ExceptionAssert.ThrowsArgument* should force invariant culture

[pranavkm]

Based on aspnet/Mvc#4260, it looks like we're verifying the exception message returned by ArgumentException without fixing the culture:

Stack Trace:
           … Xunit.Assert.Equal(String expected, String actual, Boolean ignoreCase, Boolean ignoreLineEndingDifferences, Boolean ignoreWhiteSpaceDifferences) dans C:\BuildAgent\work\cb37e9acf085d108\src\xunit.assert\Asserts\StringAsserts.cs:ligne 239
           … Microsoft.AspNetCore.Testing.ExceptionAssert.Throws[TException](Action testCode, String exceptionMessage)
           … Microsoft.AspNetCore.Testing.ExceptionAssert.ThrowsArgumentNullOrEmpty(Action testCode, String paramName)
           … Microsoft.AspNetCore.Mvc.Core.Test.ControllerBaseTest.Redirect_WithParameter_NullOrEmptyUrl_Throws(String url) dans C:\Users\Chalet K‚vin\Documents\GitHub\Forks\Mvc\test\Microsoft.AspNetCore.Mvc.Core.Test\ControllerBaseTest.cs:ligne 85
    Microsoft.AspNetCore.Mvc.Core.Test.ControllerBaseTest.Redirect_WithParameter_NullOrEmptyUrl_Throws(url: "") [FAIL]
      Assert.Equal() Failure
                                         � (pos 32)
      Expected: úúú be null or empty.\r\nParameter name: url
      Actual:   úúú be null or empty.\r\nNom du paramŠtreÿ: url

[dougbu]

@pranavkm we also use Assert.Throws<T>() in a lot of our code. Since xUnit doesn't switch to a particular culture (see Record class), this looks like the wrong place for a "fix". Suggest instead using [ReplaceCulture] or using (new CultureReplacer()) blocks (occasionally) in the test themselves.

[pranavkm]

The unfortunate part is we construct the exception messages in the Assert.ThrowsArgument... method. We could skip that line entirely (split on new lines and such) which would be a saner thing to do. The user really has no expectation that we're generating non-localized strings here.

[dougbu]

How does that change the need to fix the tests rather than this infrastructure piece which works (and should work) much like xUnit's methods?

[Eilon]

This doesn't sound right to me. What if we want the test to run in the local culture? I think the main problem is that we shouldn't be verifying the error messages of code we don't own.

[pranavkm]

Yup. We have a bunch of places where we do a partial match on the exception message. We just need to do something similar here i.e. not verify system messages.