From: Shawn Pearce <spearce@spearce.org>
Date: Mon, 14 Aug 2006 04:17:05 GMT
Subject: Re: Compression and dictionaries
Message-ID: <20060814041705.GD18667@spearce.org>
In-Reply-To: <9e4733910608132107j7bca0271g360de3447febbf51@mail.gmail.com>

Jon Smirl <jonsmirl@gmail.com> wrote:
> The zlib doc says to put your most common strings into the fixed
> dictionary. If a string isn't in the fixed dictionary it will get
> handled with an internal dictionary entry.  By default zlib runs with
> an empty fixed dictionary and handles everything with the internal
> dictionary.
 
> Since we are encoding C many strings will always be present (if,
> static, define, const, char, include, int, void, while, continue,
> etc).  Do you have any tools to identify the top 500 strings in C
> code? The fixed dictionary would get hardcoded into the git apps.

Actually GIT itself may also benefit from other strings beyond
those common found in C-like languages:

	'10644 '
	'40000 '
	'parent '
	'tree '
	'author '
	'committer '

as these occur frequently in trees and commits.
 
> A fixed dictionary could conceivably take 5-10% off the size of each entry.

Could be an interesting experiment to see if that's really true
for common loads (e.g. the kernel repo).  I don't think anyone has
tried it.

-- 
Shawn.