From: Wietse Venema (no email)
Date: Mon Mar 03 2003 - 09:06:33 EST
Jos? Luis Tall?n:
>
> >There is an unfinished dict_tcp module in the Postfix source code.
>
> Cool
>
> >It compiles and works. But that does not matter, because the design
> >needs to be revised.
>
> Any guidelines / requirements / constraints so that we can start thinking ??
It's the usual:
- The Postfix client side must be suitable for a hostile environment,
so it must recover when the server
- disconnects unexpectedly (like while the client sends a request),
- sends incorrectly formatted responses,
- sends overly long responses,
- does not respond within some time limit,
- etcetera
- Queries and replies must be easy to parse and generate with
scripting languages like PERL.
- The encoding of queries and replies must be unambiguous: don't
use newlines as terminators if there is any chance that such
characters will be embedded in the data.
- I want to use a format where queries and replies are sequences
of named attributes. This way the protocol is extensible and can
be also be used for future multi-parameter queries and whatnot.
I was thinking of using an attr_print64/scan64() like format. This
uses base64 as the encoding. However, the present attr_print64/scan64()
implementation still needs to be hardened before it can be used in a
hostile environment. See the above requirements list.
Oh, and Liviu Daia would like to have a version that can talk UDP.
> As always, thanks for a wonderful program.
You're welcome.
Wietse
|
|
|