> The only disadvantage I see to this is one of performance. Exim starts
> up new processes and re-initializes itself frequently. Thus, it would be
> loading the modules and running their initializations very often. Maybe
> this wouldn't matter much. (The modules would remain in a cache, and
> their initialization could be made fast.)
>
I willingly admit that I am not an expert on this area, but I
believe that it should be possible to:
Have the 'main' exim process load an initialise the modules and
to use some form of shared memory/IPC schema to keep state. This
should work fine so long as you start up new processes with a
fork and an exec, I believe, it probably gets trickier if you use
a system() call.
This way you don't need to worry about hoping that the modules remain in
the cache, nor that the initialization has to be fast. On the downside
leaks particularly affect you, and you _may_ be susceptible to certain
forms of exploits.
Also, whilst I am here, it would be nice to see exim make use of
facilities provided by userv, I'd provide a URL, but Chiark appears
to be inaccessible from Cambridge at the moment. Ian Jackson can
provide details, or look somewhere under
http://www.chiark.greenend.org.uk/~ijackson/
for more details.
Julian (time to lurk again)
