide whether evalue and call should also come from p$error.

A related question: knitr also changes the error text to rlang's format (Error:\n! boom instead of Error in …: boom). The test only checks starts_with("Error"), so it still passes, but we should decide whether that text is what we want to show.

Why CI didn't catch this

No CI workflow installs R, and every test in capture_tests.rs skips itself through r_available() when R is missing. These tests therefore only run on developer machines that have R. It's worth adding R (with knitr and rlang) to at least one CI leg.

Found while investigating pandoc 3.13 test failures; this failure is unrelated to pandoc.

Dominant language
Rust
Stars
302
Forks
22
Avg merge
5h 2m
Merged PRs (30d)
213

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from quarto-dev/q2

All issues in quarto-dev/q2

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.