A caching layer for git checkouts

[happz]

Based on https://issues.redhat.com/browse/TFT-2461:

discover:
  - name: first
    how: fmf
    url: http://github.com/kernel/big-pile-of-tests
    test:
      - networking/ipv6/misc/checksum
  - name: second
    how: fmf
    url: http://github.com/kernel/big-pile-of-tests
    test:
      - networking/ipv6/misc/checksum

is supported, but will consume a lot of resources - network traffic, disk space, time, etc. - by cloning a very large repository twice.

As discussed in the tmt hacking meeting, this is a complex subject. tmt lacks a caching layer for git repos, and the general mood was it should get one. git clone would then be able to save some space and network by re-using already cloned data for read-only access.

• Could we do some kind of caching?
  • Only the git metadata?
  • We could share the cached copy for the whole tmt run instance
  • There is some proof-of-concept already used in workflow-tomorrow
• What about handling different refs?
  • We could track cache per url-ref combination
• What is the essential problem?
  • Disk space or network connection bandwidth?
  • Seems the disk space is the more pressing one (for the kernel team)
  • Would be good to track disk/network issue separately
• Development branch means that the whole history is currently fetched