Entrace base errors?

[lionel-]

This is a follow-up to #229 and how dev rlang always shows a backtrace on error in non-interactive sessions. It might be helpful and more consistent to entrace base errors with rlang::entrace(). Unfortunately, simply setting options(error = rlang::entrace) doesn't work because rmarkdown::render() captures the errors before entrace() has a chance to be called.

Would it be worth it to find a solution to that problem? The easiest way might be to add an option to entrace errors in rmarkdown itself.

[lionel-]

Proof of concept at fix-top-trace...add-base-entrace:

remotes::install_github("tidyverse/reprex@add-base-entrace")

When prompted, upgrade knitr and evaluate.

• r-lib/evaluate needs support for calling handlers: r-lib/evaluate@master...lionel-:add-calling-handlers
• yihui/knitr needs support for the calling.handlers chunk option: yihui/knitr@master...lionel-:add-calling-handlers

[hadley]

We should also provide some optional way to activate this for warnings too.

[lionel-]

options(warn = 2) should work. But won't that create misleading reprexes if we enable it by default?

[lionel-]

Or should we capture all warnings, with an accompanying trace?

[lionel-]

r-lib/rlang#714

[jennybc]

@lionel- Has the world changed since we last discussed this? What's the best way to support entrancing base errors?

[lionel-]

Adding entrace as a calling handler. That's also the correct way to entrace rlang errors. Cf the weird dplyr example I've just posted in the other thread.