Two potential issues related to rlang errors and backtraces

[DavisVaughan]

Two rlang snapshot tests are failing with new testthat

══ Failed tests ════════════════════════════════════════════════════════════════
── Failure ('test-c-api.R:1299:3'): re-encoding fails purposefully with any bytes ──
Snapshot of code has changed:
old vs new
  Code
    r_obj_encode_utf8(bytes)
- Condition <rlang_error>
+ Condition <simpleError>
    Error in `r_obj_encode_utf8()`:
    ! translating strings with "bytes" encoding is not allowed

── Failure ('test-cnd-entrace.R:415:3'): can call `global_entrace()` in knitted documents ──
Snapshot of code has changed:
     old                           | new                                        
[11]                               |                                        [11]
[12]       f()                     |       f()                              [12]
[13]                               |                                        [13]
[14]       ## Error in h(do): foo  -       ## Error in `h()`:               [14]
                                   -       ## ! foo                         [15]
                                   -       ## Backtrace:                    [16]
                                   -       ##     x                         [17]
                                   -       ##  1. \-rlang (local) f()       [18]
                                   -       ##  2.   \-rlang (local) g(do)   [19]
                                   -       ##  3.     \-rlang (local) h(do) [20]
[15]                               |                                        [21]
[16]       rlang::global_entrace() |       rlang::global_entrace()          [22]
[17]                               |                                        [23]

@lionel- mentioned that the first might have something to do with you unwrapping rlang errors into base errors? Is that done unconditionally? It seems like if it started as an rlang error then it should not get unwrapped?

The second seems to have something to do with some new backtrace capturing? Ironically I think this was an example where we wanted to force the base R specific error for testing global_entrace()

[DavisVaughan]

False alarm!

translating strings with "bytes" encoding is not allowed

This is a base R error, so should not have been upgraded to a rlang_error

The backtrace change is in a test where rlang is rendering a document with rmarkdown/knitr before and after calling global_entrace() manually. In the "before" case, it used to show a base R error. But now knitr itself calls global_entrace() when you have a new enough rlang and evaluate installed, so both cases show the enhanced backtraces. (yihui/knitr#2388, released Dec 2025)