On Wed, 29 Jan 2003, Lawrence Greenfield wrote:
> Nonsense. A client should be implemented with the author's best guess
> of what is and isn't expensive.

That's the argument used by Netscape and Outlook -- program the client so
it works with one particular server, don't worry about everything else.
That's why those clients are so bad.

To design interoperable software, it is necessary to consider *all*
compliant environments.  Good clients work well with *all* compliant
servers.

It is hard to write a good client.  Irresponsible comments do not make it
any easier.  It is damn irresponsible to tell people "ignore what the spec
says, it will work OK with my server."

> Very few servers have modes where STATUS is more expensive than open
> connections.

You deal with a very narrow category of server in which the mail store was
specifically designed for IMAP from inception.  Your sample space is not
representative; it is a minority of the deployed server base.

> NOOPs may also be
> expensive but a client would be foolish to assume that NOOPs will be
> resource intensive.

False.  If you read the specification more carefully, you'll find that
NOOP is less resource intensive than other IMAP commands by definition,
since all other IMAP commands do the action of a NOOP.

> The performance of STATUS or NOOP or extra connections is a
> quality-of-server implementation issue.

False.  The specification says that STATUS is not required to be fast.

> If clients require fast
> STATUSes, people will select servers that give them that. If clients
> require hundreds of simultaneous connections to a single server,
> people will select servers that give them that.

False.  If a client is abusive, server managers will treat it as any other
DoS situation.  They will ban the client, and institute technical means to
lock it out.

-- Mark --

http://staff.washington.edu/mrc
Science does not emerge from voting, party politics, or public debate.

Reply via email to