Re: 1.3.0 creating bigger packs than 1.2.3

From: Linus Torvalds <torvalds@osdl.org>
Date: Apr 20, 2006, 17:03 UTC
Message-ID: <Pine.LNX.4.64.0604200954440.3701@g5.osdl.org>
In-Reply-To: <20060420164351.GB31738@spearce.org>

On Thu, 20 Apr 2006, Shawn Pearce wrote:

> 
> Oddly enough repacking the v1.2.3 pack using 1.3.0.g56c1 created an
> even smaller pack ("git-repack -a -d"):

That's "normal". Repacking without -f will always pack _more_, never less. So a different packing algorithm can only improve (of course, usually not by a huge margin, and it quickly diminishes).

> but then adding -f definately gives us the 2x explosion again:
> 
>   Total 46391, written 46391 (delta 6649), reused 37894 (delta 0)
>   129M pack-7f766f5af5547554bacb28c0294bd562589dc5e7.pack

Right. Doing the -f will discard any old packing info, so if the new packing algorithm has problems (and it obviously does), then using -f will show them.

Show 7 quoted lines

> > You could try to revert that change:
> > 
> > 	git revert eeef7135fed9b8784627c4c96e125241c06c65e1
> 
> Whoa.  I did that revert and fixup on top of 'next'.  The pack
> from "git-repack -a -d -f" is now even larger due to even less
> delta reuse:

Ok, so that wasn't it, and the new sort order is superior.

That means that it probably _is_ the delta changes themselves (probably commit c13c6bf7 "diff-delta: bound hash list length to avoid O(m*n) behavior". You can try

	git revert c13c6bf7

to see if that's it. Although Nico already showed interest, and if you make the archive available to him, he's sure to figure it out.

> With --window=50 on 'next' (without the revert'):
> 
>   Total 46391, written 46391 (delta 6666), reused 39723 (delta 0)
>   129M pack-7f766f5af5547554bacb28c0294bd562589dc5e7.pack

Yeah, that didn't do much. Slightly more deltas than without, but not a lot, and it didn't matter much size-wise.

You can try "--depth=50" (slogan: more "hot delta on delta action"), but it's looking less and less like a delta selection issue, and more and more like the deltas themselves are deproved.

			Linus