GHA CI possible?

[florianhartig]

Hi @MaximilianPi - do you think it's even thinkable to implement CI via GHA for this? Seems a potential nightmare due to the dependencies, but it would of course be great to check automatically.

[MaximilianPi]

Well, we cannot compile (test) the whole book, but I see a few options:

1. test if we can install all dependencies/libraries on all systems.
2. render the book's rmarkdown files individually (a lot of overhead and we cannot easily pinpoint errors)
3. extract all chunks and nest them automatically in testthat expressions for accurate logging, chunk names can be used as ID (requires metaprogramming but it should be easy to implement).

Personally, I'm in favour of option 3

[MaximilianPi]

I have implemented the 3. option:

• All chunks will be tested within a testthat::test_that("", {testthat::expect_error({ chunk }, NA)} closure
• Chunks will be skipped if purl=FALSE in the chunk options ('''{r, purl=FALSE}...)
• Name of chunks will be used as identifier
• Specify Rmds to be tested here: https://github.com/TheoreticalEcology/machinelearning/blob/Changes_from_Matthias/setup_tests.R
• Tests can be locally run via source("setup_tests.R")

To do:

• Name all chunks! (Well we/I should have done this in the first place...)
• As we cannot test all chunks (e.g. the computationally expensive chunks such as the data augmentation, CNNs, GANs, etc.), go through the markdowns and set purl=FALSE if chunk cannot be tested