Unstable tests based on AssertRefLogEntry

[AArnott]

Several tests are unstable and fail all too frequently on CI and perhaps locally.

From this run of this commit the windows agent happened to fail with this log:

  X LibGit2Sharp.Tests.BranchFixture.CanCreateBranchFromCommit [63ms]
  Error Message:
   Assert.InRange() Failure
Range:  (1/31/2020 4:54:25 AM +00:00 - 1/31/2020 4:54:25 AM +00:00)
Actual: 1/31/2020 4:54:24 AM +00:00
  Stack Trace:
     at LibGit2Sharp.Tests.TestHelpers.BaseFixture.AssertRefLogEntry(IRepository repo, String canonicalName, String message, ObjectId from, ObjectId to, Identity committer, DateTimeOffset before) in D:\a\libgit2sharp\libgit2sharp\LibGit2Sharp.Tests\TestHelpers\BaseFixture.cs:line 471
   at LibGit2Sharp.Tests.BranchFixture.CanCreateBranchFromCommit() in D:\a\libgit2sharp\libgit2sharp\LibGit2Sharp.Tests\BranchFixture.cs:line 205

It looks to me like the before local variable is still set to a later value than is recorded in the commit at times, for whatever reason.

[AArnott]

I see a bunch of tests that use

libgit2sharp/LibGit2Sharp.Tests/BranchFixture.cs

Line 199 in 2407477

var before = DateTimeOffset.Now.TruncateMilliseconds();

As a marker. Should they all be updated to subtract one second? Or should we update AssertRefLogEntry to deduct one second from the before arg passed in?

[AArnott]

Case in point: another agent from that same commit failed another test with exactly the same pattern.

  X LibGit2Sharp.Tests.BranchFixture.CanCreateBranchUsingAbbreviatedSha [77ms]
  Error Message:
   Assert.InRange() Failure
Range:  (1/31/2020 4:57:07 AM +00:00 - 1/31/2020 4:57:07 AM +00:00)
Actual: 1/31/2020 4:57:06 AM +00:00
  Stack Trace:
     at LibGit2Sharp.Tests.TestHelpers.BaseFixture.AssertRefLogEntry(IRepository repo, String canonicalName, String message, ObjectId from, ObjectId to, Identity committer, DateTimeOffset before) in D:\a\1\s\LibGit2Sharp.Tests\TestHelpers\BaseFixture.cs:line 471
   at LibGit2Sharp.Tests.BranchFixture.CanCreateBranchUsingAbbreviatedSha() in D:\a\1\s\LibGit2Sharp.Tests\BranchFixture.cs:line 119

[bording]

I was looking at that tonight, and thought #1762 might be a fix, but it looks like the range in the assert isn't the problem. Despite capturing the before value before test operation, sometimes the operation has a value that is earlier.

While we could just make the tests pass by opening up the range, I'm not sure that's the right thing to do. Seems like this could actually be a legit problem in how those values are being calculated.

[AArnott]

Seems like this could actually be a legit problem in how those values are being calculated.

Well, I guess I'll let you decide the fate of my PR that I just filed. But unstable tests cost a lot, so my little vote would be, accept my PR and leave this bug open to track investigating the underlying product failure if you want to.
But IMHO, no one will care about half a second rounding error like that. We've got bigger issues to spend time on... like #1618. :)

[bording]

But unstable tests cost a lot, so my little vote would be, accept my PR and leave this bug open to track investigating the underlying product failure if you want to.

Seems like a pragmatic way to move forward. Let's see if the PR is green, and if so I'll merge it.

[bording]

Reopening the issue so we can keep it around to investigate why #1765 was needed as a workaround.