Re: [Boost-users] What's happened to Ryppl?

Subject: Re: [Boost-users] What's happened to Ryppl?
From: John Wiegley (johnw_at_[hidden])
Date: 2011-01-24 22:36:02

Dave Abrahams <dave_at_[hidden]> writes:

>> Exactly.  I'm 98% of the way toward a branchified sequence today.
>
> Awesome!

OK, branchification is working!  All that's left is submodulization as part of
the same run.

This will actually not be very difficult, just time consuming to run.  I'll
use Eric's manifest.txt file, plus 'git log --follow -C --find-copies-harder'
on each element of each submodule, run against the flat history.  The man page
says this is an O(N^2) operation -- where N is very large in Boost's case --
so I may end up having to do some pruning to keep it from getting out of hand.

Actually, the speed of this script is already too slow, so I'm rewriting it in
C++ today both for the native speed increase (10x so far, for dump-file
parsing), and because it lets me use libgit2 (https://github.com/libgit2) to
create Git objects directly, rather than shelling out to git-hash-object and
git-mktree over a million times.  That alone takes over 15 hours to do on my
Mac Pro.  Don't even ask how long the git gc takes to run!  (It's longer).

If anyone wonders whether my process -- which works for any Subversion repo,
btw, not just Boost -- preserves more information than plain git-svn: consider
that my branchified Git has just over one million Git objects in it, while the
boost-svn repository on ryppl has only 593026 right now.  That means over 40%
of the repository's objects got dropped on the cutting floor by git-svn's
hueristics.

John