Some tests are Locale dependent

[radeusgd]

I have a default locale on my PC set to pl_PL, in this locale floating-point numbers are comma-separated instead of being dot-separated as in en_US locale.

At least one of the tests fails because of this, namely tests/run-macros/f-interpolator-tests.scala, because it relies on a check file generated on an en_US locale containing amongst others number 10.00.

Expectation

The tests should work on all locales (for example by forcing sbt to use the supported en_US locale) or at least emit a warning which locale is required for a correct test run.

[radeusgd]

Interestingly, I tried a quick workaround by running export LANG=C before launching sbt, and while in the REPL Locale.getDefault() returns en_US, the test (run by dotty-bootstrapped/testCompilation f-interpolator) still seems to be using my pl_PL locale.

[radeusgd]

The following patch fixes the issue:

diff --git a/compiler/test/dotty/tools/vulpix/RunnerOrchestration.scala b/compiler/test/dotty/tools/vulpix/RunnerOrchestration.scala
index bec8b83b4..779df71b7 100644
--- a/compiler/test/dotty/tools/vulpix/RunnerOrchestration.scala
+++ b/compiler/test/dotty/tools/vulpix/RunnerOrchestration.scala
@@ -159,7 +159,7 @@ trait RunnerOrchestration {
       val url = classOf[ChildJVMMain].getProtectionDomain.getCodeSource.getLocation
       val cp = Paths.get(url.toURI).toString + JFile.pathSeparator + Properties.scalaLibrary
       val javaBin = Paths.get(sys.props("java.home"), "bin", "java").toString
-      new ProcessBuilder(javaBin, "-Dfile.encoding=UTF-8", "-Xmx1g", "-cp", cp, "dotty.tools.vulpix.ChildJVMMain")
+      new ProcessBuilder(javaBin, "-Dfile.encoding=UTF-8", "-Duser.language=en", "-Duser.country=US", "-Xmx1g", "-cp", cp, "dotty.tools.vulpix.ChildJVMMain")
         .redirectErrorStream(true)
         .redirectInput(ProcessBuilder.Redirect.PIPE)
         .redirectOutput(ProcessBuilder.Redirect.PIPE)

I force the locale settings when spawning the JVM process that is executing the compiled tests before checking them with the checkfile.

Is such a solution ok to be added as a PR or do you think it should be solved with some different approach?

[som-snytt]

The f-interpolator tests were updated on scala2 for this reason at some point.

Edit: I was thinking of this test where expected result is formatted locally: scala/scala@c47a60b