fix(ci): do not install_egg_info to fix importlib_metadata not found

[joshuarli]

pluggy (via pytest)'s complaining about importlib_metadata again. This happened with the Riak test suite, which was removed (#15072) after I noticed it was deprecated. Back then, that was the path of least resistance, but now the issue's come back.

Related: pytest-dev/pluggy#205

[joshuarli]

Dang, adding the importlib backport explicitly led to

  File "/home/travis/virtualenv/python2.7.15/lib/python2.7/site-packages/importlib_metadata/_compat.py", line 19, in <module>

    from backports.configparser import ConfigParser

ImportError: No module named backports.configparser

And pinning pluggy back down to 0.11.0 results in the same. Actually forcing pluggy back down will break during runtime:

pluggy.manager.PluginValidationError: Plugin 'metadata' could not be loaded: (pluggy 0.11.0 (/home/travis/virtualenv/python2.7.15/lib/python2.7/site-packages), Requirement.parse('pluggy<1.0,>=0.12'), set(['pytest']))!

[joshuarli]

Yep, looks like it was because we did python setup.py install_egg_info. Not entirely sure why but there was talk among pytest/pluggy devs about eggs not being too well supported.

EDIT: see #15255 (comment)

[wedamija]

Seems ok, not sure if there was some reason we installed eggs in ci?

[joshuarli]

Don't really know much about eggs, but I suspect it has something to do with testing sentry releases...? @mattrobenolt is this install_egg_info thing in ci good to be removed?

Edit: @BYK see above^

[joshuarli]

Hmm, ci in master seems to be working again. Probably some weird caching stuff, but I wanna say travis caching is per pr/branch. Wonder if the real solution was just to nuke travis cache (which was done during this pr). Because I was confused reading pytest-dev/pluggy#205 since the egg issues seem to have been fixed.

Anyways, I don't think these changes hurt anything.

[BYK]

So this was added by @mitsuhiko in #10654. Not sure you asked for my review but I don't have anything intelligent to say 😀

Yielding to @mitsuhiko

[joshuarli]

Oh, matt just guessed you'd have more context. (I didn't try and follow the git blame trail, haha)

[mitsuhiko]

@BYK it clearly says workaround ;) You overstate my comprehension of the nightmare that is python packaging.

[BYK]

@mitsuhiko, what I meant was I have no idea whether we still need this workaround or not and you were the closest to this car crash 😁

[mitsuhiko]

From my understand we still need it. Yes. Otherwise travis will break on certain dependency changes.