Add workaround for a std::regex's bug on non-UTF8 MBCS locales

[yutannihilation]

Fix #31

This pull request fixes #31 by wrapping functions that call std::regex with withr::with_locale().

Implementation

First, I investigate which R functions generated by Rcpp use std::regex, and found all of them do. So, it's probably better to fix at the very near location of these functions. But, since they are automatically generated, we cannot modify RcppExports.R. So, there probably are two choices:

1. Write wrapper functions one by one (e.g. pz_parzer_lat_safe() for pz_parzer_lat())
2. Modify function body within R script

While 1. is easy, I'm afraid this is human-error prone (e.g. use the original function without _safe mistakenly, forget to add the wrapper when adding a new Rcpp function, etc). So, I chose option 2 in this pull request. (edit: I got the suggestion to override .Call(). It seems far better than my tricky implementations, so I switched to it)

Testing

The hardest part is how to test this. The problem is, as the nature of this bug, the test case doesn't fail, but hangs forever. So, I think it's safe to call it on an external process, so that we can set timeout.

However, this turned out a bit tricky. To test the function, we need to use pkgload::load_all() (or devtools::load_all()) to load them on the external process. The problem is, on Windows, this invokes recompilation (I don't know why), which makes the tests fail as it doesn't finish within the specified timeout. I tweaked to compile it beforehand with pkgbuild::compile_dll(), but anyway it means we need to compile twice on Windows, so the Windows check is a bit slow (I think this is acceptable).

 ** running tests for arch 'i386' ...

  Running 'testthat.R' [53s]

 OK

** running tests for arch 'x64' ...

  Running 'testthat.R' [68s]

 OK

[yutannihilation]

Thanks to the the power of Twitterverse, a better trick is suggested.

https://twitter.com/hadleywickham/status/1355519666660192259

[sckott]

thanks, having a look

[sckott]

Do we know if Japanese locale is available on the GH Actions Windows builder? https://github.com/ropensci/parzer/pull/32/checks?check_run_id=1797664773

Maybe you can try adding to the gh actions config file to show testthat output

- name: Show testthat output
   if: always()
   run: find check -name 'testthat.Rout*' -exec cat '{}' \; || true
   shell: bash

[yutannihilation]

Yes, it's available. Do you mean I should make GHA fail when the locale is unavailable?

[yutannihilation]

Confirmed no test is skipped.

https://github.com/ropensci/parzer/pull/32/checks?check_run_id=1826691377#step:12:394

[sckott]

Okay, great. And I assume this works for you on your machine?

[yutannihilation]

Yes, it works on my Windows, and I also confirmed this fixes the problem of the original reporter of this issue (for reference, here's the conversation (in Japanese): https://twitter.com/niszet0/status/1357863004847497221).

[sckott]

thanks!

[yutannihilation]

Thanks for merging!