Make foliage repositories buildable in Nix

[michaelpj]

The main thing this would require is to be able to do all the fetching in a Nix-friendly way, which would require #45 and also some cleverness in working out what foliage needs, e.g. we might need something like:

$ foliage list-requirements
<url1> <hash1>
<url2> <hash2>
...

Then we could use IFD to make a derivation that prefetches everything into a directory, or use the intercepting-webserver trick.

[andreabedini]

💯

I have been thinking about this. Let's start with #45 first but in the meantime:

• We need to figure out how to tell foliage where the fetched inputs are. Maybe just with a decent convention? Switch the cache to be content addressed?
• Maybe we should integrate this into the metadata file and offer a way to produce it separately?

[michaelpj]

We need to figure out how to tell foliage where the fetched inputs are.

Yeah, this is a bit annoying. I think it would be fine to have the answer be "pre-populate the cache" or something. It's some Nix code that we have to write, mostly.

(Amusingly, this is actually a perfect case for recursive Nix (we replace curl with a script that calls fetchurl with the provided hash).)

Maybe we should integrate this into the metadata file and offer a way to produce it separately?

Not sure what "this" is? The "prefetched source location"? I do think we might want something like that, although I think we need a top-level metadata file really for some repository-level options. That might be useful anyway for things like "default value of signing/writing metadata" etc.