Re: icp bitrot?

On Wed, 2002-10-09 at 07:54, Henrik Nordstrom wrote:
> Duane Wessels wrote:
>
> > > Actually, I'm creating async storeGetPublic* functions. This requires
> > > breaking every synchronous function that use storeGetPublic up into a
> > > start and completion routine. ICP uses storeGetPublic...
> >
> > What will this do to ICP performance?
>
> Probably not very nice things, but it should depend a lot on the type of
> store index used. The reason for the change has been discussed many
> times, and is to allow for on-disk cache indexes. When using a on-disk
> cache index we do not really have a choice but to have all
> storeGetPublic() calls async in nature..
>
> The response time should only increase marginally when using a in-core
> store index like today. We are then looking at a non-linear code flow
> with a callback in the middle rather than a standard call/return
> function call, but the same basic operations taking place in the same
> order with no interruptions (I suppose.. don't know exacly how Robert is
> implementing this).

I've been doing it my cxx test branch, experimenting with how callback
APIs might look.

It's currently implemented by simply calling the synchronous
storeGetPublic, and then calling back with the result.

The main point is to get the API change from sync to callback
implemented and tested.

Rob