Add git-number script to calculate generation numbers for commits.
Compatible with any git topology (multiple roots, weird branching/merging, etc.)
I can't get it to be any faster (in python). Suggestions welcome :).
On z600/linux, this takes 5.1s to calculate the initial count for 2e3de954ef0a
(HEAD on src.git at the time of writing). Subsequent lookups take ~0.06s. For
reference, this machine takes 3s to just list the revisions in sorted order
without any additional processing (using rev-list).
All calculations are stored in a git-notes-style ref with the exception that the
leaf 'tree' object which would normally be stored in a git-notes world is
replaced with a packed binary file which consists of records [hash int]. Each
run of this script will create only 1 commit object on this internal ref which
will have as its parents:
* The previous git number commit
* All of the target commits we calculated numbers for.
This ref is then excluded on subsequent invocations of rev-list, which means that
git-number will only ever process commit objects which it hasn't already
calculated a value for. It also prevents you from attempting to number this
special ref :).
This implementation only has a 1-byte fanout which seems to be the best
performance for the repos we're dealing with (i.e. on the order of 500k commit
objects). Bumping this up to a 2-byte fanout became extremely slow (I suspect
the internal caching structures I'm using are not efficient in this mode and
could be improved). Using no fanout is slower than the 1 byte fanout for lookups
by about 30%.
R=agable@chromium.org, stip@chromium.org, szager@chromium.org
BUG=280154,309692,skia:1639
Review URL: https://codereview.chromium.org/26109002
git-svn-id: svn://svn.chromium.org/chrome/trunk/tools/depot_tools@236035 0039d316-1c4b-4281-b951-d872f2087c98
7 files changed
