Coda File System

codadev mailinglist & important vice.rpc2 changes

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Wed, 4 Aug 1999 11:19:16 -0400
Hi everybody,

First of all, I invite anybody who is interested in getting their hands
into the dirty parts of Coda (some people call this 'the sourcecode') to
subscribe to codadev_at_coda.cs.cmu.edu.

    echo | mail -s subscribe codadev-request_at_coda.cs.cmu.edu

As our development team has become less centralized, and it is not
possible to just walk into another developer's office across the
corridor, we need an adequate forum for code-hacking discussions.

Don't worry about all interesting discussions leaving codalist_at_coda.
This list will continue to be the main forum for most questions and
discussions about installation, administration, and `wishlist' ideas.

Codadev is intended to be a relatively low-volume list that covers more
in-depth/bleeding edge/development issues such as:

- Mayor changes in the CVS tree.

    Although we all are writing nice `changelog' entries when we do CVS
    checkins, they sometimes need some additional info/warnings.

- Bouncing (realistic) ideas off others.

    f.i. If a volume is mounted in 2 places, should we make the second
    place where we see the mountpoint a symlink to the first. _I can
    probably do this in a couple of days_, but I haven't considered how
    to handle situations like when that first mountpoint disappears.

    The main difference with `wishlist' ideas, is IMHO whether you
    actually looked at the code to see if it can be done.

- Programming/debugging related questions.

    Does anybody but myself know how to dump/examine the stacks of all
    server lwp's to debug server-deadlock problems. 

I guess you get the idea. I will set up a web-thingy so that people can
follow this list on the web as well.

Jan

Following is the first codadev_at_coda email.
======================================================================
From: Jan Harkes
Subject: Incompatible RPC2 protocol changes

With the introduction of writeback caching, an extra field was added to
the ViceGetVolumeStatus rpc. This breaks at least "cvs listvolume" and
"repair". Furthermore, when non-wb clients attempt to connect to one of
the new WB servers the server cannot set up the new server->client
writeback connection and will drop the connection to the client.

The choice here is to either patch things up to be compatible, or to
completely break the rpc2 protocol and throw in as many other pending
incompatible changes as we can find.

I believe we should break things. Yesterday I've made the necessary
changes to fix the last CountedBS issues that were pending as of Oct 7th
last year (see BUG#695). I also split ViceGetAttr from the ViceFetch
rpc.

When I check my changes in, any client build from the head of the CVS
tree will not be able to talk to existing servers. But at least that is
less deceptive than the current situation of the development tree, where
some things silently fail.

Jan

btw. I am planning to commit this before the end of the week, and
release the head of the cvs tree as coda-5.3.0.
Received on 1999-08-04 11:21:52