On Wed, 2004-03-10 at 15:46, Phil Pennock wrote:
> We're currently looking at some overhauls to our mail-system, for
> efficiency purposes. We have most of it fairly well designed and are
> approaching the stage where we'll start programming. One part of the
> changes involves a new extension to the appendfile transport, to be able
> to create an extra "flag" file with some data encoded in it. How likely
> is it that this sort of unusual feature would make it back into the
> upstream code-base? Will it be more or less favourably received if it
> uses #ifdef wrappers, given that it will preserve existing default
> behaviour and will require setting a transport option to activate?
I've got a mental note to start raising issues about compile time flags
- although I was thinking more in terms of binary distribution
packaging. I'll chip in a summary here since it has some relevance....
I would like to reduce as far as possible the amount of compile time
tailoring that is done, and move as much as possible to run time,
although with an eye on limitations of efficiency and dependencies etc.
That would mean that binary distribution sets would really need to have
some form of loadable module arrangement - ie you want mysql then you
load the exim_mysql loadable module in.
In this case, assuming the code does do something vaguely sensible (!),
then I would want it in the main build and selected by a config option
to the transport -- aside from anything else we have stated that these
things must be documented, and having the documentation strewn with
"This feature is only available if you set the BLURP_FURBLE_WHIZZ option
and compiled whilst the moon was full" is going to make things unclear.
Nigel.
--
[ Nigel Metheringham Nigel.Metheringham@??? ]
[ - Comments in this message are my own and not ITO opinion/policy - ]
