Transmit R session's LANG to the child process for the test runner

[jennybc]

Companion to posit-dev/ark#270.

Addresses #1808 (comment).

How to test:

Pre-requisite: you will need to be on macOS to truly test this. (Windows is actually (!!) unaffected by this problem, but I have still tested these 2 PRs together on Window and all seems well.)

Get the source of this R package: https://github.com/jennybc/localetest.
Note this is a public package, but I don't explain its purpose anywhere. It just looks like an experiment.
Perhaps using usethis::create_from_github("jennybc/localetest").
This package contains exactly 1 snapshot test, specific for macOS, the only affected OS.

Run the test suite via devtools::test() in the Console or via Cmd + Shift + T (basically use any method other than Positron's test explorer).
The test should pass.

Now run the test suite via Positron's test explorer (accessible via the flask icon).
With a Positron dev build from this PR (along with the amalthea PR), the test should also pass.

With a Positron release build, the snapshot test should fail and will look something like this (i.e. the LANG env var is undefined and several locale categories are "C" instead of "en_US.UTF-8".

Snapshot of code has changed (variant 'darwin'):
old[1:11] vs new[1:11]
  Code
    Sys.getenv("LANG")
  Output
-   [1] "en_US.UTF-8"
+   [1] ""
  Code
    Sys.getlocale()
  Output
-   [1] "C/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8"
+   [1] "C"
  Code
    Sys.getlocale("LC_COLLATE")
and 1 more ...

old[13:23] vs new[13:23]
  Code
    Sys.getlocale("LC_CTYPE")
  Output
-   [1] "en_US.UTF-8"
+   [1] "C"
  Code
    Sys.getlocale("LC_MONETARY")
  Output
-   [1] "en_US.UTF-8"
+   [1] "C"
  Code
    Sys.getlocale("LC_NUMERIC")
and 1 more ...

     old                        | new                            
[25] Code                       | Code                       [25]
[26]   Sys.getlocale("LC_TIME") |   Sys.getlocale("LC_TIME") [26]
[27] Output                     | Output                     [27]
[28]   [1] "en_US.UTF-8"        -   [1] "C"                  [28]

* Run `testthat::snapshot_accept('darwin/locale')` to accept the change.
* Run `testthat::snapshot_review('darwin/locale')` to interactively review the change.

[jennybc]

Typical value for locale on macOS;

{
  LANG: "en_US.UTF-8",
  LC_ALL: "en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8",
  LC_COLLATE: "en_US.UTF-8",
  LC_CTYPE: "en_US.UTF-8",
  LC_MONETARY: "en_US.UTF-8",
  LC_NUMERIC: "C",
  LC_TIME: "en_US.UTF-8",
  LC_MESSAGES: "en_US.UTF-8",
  LC_PAPER: "",
  LC_MEASUREMENT: "",
}

[jennybc]

And here's how it looks in my Windows VM:

2024-03-19 14:17:47.351 [info] Locale info from active R session: {
  "LANG": "en_US.UTF-8",
  "LC_ALL": "LC_COLLATE=English_United States.utf8;LC_CTYPE=English_United States.utf8;LC_MONETARY=English_United States.utf8;LC_NUMERIC=C;LC_TIME=English_United States.utf8",
  "LC_COLLATE": "English_United States.utf8",
  "LC_CTYPE": "English_United States.utf8",
  "LC_MONETARY": "English_United States.utf8",
  "LC_NUMERIC": "C",
  "LC_TIME": "English_United States.utf8",
  "LC_MESSAGES": "",
  "LC_PAPER": "",
  "LC_MEASUREMENT": ""
}

[juliasilge]

Yep, here is what I see:

2024-03-20 13:15:59.975 [info] Locale info from active R session: {
  "LANG": "en_US.UTF-8",
  "LC_ALL": "en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8",
  "LC_COLLATE": "en_US.UTF-8",
  "LC_CTYPE": "en_US.UTF-8",
  "LC_MONETARY": "en_US.UTF-8",
  "LC_NUMERIC": "C",
  "LC_TIME": "en_US.UTF-8",
  "LC_MESSAGES": "en_US.UTF-8",
  "LC_PAPER": "",
  "LC_MEASUREMENT": ""
}

[jennybc]

I originally thought I would set individual components, such as LC_CTYPE and LC_TIME, but reading about related child_process.spawn() + locale issues in VS Code and extensions suggested that LANG is the best place to intervene. It appears to lead to correct values of other aspects of the locale and strikes me as the cleanest solution.

[jennybc]

Recording more background here in case we need to revisit this and also so I can close my browser tabs.

Here's a precise description of the problem. The info below applies specifically to the environment when running testthat tests. It's not exactly what you get with Sys.getlocale(), because testthat, for example, sets LC_COLLATE to "C".

category	macOS Positron R Console & Terminal	macOS Positron child_process	Windows Positron R Console & Terminal	Windows child_process
LC_COLLATE	C	C	C	C
LC_CTYPE	en_US.UTF-8	C	English_United States.utf8	English_United States.utf8
LC_MONETARY	en_US.UTF-8	C	English_United States.utf8	English_United States.utf8
LC_NUMERIC	C	C	C	C
LC_TIME	en_US.UTF-8	C	English_United States.utf8	English_United States.utf8

---

Background on "Internationalization Variables" (see section 8.2 especially) for POSIX.1-2017:

https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html#tag_08

Good overview of LANG and more detailed categories, such as LC_COLLATE and friends.

---

Example of a past issue around child processes and LANG in VS Code itself or in extensions:

microsoft/vscode#85675

That issue had a fix applied, which seemed to cause other troubles, and was eventually reverted. My main take-away is that we aren't the first to encounter locale troubles with child_process.spawn() and what I'm doing is reasonable.

[juliasilge]

When I use the current release build, I see devtools::test() pass and the test explorer fail. ✅

I installed both this PR and the amalthea PR, and then I see both devtools::test() and the test explorer pass. ✅