Implement builtins.path

[ryantrinkle]

This Builtin should be added here.

This doesn't have a .exp file; I think that means it just needs to evaluate without error.

Fixes eval-okay-path (remove it from newFailingTests list).

[jwiegley]

This could be challenging, since I think these "enriched paths" are supposed to be interchangeable with other paths.

[xplat]

nix actually handles this by figuring out the store path and using that as a plain path object. There are no actual 'enriched path objects' in nix libexpr.

[xplat]

Actually now that i check more closely, builtins.path returns a string, not a path!

[domenkozar]

Right:

nix-repl> :t builtins.path { name = "patchelf"; path = ./patchelf; }
a string with context

[jglukasik]

I started looking into this today. Haven't made much progress, but I'm getting more familiar with the codebase. Will try again tomorrow.

[jwiegley]

Great! Once this is closed we’ll be able to turn on the full language tests for Travis!

[mightybyte]

@jglukasik Definitely ask for some help from someone on this tomorrow. @shlevy will have all the knowledge you need on the nix side of things. It would be great to get this one finished this weekend.

[jglukasik]

After some discussion with @mightybyte , we've realized this might be a bit too large of a first change for me. I'll let someone else tackle this.

[jwiegley]

I'm going to work on this on the plane tomorrow.

[shlevy]

@jwiegley This requires nar serialization, which @imalsogreg is working on

[jwiegley]

I realized that soon after looking into it, so I’m going to hold off and leave you guys to it.

[jwiegley]

@shlevy since were not reimplementing the store, what are you think about binding to libstore through the FFI?

[ryantrinkle]

We could also shell out as I did for derivationStrict; it's dirty, but if it's easier it might be worth it since we should eventually try not to have a dependency at all.

[jwiegley]

I’m fine with shelling out right now, since it would let us move on to other problems.

[shlevy]

Eh, libnixstore isn't really stable at all and is not a nice interface at all. Not sure why we'd prefer that to implementing the daemon protocol.

[ryantrinkle]

@shlevy Do you think it makes sense to shell out in the immediate future to get this working, and then use the daemon protocol as the next step?

[shlevy]

If that gets us something useful, sure. First step of the required logic to do readonly mode properly (so no need to talk to the daemon) is up at haskell-nix/hnix-store#7

[Anton-Latukha]

layus was implementing path in #755.
It for ported to the current code in #1032. Passes the basic test (./data/nix/tests/lang//eval-okay-path.nix).