enable switching to the faster, time-stamp-based modification detection in auto_test() & auto_test_package()

[katrinleinweber]

I haven't bench-marked the difference, just found it neat to have this option in the auto_test() functions, as well.

[katrinleinweber]

Travis says: LaTeX Error: File 'xkeyval.sty' not found. Since there are several other errors in other PRs, I'm not sure whether mine triggered this or not. devtools::check() on my machine does not produce any warnings or errors, but doesn't do "checking PDF version of manual".

[katrinleinweber]

Hello @krlmlr! There are currently no tests for the two functions I modified here. Would the recursion of testing auto_test_package() while testing testthat even be possible? 😉 Either way, I'm not sure how to fulfil the codecov target, sorry.

[krlmlr]

Thanks. I guess "user testing" is fine for this function.

Have you seen my brushthat package? I'd be very much interested in auto_test() functionality there, too!

[katrinleinweber]

On my machine, auto_test_package() against testthat [c7e83308](https://github.com/hadley/testthat/tree/c7e8330867645c174f9a286d00eb0036cea78b0c) yields the same results as my modification auto_test_package(hash = FALSE): some mismatches in the reporters/..txt files.

I also ran the following user test. Is that a valid approach in your view?

• in my macOS: checking /Library/Frameworks/R.framework/Versions/3.4/Resources/library for the largest dir => dplyr (well, 10 MB)
• in Terminal: git clone https://github.com/tidyverse/dplyr.git --depth=1
• in RStudio: uninstalled testthat and reinstalled with devtools::install_github("hadley/testthat")
  • in RStudio: associate new project with cloned dplyr folder & open
  • auto_test_package() => some skipped, none failed
  • back in testthat on my branch: Clean & Rebuild
  • back in dplyr: auto_test_package(hash = FALSE) => some skipped, none failed

This, together with the checks here seems to say: I regress nothing ;-D

I guess that to get more conclusive tests results, one would need to measure the reaction time of watch(hash = TRUE) vs. watch(hash = FALSE) in an even bigger package, but that is out-of-scope for me.

Please just consider this PR as enabling an existing, low-level option in the high-level functions as well.

[katrinleinweber]

Hello @krlmlr!

[…] "user testing" is fine for this function.

Does that mean the codecov/patch check can be ignored and this PR will be merged? Kind regards!

[hadley]

Thanks!

[katrinleinweber]

You're welcome :-)