Regression in R 3.4

[t-kalinowski]

Hi,

I recently expanded the CI tests for the keras R package and I noticed that R CMD check would fail due to vignette chunks being evaluated that shouldn't be evaluated. It seems that knitr::opts_chunk$set(eval = FALSE) isn't respected in the latest few versions of knitr if R <= 3.4. Installing knitr version 1.30 makes my tests pass again.

[yihui]

Could you read and follow the issue guide exactly to provide the required information? Otherwise it is usually hard for us to help you. In your case, it's difficult to know what the problem could be without a reproducible example. If it is impractical to provide a reproducible example, it will be helpful to let us know which exact version broke your package (e.g., you said 1.30 works, and does that mean 1.31 broke it?). Thanks!

[t-kalinowski]

Hi, apologies for not fleshing this out fully. I made a skeleton R package with some GHA workflows that demonstrate the issue:
You can see the failed run here: https://github.com/t-kalinowski/knitr.regression/runs/3735193979
relevant output, present in R 3.4 but not 3.5+:

── R CMD check results ───────────────────────────── knitr.regression 0.1.0 ────
Duration: 25.1s

❯ checking running R code from vignettes ...
     ‘a_vignette.Rmd’ using ‘UTF-8’ ... failed
   ERROR
  Errors in running code in vignettes:
  when running code in ‘a_vignette.Rmd’
    ...
  
  > knitr::opts_chunk$set(eval = FALSE)
  
  > stop("I shouldn't run!")
  
    When sourcing ‘a_vignette.R’:
  Error: I shouldn't run!
  Execution halted

1 error ✖ | 0 warnings ✔ | 0 notes ✔
Error: Error: R CMD check found ERRORs

[t-kalinowski]

and does that mean 1.31 broke it?

Yes

[yihui]

Hmm... That's weird. I looked at the changes in 1.31 but didn't see a possible change that could have led to the breakage: v1.30...v1.31 I'll investigate further. Thanks a lot for the minimal reproducible example! I really appreciate it.

[yihui]

Identified the offending commit: d35590d but still need to understand why the problem only occurs with R 3.4.4. @cderv Do you have R 3.4.4 to test with?

[yihui]

@cderv Please ignore this issue. I've figured it out.

@t-kalinowski It's fixed now. Unfortunately I just made a CRAN release this morning, so I can't make another one in a short time (may need to wait for at least a month for the next CRAN release). Thanks for your patience!

[halldc]

Hi @yihui - I'm encountering this same issue on R 4.1.1 when running rcmdcheck::rcmdcheck().

[yihui]

@halldc Could you please provide a reproducible example? Thanks!

[halldc]

@yihui - building on @t-kalinowski's excellent reprex: https://github.com/halldc/knitr.regression/runs/3737702441

[halldc]

@yihui - I suspect this is related to #2036 interacting with {rcmdcheck}.

[yihui]

@halldc You are correct---the check in Github action didn't use the --as-cran flag. Unfortunately I was wrong in #2036 (review) that it should be rare to run R CMD check without --as-cran.

For now, the workaround is to either use --as-cran (https://github.com/yihui/knitr.regression/commit/0af676ac41cd5f29994dc5d114eb2f825c618337), or set _R_CHECK_VIGNETTES_SKIP_RUN_MAYBE_ to TRUE like this: https://github.com/rstudio/keras/pull/1275/files.

I'll fix this problem in the next version of knitr.

[halldc]

Thanks @yihui! Yep, I reached the same conclusion and I'm using the second option for now. 🙂

FYI - I'm not using --as-cran because the package is an internal package at a company. Many of the CRAN-specific checks are inappropriate in this setting (e.g. warnings about unreachable intranet URLs).

[yihui]

I have submitted a new version of knitr to CRAN, which is pending manual inspection right now (because CRAN requires at least 7 days between two releases, otherwise a CRAN maintainer will have to manually check it). Hopefully this problem will go away in a day or two. Apologies for the trouble, and thanks for your patience!

[halldc]

Thanks for the quick resolution, @yihui ! 🎉

[yihui]

You are welcome! The new version was accepted on CRAN about 7 hours ago.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue by following the issue guide (https://yihui.org/issue/), and link to this old issue if necessary.