Add analyzer and getter for source manager

[sdboyer]

Two things in this:

• analyzer, which is the simplest possible implementation of a gps.ProjectAnalyzer. That's the thing that's responsible for pulling manifest and lock data out of dependency projects.
• A dumb-as-rocks singleton setup of a gps.SourceManager, which is going to be needed for all solving operations, and will have quite a number of uses outside of that, as well.

I took care in gps to ensure that the default gps.SourceManager (*SourceMgr, returned from gps.NewSourceManager()) is completely encapsulated - no package-level state. (There's still some work to be done related to exclusivity controls at the filesystem level, as well as signal handling, but we don't have to care about that right now). That makes it safe to take this dumb approach to setting up our sm.

Longer term, though, it'd probably be preferable to inject the sm we use, rather than relying on a global.

[sdboyer]

Also worth noting that this does codify a relationship with GOPATH - we add a new dir, $GOPATH/depcache, for local copies of source repositories.

[freeformz]

I'm not sure what the point of getSourceManager is. Either have a singleton and use it or don't. If you have a singleton any setup should be done in init() IMO.

[sdboyer]

Totally fair point. I had two reasons for not doing it in init():

some commands - e.g. init - may not actually need the source manager. And, because setting up a *SourceMgr entails writing out a file to disk (to maintain single-process access to the cache dirs), that's something I try to avoid unless it's strictly necessary.

Second, because *SourceMgr.Release() needs to be called in order to remove that lock file, the point at which the manager gets initialized is important - you generally want to defer sm.Release() right afterwards. Controlling responsibility over that becomes even more awkward when that's done in init().

[sdboyer]

Hmm...but...that Release() is still awkward to manage, as this PR is currently written. Maybe it'd be better to drop the package-level var and just have getSourceManager() try to set up a new one each time. Then its only purpose is to apply defaults, not control instance access.

Yeah, I like that better.

[sdboyer]

Also LOL i wasn't actually even writing to the sm package var

[freeformz]

If I call this twice, I assume that I get an error on the second call based on what we discussed in slack?

[sdboyer]

correct. it would be a gps.CouldNotCreateLockError, which has enough info in it to differentiate from e.g. fs perm problems

[sdboyer]

So now, you'd have to pick one place that you create the sm, which then is also the point where you have to decide about Release()ing it.

[freeformz]

tbh I would just push all of this into a PR that actually uses it.

[sdboyer]

sure, i can do that, too

[sdboyer]

Closing in favor of #8