bpo-38780: Avoid errors in SyslogHandler when socket is not set up.

[jaraco]

https://bugs.python.org/issue38780

[vsajip]

Why is it better to have a NullSocket rather than None? It seems like None would catch more error cases than a NullSocket which accepts an erroneously made method call which then just does nothing.

[jaraco]

Why is it better to have a NullSocket rather than None? It seems like None would catch more error cases than a NullSocket which accepts an erroneously made method call which then just does nothing.

I guess my thinking was that because the None required an if None around each invocation of it, it would be better to embed that Noneness into the underlying object. The point of the PR is to suppress errors when the handler is not fully configured. I was tempted for the NullSocket to be even less fitting to the implementation, and using something like a MagicMock or an object that accepts any interface with a trivial response, but ultimately settled on this narrow fit to cover the known needed cases.

[jaraco]

After some analysis and attempting to write tests, I determined that this approach is inadequate, so I'm closing this PR, although I'll leave the branch and test branch around.

[jaraco]

I've moved the branch to /refs/archive/bugfix/[bpo-38780](https://bugs.python.org/issue38780) in jaraco/cpython.