Fix missing default during pytest_configure - #5435

@josePhoenix From what I've seen are the test failures not because of this PR. So you don't need to worry (yet) about these.

I think someone (just semi-randomly pinging @pllim) will restart those tests soon.

Restarted.

Travis passed but coverage decreased 0.4% (probably unrelated to this PR but due to #5418).

It is kinda a bug fix but not a very visible one. Does this need change log?

👍 from me too - thanks @josePhoenix!

I think it should get a changelog entry, though, as @josePhoenix's case is not necessarily that unique. @josePhoenix, can you add this to the 1.2.2 section of CHANGES.rst under "Other Changes and Additions"?

(aside: @pllim or maybe @astrofrog - do you know if there's any way I can add files with github's new "edit a PR branch" feature? If there was I could just add the changelog entry myself... I thought someone mentioned this on the astropy-dev thread but don't see a way to do it?)

@eteq apparently you can now commit directly to this PR branch. So that's how you add the entry. I tried that feature only once but it failed miserably because i had to rebase and it won't let me push forcefully.

Cheers! I'd forgotten I needed to do that.

Oh, whoa, that did work! Of course now the changelog conflicts... but still. Neat!

@josePhoenix - one other question before I try to merge this fixing the conflicts: can you confirm making this change get py.test working in your package? Or is there something else that needs changing? (Which it would make sense to bundle in with this).

Turns out, there are a bunch more unchecked exception cases in this plugin. This PR needs more attention than I can give it at the moment, I'm afraid. There are also calls to deprecated methods in pytest that need to be updated.

Yes, these other exceptions are a known issue (see #5277). I was under the impression that you only needed this one-liner change to make your own package using Astropy template work.

By "unchecked exception cases" I mean cases where the code could raise an exception and the exception is not checked. For example, the one I fixed in the first commit in this PR. It turns out that there are a lot more of these in pytest_plugins.py, which seems more or less orthogonal to the issue you linked to.

I've opened #5446 instead, as the docs are currently misleading. Fixing pytest-command support seems to be more involved than anticipated. (Once I fixed one missing default, another appeared. And when I fixed that one, another.)

Addendum: in an offline chat w/ @josePhoenix it looks like this might actually be due to a change to how options are set as defaults. So this might be a new aspect of #5277 after all: options now may have to specify defaults, which would be why this change was needed if you try to invoke it from py.test 3.x ...

Why, pytest? Why?! 😱