On Tue, Mar 01, 2011 at 02:07:25PM +0100, Michael Hanselmann wrote:
> Am 1. März 2011 11:32 schrieb Iustin Pop <[email protected]>:
> > On Tue, Mar 01, 2011 at 11:00:01AM +0100, Michael Hanselmann wrote:
> >> What do you think of such an implementation?
> >
> > If that's all what all LUs will do, then I fail to see why adding this
> 
> I assume you didn't really mean all LUs here. It'd just be the LUs
> generating other jobs (e.g. NodeEvacStrategy).

Yep.

> > line (as a code line) to all such LUs  is better than a declarative
> > approach, e.g. (whatever the name):
> >  self.generated_jobs = ops
> 
> I don't like code with side-effects where they can be avoided.

Hah, that's why you gave an example that specifically had side-effects?  ;)

> If you
> don't want the jobs to be submitted directly from the opcode, how
> about returning a list of opcodes from Exec?
> 
> If they're stored in a recognizable (as in isinstance()) container
> class, mcpu.Processor or a wrapper function can submit them.

That's also fine. Note we already have dry_run_result as a "side-effect"
(why you call it that I'm not sure).

thanks,
iustin

Reply via email to