Avoid hitting RubyGems remotes when all missing gems are in --without groups

[segiddins]

What was the end-user problem that led to this PR?

The problem was we would attempt to fetch dependency info from a RubyGems remote even when it was only needed for gems whose groups were excluded.

Was was your diagnosis of the problem?

My diagnosis was that the test we had for this case was insufficient, since it checked only that stderr was empty, and since Bundler 1 prints errors to stdout... it went undetected for a year.

What is your fix for the problem, implemented in this PR?

My fix is to only report that deps are missing when they are either requested or are path/git/gemspec-sourced.

Why did you choose this fix out of the possible options?

I chose this fix because it keeps the behavior of always ensuring all git repos are checked out.

[indirect]

Any chance we can also skip checking out git sources if they only contain gems that are excluded? That would fix the regression I found in 1.8 last year :P

[segiddins]

Any chance we can also skip checking out git sources if they only contain gems that are excluded? That would fix the regression I found in 1.8 last year :P

I don't know how well I can make that work with rspec ./spec/install/git_spec.rb:39 # bundle install git sources should check out git repos that are missing but not being installed

[indirect]

@segiddins that's when there's no lockfile and Bundler needs to use the gemspec from that git repo to successfully resolve. I'm talking about when we're running in deployment/frozen mode, and there's already a lock, so we don't actually need the gemspec for the without gem.

[segiddins]

@indirect I think I did it?

[indirect]

🎉

[segiddins]

r?

[indirect]

My test for the regression is still failing inside Definition#converge_paths, where specs_changed? requires the directory to exist. Are we checking every source for changes, whether or not we need any gems from that source, even in frozen mode?

[indirect]

(The repro is here if you want to see it for yourself: https://gist.github.com/indirect/1cc86d686f40a40d68b895834ab11cf6)

[segiddins]

That looks like a different issue? That gemfile doesn't have any Git sources.

[segiddins]

Lol I think I have a fix for it anyways

[indirect]

lol 🙌🏻

[segiddins]

@indirect take a look :P

[indirect]

you did it! 🎉🎉🎉

[indirect]

@bundlerbot r+

[bundlerbot]

📌 Commit 1947060 has been approved by indirect

[bundlerbot]

⌛ Testing commit 1947060 with merge 1102ec1...

[bundlerbot]

☀️ Test successful - status-travis
Approved by: indirect
Pushing 1102ec1 to master...