This is a proposal to move to a API call based dependency resolution, rather than the Git based index clone. This is similar to what NPM uses.

Mainly, git based index clone has scalability issues:

• The initial clone is a burden. Even after squashing it still takes a few megabytes, which is poor UX on throttled mobile network. A NPM-like solution consumes much acceptable traffic.
• It doesn't scale. If we had as much packages as NPM, the index will be too large to download.

Then, here are some random notes:

• Resolution caching should be done at HTTP level. This approach is good because it is also honoured by the CDN (if we eventually use one).
• When working offline we can rely on crates we cached locally to resolve dependencies.
• To replace the functionality of the crates.io-index repository, we can have daily database dumps, and offer an API and/or UI to display the latest updated crates.