Autoclosing issue in package-template closes issue in affiliated package

[bsipocz]

It seems like using the "autofix / close issue" feature in package-template closes the matching issues in packages where the template was used. I'm not sure this is considered to be a bug or just an antifeature of github.

Example:
astropy/photutils#69 (comment)

[astrofrog]

Woah, ok. That's... dangerous. Guess we need to not use this functionality in the package-template commit log?

cc @mdboom @embray @eteq

[eteq]

Whoa, yeah, that's bizzare and clearly we have to be cautious here. I'm going to ask github support about this and I'll post here if they have any suggestions.

[mdboom]

Ouch. They must have some way to avoid this when a repository is explicitly created as a fork of another one, since this doesn't happen (I don't think) to mdboom/astropy when astropy/astropy autocloses a bug. The difference with affiliated packages is that they are not explicit forks (github doesn't have metadata about that), they just pull in commits from package-template. That's a good one, and I don't even see an obvious way that github could avoid doing what it's doing. If we still want to use this feature it might work to say explicitly: Fixes astropy/package-template#X rather than just #X.

[eteq]

@bsipocz @astrofrog @mdboom @embray - github support's advice was pretty much just "You might like to use the "Fixes User/Repository#Num" format instead, which should only close the issue in the repository specified, regardless of where your commit appears". So I think we'd better make sure we stick with astropy/package-template### from now on.

[astrofrog]

Ok, sounds good! Not sure where we should document this...

[bsipocz]

Another similar issue is that package template tags are being inherited in packages and are soo damn difficult to clean up years after the fact (e.g. we have 1.0 etc in astroquery, that are coming back each time I fetch from someone else's fork).

For the issues, we can indeed deal with it with using the astropy/package-template###, as suggested above. Also, I believe it becomes less of an issue since we switched to cookiecutter rather than suggesting to bring the repo history.

So overall, I suggest to close this, as there isn't much we can do here (we could rewrite the commit messages and force push, a tiny bit drastic, but that's the only thing we could do about the past).

[pllim]

This issue will still be pinned even after close.