Re: [exim-dev] parsing costs of acl config

On 2013-07-07 at 16:11 +0100, Jeremy Harris wrote:
> Does anyone know if the saving of parse time is significant?

Changing it will impact all the deliveries and re-exec there. Frankly,
given that the file will be permanently in buffer-cache on any
mail-server (and if not, performance isn't an issue) and all that
happens is parsing into in-memory structures, I'd be surprised if it's
that significant.

> I could put in a lazy evaluation if needed, or merely remove
> the current "skip".

In an ideal world, the "skip" would set some offset pointers to enable
later re-reading, then skip as normal, and ${acl} expansion would detect
those being set and determine that the ACL section then needs to be
ready. If that's the implementation you're thinking of for "lazy
evaluation", then yes.

In practice: your work, your call, benchmarking would be good but not
entirely practical for most of us at this point.

Todd is working on some Vagrant stuff that might change that and make it
easier to benchmark?

-Phil