[Bug]: The order of type definitions influences mapping subtype relation about ballerina-lang

[lochana-chathura]

The BDD for diff(M1, M2) is as follows.

• For the failing cases,

(map<any> & map<any> & !map<anydata>) | (!map<any> & map<anydata> & !map<any> & !map<anydata>)
= (map<any> & !map<anydata>) | (map<anydata>  & !map<any>)
= map<any> & !map<anydata> | never
= not empty

• For the passing cases,
  The tree is same with map<any> and map<anydata> types interchanged.

(map<anydata> & map<anydata> & !map<any>) | (!map<anydata> & map<any> & !map<anydata> & !map<any>)
= (map<anydata> & !map<any>) | (map<any> & !map<any>)
= never | never 
= empty

[lochana-chathura]

Checked in nBallerina, there the diff(M1,M2) always be never. In our case, BDDs of M1 and M2 are not identical in terms of the indexes. (We have two map<any>/map<anydata> atoms with different indexes) However, in nBallerina we have identical BDDs (Same index for all map<any>/map<anydata>) thus the diff() results in never.

[lochana-chathura]

Overriding MappingAtomicType and ComplexSemTypes' equals() and hashCode() solved the issue for the original sample. However the following still fails as deep equality is not considered in subtypeDataList.

type M map<(1|2|3)>;
type N map<(1|2|3)>;

// @type M1 = M2
type M1 M|map<(1|2)>;

type M2 N|map<(1|2)>;

[lochana-chathura]

I was able to reproduce the same scenario(for the original sample) in nBallerina by modifying mappingAtom() as follows.

   isolated function mappingAtom(MappingAtomicType atomicType) returns TypeAtom {
       lock {
           TypeAtom? ta = self.atomTable[atomicType];
           if ta != () {
               TypeAtom result = { index: self.atomTable.length(), atomicType };
               return result;
           }
           else {
               TypeAtom result = { index: self.atomTable.length(), atomicType };
               self.atomTable.add(result);
               return result;
           }
       }
   }

The BDD for diff(M1,M2) I get there is different and it resulted in the correct emptiness. So there is a bug in diff() in jBallerina.

[lochana-chathura]

The root cause was found to be a bug in mappingInhabited() when porting. Both nBallerina and jBallerina produce the same BDDs. The previous BDD explained here turned out to be a mistake.

[lochana-chathura]

Fixed in PR #42791