lazy fetch only needed files + pin git repo to commit hash about gitfs

[milahu]

lazy fetch only needed files

this is solved in google's slothfs (aka gitfs), but only works with "Gerrit/Gitiles based Git hosting" = no github

background: access monorepos from thin clients

the tree contains a significant amount of unused data

no github

on github, we can use api.github.com to browse commits and trees
and use raw.githubusercontent.com to fetch blobs

this works well for small projects, where only few files are actually needed (syscalls: read, readdir, stat, attr)
for larger projects, this can be too slow, so git clone becomes more attractive

pin git repo to commit hash

the goal here is file-deduplication on fat clients.
fat client = the full git repo was fetched with some depth (mostly-shallow clone),
so we have local access to old versions

what i want is something like this:

gitfs interface

/commit/:sha/tree/ -> root directory by commit sha
/tree/:sha/ -> root directory by tree sha (can be a subfolder of the repo)
/blob/:sha -> file by blob sha

sample:

$ gitfs mount https://github.com/presslabs/gitfs.git /tmp/demo
$ cd /tmp/demo
$ less commit/12886ec5b9c7e103bfcab0cd37a8333873382fae/tree/README.md

this would show exactly this file:

https://github.com/presslabs/gitfs/blob/12886ec5b9c7e103bfcab0cd37a8333873382fae/README.md
https://raw.githubusercontent.com/presslabs/gitfs/12886ec5b9c7e103bfcab0cd37a8333873382fae/README.md