On Mon, 13 Jan 2003, Wietse Venema wrote:

> I fail to see how prefixes for some parameters for some processes
> helps to document wich parameters are being used by which Postfix
> daemon processes. Sounds to me that we made things a lot more
> complicated.
>

Last post by me in this thread, promise... Anyway the idea would be to
prefix only parameters that currently have reasonable semantics when used
with "-o" options. "Reasonable" semantics mean: you get sensible *and*
useful behaviour when different processes have different values.

So hash_queue_depth does not get prefixed, but "smtp_connect_timeout"
does.

The set of prefixed parameters is precisely the set that work with "-o"
but the prefix obviates the need for "-o" (except possibly for an option
to set the prefix!).

The documentation would include a machine generated list of parameters
that are subject to "prefixed" lookup and the daemons that perform
prefixed lookups against the parameter in question.

I am afraid this is all too general. Just making some delivery agent
parameters transport name dependent would I think be both useful and
simple. I must admit that I can't predict whether it would increase
confusion.

How about the following scheme:

- Only parameters starting with "default_" support "<transport>_"
alternatives.

- The transport parameters which support prefixes are renamed:

        - default_smtp_connect_timeout
        - default_smtp_helo_timeout
        ...

- The "<transport>_" prefix replaces the "default_" prefix.

Again in case this is lost in the long discussion up til now, I am looking
to minimize "-o" use in master.cf and my remaining use cases are per
transport tuning parameters for delivery agents.

I am already using "-o smtp_mumble=$<transport>_mumble", so my settings
are in main.cf. The proposal would eliminate the need for this hook.

Over and out. Even if no change results from all the hot air in this
thread, perhaps at least we will know why the the unchanged design is the
best compromise.

-- 
	Viktor.

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]