improve behaviour of inheritance

[quentin]

Hi, this is a change to improve behavior of components inheritance.

I added some tests.

Instantiate inherited component

It used to be impossible to instantiate components declared by an inherited base:

.comp Base {
  .comp Nested {
  }
}
.comp Derived : Base {
  .init n = Nested
}
============
Error: Referencing undefined component Nested in file foo.dl at line 6
    .init n = Nested
----^----------------

Clauses from base's nested instances

Clauses within nested instances of base components were be missing:

.comp Base {
  .decl rel(s:symbol)
  .comp Nested {
    rel("Nested").
  }
  .init n = Nested
  rel("Base").
}
.init b = Base
---------------
b.rel
s
===============
Base
Nested <------------------ expected
===============

But the clause from Nested would disapear when inherited:

.comp Derived : Base {
  rel("Derived").
}
.init d = Derived
---------------
d.rel
s
===============
Base
<----------------- missing "Nested"
Derived  
===============

Open question: overriding and clauses from nested instance of base

As a side-effect of fixing previous point, I allowed clauses from nested instance of base component to be inherited in the derived component:

.comp Base {
  .decl rel(s:symbol) overridable
  .comp Nested {
    rel("Nested"). // <------ not excluded by Derived .override directive
  }
  .init n = Nested
  rel("Base"). <----- excluded by Derived .override directive
}

.comp Derived : Base {
  .override rel
  rel("Derived").
}
.init d = Derived
---------------
d.rel
s
===============
Nested  <--------------- do we want this entry ?
Derived
===============

Arguments in favor

The rationale for this behavior is that we may want to expect the same result whether .init n = Nested is located in Base or Derived:

.comp Base {
  .decl rel(s:symbol) overridable
  .comp Nested {
    rel("Nested").
  }
  rel("Base").
}

.comp Derived : Base {
  .override rel
  .init n = Nested // <---- now instantiated from Derived
  rel("Derived").
}
.init d = Derived
---------------
d.rel
s
===============
Nested  <--------------- do we want this entry ?
Derived
===============

The second argument is that we get the same behavior when the clause is "orphan" (relation is not visible in a lexical lookup):

.comp Base {
  .comp Nested {
    rel("Nested"). // <------- "orphan" clause
  }
}

.comp Derived : Base {
  .decl rel(s:symbol)
  .init n = Nested
  rel("Derived").
}
.init d = Derived
---------------
d.rel
s
===============
Nested  <--------------- expected
Derived
===============

Opposite argument

The rationale for the opposite behavior would be that a clause relates in priority to the visible relation declaration in a lexical lookup (when one can be found). Hence if there is an override applying to that relation declaration, the clause should be discarded.

Implementing this behavior would be a bit more complicated.

[codecov]

Codecov Report

Merging #2220 (bc8d8dc) into master (6eb3ad0) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #2220   +/-   ##
=======================================
  Coverage   75.25%   75.26%           
=======================================
  Files         454      454           
  Lines       27572    27587   +15     
=======================================
+ Hits        20750    20764   +14     
- Misses       6822     6823    +1     

Impacted Files	Coverage Δ
src/ast/analysis/ComponentLookup.h	100.00% <ø> (ø)
src/ast/analysis/ComponentLookup.cpp	95.45% <100.00%> (+2.12%)	⬆️
src/ast/analysis/typesystem/TypeConstraints.cpp	90.11% <100.00%> (ø)
src/ast/transform/ComponentInstantiation.cpp	92.39% <100.00%> (+0.02%)	⬆️
src/include/souffle/SouffleInterface.h	88.99% <100.00%> (ø)
...ouffle/datastructure/ConcurrentInsertOnlyHashMap.h	81.30% <0.00%> (-1.63%)	⬇️
src/include/souffle/utility/ParallelUtil.h	83.72% <0.00%> (-1.56%)	⬇️
src/include/souffle/datastructure/LambdaBTree.h	74.04% <0.00%> (+0.76%)	⬆️
src/include/souffle/datastructure/UnionFind.h	97.80% <0.00%> (+2.19%)	⬆️

[b-scholz]

Is somebody against the proposal?

Could you give a practical example that made you change the component model?

[b-scholz]

There are examples as part of the PR. Thanks for this.