looks good to me.

This is fine and I'm on the edge of pressing the button, but... I think that adding the test is overkill. Here's the larger issue: test code is still code, still subject to bugs, still a potential cause of puzzlement among developers, still something that needs maintenance. The test here is for a very specific problem that has been solved in a general and obvious way--there are probably hundreds of similar places in the code where problems like this have been solved, or never arose, and it would do us no good to put in tests for them. Similarly, I think that in the long run having this test included will be a net drag, not a net gain, for the development and maintenance process.

Aha, I didn't update my web page before writing that comment, so I see my point is moot for this PR. Nevertheless, I think that this issue of how to ensure our test suite is lean and effective, and minimally puzzling, is something we need to consider more carefully in the future.

@efiring Want me to open a PR to remove that test?

Only if you agree with my rationale and feel motivated. We can discuss this on Monday if you like. It's really more the big picture that I'm concerned about.

I also half-viewed those tests as a down-payment on more through quiver tests (as this was a code path we do not otherwise touch) so it is getting us some additional (smoke-test) coverage.