Call super in Formatter.close methods to ensure sync is reset

[rahearn]

There were two instances (that I could find) of overridden methods in classes inheriting from BaseFormatter that likely should have called super to complete the cleanup.

[JonRowe]

Thanks for spotting this, theres an unnecessary change to be reverted in the base formatter, and ideally some sort of test to be added to show the problem this causes if any

[rahearn]

@JonRowe @pirj I was finally able to set up a 2.7 environment that did successfully reproduce the CI failures on 2.7, so I can dig into this.

I have a question about preferred implementation. The test failures are because in three situations it's possible to call formatter.close without ever calling formatter.start. Two of those are formatter unit tests, so it's easy enough to just add start to the test setup. The other is in https://github.com/rspec/rspec-core/blob/main/lib/rspec/core/bisect/coordinator.rb#L42 where BaseTextFormatter is used but outside of the normal Reporter structure so start is indeed never called. Which of these solutions would be preferred?

1. Add some sort of sync_started guard to BaseFormatter that is initialized to false and set to true when start is called that can be checked before resetting the sync.
2. Add a specific call to @notifier.publish(:start, count: 0) at the beginning of the Coordinator#bisect method
3. something else?

I'm leaning towards the first option since it will protect against future uses of formatters outside of the Reporter notification system in addition to fixing the current problem. Let me know if you disagree

[JonRowe]

I'm glad this passes as is, but is it in the right place? This is actually implemented in the base formatter, and if this is testing the super call then it doesn't cover the base formatter and the json formatter, prehaps some of these should be a shared example we call across all the formatters that need it?