branch 2.14 version of #4918

[pjfanning]

seems to show that the issue in #4917 / #4918 is new-ish

[cowtowncoder]

@pjfanning Ah ok so this is where we get the tests.

I can merge into 2.14 and forward, although typically we don't really do PRs against un-maintained branches.
But I guess in this case there's also the question of how to coordinate fix vs test that fails for ... 2.18+ only?

[pjfanning]

Starts failing in 2.15 - 2.14 is the last branch where it works.

[cowtowncoder]

Starts failing in 2.15 - 2.14 is the last branch where it works.

Ok. So I think test should:

1. Be disabled in intermediate versions where it'd fail (I don't see much point in fixing 2.14 as there's not going to be releases)
2. Only merged in for version being fixed (2.17 or 2.18; maybe it's safe enough to go in 2.17... or possibly 2.16)
3. Or if we really want to, be merged in 2.14 as passing and then add fix in 2.15 along with merged test

I am leaning towards (2) but can go with other options as well. While nasty bug, I don't think I'll be tempted to release 2.14 - 2.16 any more (2.16 is nominally open until 2.19.0)