Move completions to new testing infrastructure

[gabro]

Still WIP.

This closes #59. I've picked up @olafurpg's work and simply changed the implementation to use safeCompletionsAt. I'll add some extra test cases and merge this, so that we can use it as a base for #62 and #63

[olafurpg]

I just checked out your branch and got it working with CURSOR, that seems to fix the StringIndexOutOfBounds error. Will soon push a fix!

[olafurpg]

I made the enough changes to get the case test passing, we can leave further completion improvements to #63 and #62

[olafurpg]

I'm still not happy with the code organization for the compiler. It seems we should delegate to CompletionProvider/SignatureHelpProvider/HoverProvider to create the compilation unit, since they have different needs for CURSOR. I will try to refactor things a bit more.

[olafurpg]

• Compiler is now ScalacProvider and does nothing but return an interactive.Global
• logic for SignatureHelp/Completions/Hover providers is now fully contained in those objects, with exact same implementation being shared for tests and from the server

I hope this makes sense. There is still quite a bit of duplication for the three providers that can be abstracted away by changing the superclass LanguageServer, but we can leave that for later.

The signature help bug fix is there because I thought it was a regression from the refactorings, sorry for mixing that together.

[olafurpg]

PS, can you maybe open PRs from your personal fork @gabro ? it seems CI runs test on every push to any branch on origin and I get email notifications when those tests fail.

[olafurpg]

Seems CI is unhappy :(

oci runtime error: exec failed: container_linux.go:265: starting container process caused "could not create session key: disk quota exceeded"

Will try to restart tomorrow!

[gabro]

@olafurpg ah sure, I've been opening on the main repo out of habit at work, but I didn't realize Travis would notify you. I'll use my fork from now on.

Also, thanks for the nice refactoring. I think it makes sense to delegate responsibility over the compiler configuration to the single providers.
I wouldn't worry about duplicating code for now: it's way easier to deduplicate later, than to untangle a huge mess like it was about to become. 👍

[olafurpg]

Having a well-defined API between the server and Global will become even more important once we add support for multiple scala versions via classloading. Travis is happy now :D

[laughedelic]

I'm also used to open PRs from the branches in the repo itself at work (isn't it the main point for giving the collaborator permissions?). But I don't mind either way. Anyway, if it's just about Travis notifications, I usually just rurn them off:

notifications:
  email: false

Or it can be configured for only certain emails and only certain events:

notifications:
  email:
    recipients:
      - one@example.com
      - other@example.com
    on_success: never  # default: change
    on_failure: change # default: always

Also docs say that

Pull Request builds do not trigger email notifications

So the problem is in deduplicating the triggering events: pr only for PRs, push only for master. I think I know how to solve it.

[gabro]

I think there's no need to complicate the Travis settings, if the alternative is using a fork model for PRs. I wouldn't lose too much time on this :)

[olafurpg]

Another aspect is that I typically try to keep the number of branches in the origin fork clean. Contributors who fork the repo pull all branches and I don't like it I fetch 100s of WIP branches (many forgotten and will never get merged) when contributing to a project. master and gh-pages is cleaner, IMO

[laughedelic]

Makes sense 👍