[openib-general] configuration management for OpenIB

[Roland Dreier]

    Tom> Doesn't git have a very inefficient way creating blame files
    Tom> or creating a line-by-line versioned file?

Good point.  Yes, this is one other advantage that the hg data
structure design has -- it makes "hg annotate" possible and rather
fast.  For example, on a kernel tree with ~33000 changesets, I get:

$ time hg annotate include/linux/kernel.h  > /dev/null
real    0m1.584s
user    0m1.180s
sys     0m0.218s

I picked <linux/kernel.h> semi-randomly as a file likely to have a lot
of changes -- it is touched by 77 changesets in this repo's history.

On the other hand, Linus would argue that the git design of tracking
history over the whole tree rather than by individual file has its
advantages for other operations (in addition to being conceptually
very clean).

    Tom> Or does repacking fix this?

No, repacking doesn't change the data structures in any fundamental
way.  It just puts multiple blobs in a single file.

 - R.