Fork me on GitHub
#clara
<
2016-03-22
>
devn00:03:39

I guess I'm still interested in understanding exactly what's going on here

mikerod14:03:54

devn: did you have an example of what broke somewhere?

devn17:03:00

That's the situation, which I guess makes good sense to me.

devn17:03:39

We don't need to retract here. I pushed back against the use of it, but it worked, until we bumped to 0.11.0 of course simple_smile

mikerod20:03:27

devn: yeah, in that particular scenario the previous behavior was questionable

mikerod20:03:59

The RHS is retracting the facts that are supporting the truth of the rule in general

mikerod20:03:48

This goes against the concept of logical truth maintenance of inserted facts. Retracts will often lead you into bad situations when it comes to truth maintenance. In this particular scenario, the rule shouldn't insert anything because anything it inserts loses its logical support immediately, due to the retract of all the recommendations.

mikerod20:03:12

Looks like Will put more details on the Google groups though.

mikerod20:03:17

In our use-cases, we avoid retracts pretty much completely. We like to rely on the logical truth maintenance of inserts to do any "retracts" for us. It does involve adding new fact types occasionally, which Clara's default is via :type metadata or the Java class hierarchy (on Java side) or JS hierarchy (on JS side)