Date: Mon, 8 Jul 2002 15:15:12 -0700 (PDT) From: Doug Barton <DougB@FreeBSD.org> To: Terry Lambert <tlambert2@mindspring.com> Cc: arch@FreeBSD.org Subject: Re: Package system flaws? Message-ID: <20020708150549.D84324-100000@zoot.corp.yahoo.com> In-Reply-To: <3D295B77.E62ED5FC@mindspring.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 8 Jul 2002, Terry Lambert wrote:
> Metadata is metadata.
For sufficiently broad definitions of metadata, ok.
> If it's good to put metadata in a file seperate from the data it
> describes, then the judgement of "goodness" is universal.
This I disagree with. Insert obligatory quote about "foolish consistency."
> We do this because it makes the data and metadata non-severable,
See a later post by me on this thread where I clearly state severability
as a specific, and desirable goal for this application.
> it relieves us of having to consider synchronization issues which
> would otherwise arise, and we do it because it's convenient in
> terms of speed of operations involving both, and convenient in
> terms of locality of reference.
In this case, I think that the costs of synchronization are very small,
and easily addressable by existing tools. Whereas, the benefits of
severability are very great, and easily offset the costs of both not
keeping them together, and the cost of actually performing and maintaining
the severance.
> It also gets rid of the implied graph edge for locking of data and
> metadata, which can lead to an undetectable deadly embrace deadlock .
I agree that this is a factor we need to keep an eye on. However, I think
that the problem space is sufficiently small that we'll easily pass the
90/10 rule, and may even get as high as 95/5 with little additional
effort.
> All of these arguments apply equally well to bundling package
> metadata with package data: conceptually, that metadata is no
> different than file ownership, flags, or permissions.
And here is where we would need to actually define which metadata we're
considering splitting out where. I think dependancy data is a slam dunk
for being split out into a seperate file, both because it's already been
identified as something we need to know before we download the whole
binary package, and because it is rather likely to change independent of
the actual binaries themselves.
I'm studiously trying to avoid focusing on implementation details because
I'd like to spend some more time discussing the problem space, but if
it'll help us understand the problems better, I could describe what I have
in mind in a little more detail....
Doug
--
"We have known freedom's price. We have shown freedom's power.
And in this great conflict, ... we will see freedom's victory."
- George W. Bush, President of the United States
State of the Union, January 28, 2002
Do YOU Yahoo!?
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20020708150549.D84324-100000>