Re: A forward-thinking thought about deltas in FS design

At 4:40 AM -0400 7/5/03, Greg Hudson wrote:
>We've always taken it as
>axiomatic that the most recent rev of a file should be stored in plain
>text, and earlier revs should be stored as deltas against it, so that
>checkouts of the most recent rev are fast. That's what CVS does (on
>the mainline, anyway), and it was certainly important back when we had
>no delta combiner and no skip-deltas.

Well, CVS does that because RCS does that. And RCS does that because
Tichy published a paper showing that it was faster for retrieving the
HEAD revision on files of around 200 lines (with a "lemma" survey
showing that this was the average or typical file size at the time).
On that basis, he implemented RCS that way, and it got a lot of
academic macho points and took over. A few years later, people
noticed that 200 lines was no longer typical (on my project, the 50th
percentile is at 29682 lines!), and that "the RCS advantage" was
illusory at such sizes. Commercial developers also have much less
allegiance to HEAD, being into things like supporting older versions,
patching patches, and working on isolation branches (which really
kills the RCS assumptions).

So, yeah, I like where you're headed, Greg!

-- 
-==-
Jack Repenning
CollabNet, Inc.
8000 Marina Boulevard, Suite 600
Brisbane, California 94005
o: 650.228.2562
c: 408.835-8090
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org