mounting files api

[jbenet]

We've discussed making mfs/files api mountable. we should scope out that interface.

I could see something like this:

# mount mfs/files into a specific dir. 
# the `new` param here means a new dir. but could be an existing hash
> ipfs files mount new ~/arch-mirror

# copy stuff over (or wget or whatever)
> cp stuff/* ~/arch-mirror/.

> ipfs files mount-root ~/arch-mirror
/ipfs/Qmdp278UWxHu1aMvSb9tgxaMGzxtAfX7VxZYv6dLSwpJkH

# maybe even this (controversial interface, of course)
> cat ~/arch-mirror/.ipfs-root
/ipfs/Qmdp278UWxHu1aMvSb9tgxaMGzxtAfX7VxZYv6dLSwpJkH

# maybe unmount yields the root, too
> ipfs files unmount ~/arch-mirror
/ipfs/Qmdp278UWxHu1aMvSb9tgxaMGzxtAfX7VxZYv6dLSwpJkH

# mount that root again
> ipfs files mount /ipfs/Qmdp278UWxHu1aMvSb9tgxaMGzxtAfX7VxZYv6dLSwpJkH ~/arch-mirror

[whyrusleeping]

This ties in nicely to the pinning modifications that i'm planning. I want to have ipfs add make an entry somewhere in mfs in order to keep a name associated with the thing you added.

[jbenet]

@whyrusleeping sgtm. how about some of these:

• $repo/all/adds - all added roots, inc add time, unix user if exists (or http client ip for api!). i think you suggest adding this data to the root object of the add, and i really like the idea. we could call these objects add roots.
• $repo/index/by-add-time - an index of all objects by add time
• $repo/index/by-filename - an index of all objects by filename
• $repo/index/by-size - an index of all objects by sizes

notes:

• i see the mfs root at $root/files so the above would be separate from the normal mfs root. we could make all of the above rooted at $repo/files though if we want that.
• indexes are graphs that make search + accesses fast through some properties. they're analogous to database indices in say, SQLite, or some kv-store, etc.
• indexes can be computed over time, and may not need to happen on-add. (can journal an add and insert into indices later, or on a ipfs repo index build command)