zslajchrt / morpheus Goto Github PK

The current version of the alternatives iterator produces duplicate alternatives as long as the same fragment occurs in multiple disjunctive groups:

(A | B)*(A | B) produces 4 alternatives A, AB, B, AB, but it should produce A, AB, B.

A fragment can be marked as private in the morph model. In such a case it will be excluded when constructing the morph's LUB (lowest upper bound). Only the other fragments in the morph can invoke its members.

This issue is of low importance since it can be achieved by means of kernel references that can hide private fragments.

Two fragments may have contain unrelated members of the same name (i.e. not overriding). Since this ambiguity may lead to an unintended behavior, the compiler should reveal this issue and report it as an error.

Since AltMappings are generated as a string source in compile time, which must be parsed in runtime, it will be necessary to incorporate some caching into the AltMappings.apply(src: String), which is responsible for parsing the serialised form of generated AltMappings.

A fragment instance in a kernel can be specified as the identity fragment.

Two kernels have the same identity iff they are the same instances (i.e. in terms of eq) or contain the same identity fragment (i.e. in terms of ==).

Two morphs have the same identity iff their kernels have the same identity. In other words, a morph has no its identity.

Introduce new operator =*=, which compares the identities of two morphs or kernels.

Scopes will replace the singleton and frag factories.

It will automatically make possible to share hidden fragments between two morphs, the source and target ones.

Private members should not be enhanced, mainly because of performance reasons.

Sometimes types have no annotations loads during macro morph model parsing in compile time, although they have some.

There is a workaround: it helps if the morph model is parsed twice.

Dimension traits should support non-public members as well. Currently the generator ignores them and the generated dimension stub (Dim$dimension.class) do not contain any of them.

The parents of the generated fragment class (Frag$fragment.class) are not fully qualified. It results in compilation errors complaining about non-existence of types from the list of parents as long as these types belong to other package that that of the generated fragment class or no appropriate import statement is present.

A workaround: Import the whole package to which the parent fragments or dimensions belong to the source from which the fragment class is generated.