Determine whether we should use a reference repo

Problem/Motivation

Our current AMIs contain a local copy of the Drupal core repo.

We could use this repo for checkouts instead of cloning the repo across the network.

Proposed resolution

• Experimentally add this behavior to the checkout plugin.
• Determine whether there is a benefit.

Remaining tasks

User interface changes

API changes

Data model changes

Original report

Checking somebody's comment, I took a look at the logs of the fastest testbots cloning contrib projects for contrib tests, and found that some clones were in fact taking 2-3 minutes, when they should take 5-10 seconds. I was baffled because a manual clone didn't take that long. However, when I added --reference /var/cache/git/reference, the checkout took twice as long for views. However, it took 3 minutes and 45 seconds for commerce.

It's possible that this is a regression in the (old) version of git we're pinning to because PIFR 6.x-2.x doesn't have the ability to provide a valid patch for current versions of git. I'll check that...

root@www:/tmp# time git clone git://git.drupal.org/project/views.git
Cloning into views...
remote: Counting objects: 52456, done.
remote: Compressing objects: 100% (12131/12131), done.
remote: Total 52456 (delta 38684), reused 52133 (delta 38472)
Receiving objects: 100% (52456/52456), 15.09 MiB | 875 KiB/s, done.
Resolving deltas: 100% (38684/38684), done.

real	0m21.716s
user	0m1.416s
sys	0m0.096s
root@www:/tmp# rm -rf views
root@www:/tmp# time git clone --reference /var/cache/git/reference git://git.drupal.org/project/views.git
Cloning into views...
remote: Counting objects: 50329, done.
remote: Compressing objects: 100% (11245/11245), done.
remote: Total 50329 (delta 37612), reused 49738 (delta 37374)
Receiving objects: 100% (50329/50329), 14.89 MiB | 438 KiB/s, done.
Resolving deltas: 100% (37612/37612), done.

real	0m46.655s
user	0m3.692s
sys	0m0.208s