Add OR to exception in evaluatable checker

[kasimebrahim]

I needed this in asmoses when creating representation.

[linas]

Why?

AndLink is treated differentlty due to a historical mistake; I'm in the process of slowly erasing that mistake. I do not want to spread that mistake to OrLink.

I'm guessing that perhaps you should be using ChoiceLink which is a non-boolean-logic Or useful for presenting a menu of choices?

[kasimebrahim]

Yeah so do you remember the representation atomeses we were talking about on #2546?
some thing like

 
(define the-expr
  (AndLink
    (PredicateNode "$a")
    (CondLink
      (EqualLink (Variable "$knob-2") (NumberNode 1))
      (PredicateNode "$b")
      (NotLink (PredicateNode "$b")))))

I needed it to do the same when we have OrLink inplace of AndLink.

[linas]

I think what you need is to add

+++ b/opencog/atoms/core/Checkers.cc
@@ -56,6 +56,7 @@ bool check_evaluatable(const Handle& bool_atom)
                if (VARIABLE_NODE == t) continue;
                if (GLOB_NODE == t) continue;
                if (DEFINED_PREDICATE_NODE == t) continue;
+               if (COND_LINK == t) continue;
 
                // Allow conjunction, disjunction and negation of
                // predicates. Since it cannot inherit from EVALUATABLE_LINK

I haven't tested, but I think that's the actual issue you are facing

[kasimebrahim]

One more thing the CondLink is not actually CondLink, it is a KnobLink[gets converted to CondLink when execution though] which contains more information than CondLink when constructing representation.
example of an actual rep will be.

(OrLink
  (KnobLink
    (VariableNode "$knob-376fd81c")
    (PredicateNode "$2")
    (NumberNode "0 1 2")
    (FalseLink ) 
    (FalseLink )))

But I could check if it is FunctionLink?

[linas]

Test the patch above, see if that fixes it for you. Also, the checker infrastructure should be updated to print a more informative error message: e.g "found a non-evaluatable atom XYZ in a link that will be evaluated".

[linas]

Ho about if (nameserver().isA(t, COND_LINK)) ? I want to keep this as narrow/strict as possible, at least for now.

There are several meta-issues that haven't been solved; one is that static type-checking is ... OK but imperfect, and that dynamic type-checking is equally hap-hazard. Both are kind of a kludge, but I want to avoid making them even kludgier by having too many exceptions. The correct, long-term "perfect" solution is not yet clear.

[linas]

I'm assuming KnobLink inherits from CondLink ...

[linas]

Oh, well, you could just also make make KobLink inherit from EvaluatableLink, and then no changes to atomspace source are needed. So if you have some asmoses_atoms.script file then just change

KNOB_LINK <- WHATEVER

to

KNOB_LINK <- WHATEVER, EVALUATABLE_LINK

and it will all just work.

[kasimebrahim]

Thanks that works.