Lazy index loading (1/2)

[mosteo]

This is the first of two PRs that remove the need to load the full index for many operations.

The idea is that we know what indexes are configured, where they are on disk, and where each crate lives therein, so any time a crate is needed, we can load it at that time. All in all, this should speed-up things when indexes grow, and fewer external detections should be attempted.

There is a catch that makes this not as straightforward as it seems, which is that "provides" fields mean that a provided crate may require loading whichever other crates provide it, and this is unknown in advance until the whole index has been loaded at least once.

This part is what goes into the follow-up PR: the first time we "see" an index, it is loaded in full and provider -> provided mappings are stored in a new metadata file that is invalidated whenever indexes are added or updated. From then on, we know all the crates that must be loaded when one is requested, and lazy load works as intended in all cases.

Fixes #910, fixes #381

[Fabien-Chouteau]

new metadata file that is invalidated whenever indexes are added or updated

What if I add a new crate in my local index? Will this be detected?

(I didn't look at the PR content yet)

[mosteo]

What if I add a new crate in my local index? Will this be detected?

You're right, this won't be detected, although running alr index --update-all is all it takes to fix it. I don't see any simple way of detecting this that doesn't involve complex filesystem monitoring.

I will add a note about it in the publishing documentation.