'typeVariableDeclaration' now returns substituted vars or creates new subst.
Earlier it created created a new susbst which led to infinite recursion in specific cases.
Now macro expects tvdecl to be declared in TypeVariableSubstitution.
Accordingly modify usages of TypeVariableSubstitution.
Also fix (apparently broken?) SubclassPaths rulesList (used in lub relation)
and switch 'dpromote' implementation to use it
Remove duplicated processing of auxiliary bounds in type var declarations.
TypeVariableDeclaration macro now plays a role of instantiating type variables.
hasBound constraint is produced more consistently, on any expansion of TypeVariableDeclaration.
It will work mainly through reactivation due to changes to its arguments.
Introduce a reactive scheme: alpha node has "dependent" nodes which
it can invalidate transitively. Drop block initiates invalidation,
intro block actually drops the nodes.
Instead of just reporting the times each probe has been called,
collect and calculate the actual time taken by each probe location from
the whole backtrace pool.
The tests presumably fail because of a broken contract that is used
in the typesystem implementation, namey that a logical variable
contained by an constraint argument-term (as opposed to being an
argument itself) reactivates the constraint on ground/parent change
event. This assumption is false, thereby the implementation needs
to be adjusted keep compatibility with the contract.
Drop layer prototype, make layer mutable.
Keep nodes in a layer in a linked list with fail-safe iterator.
Introduce update queue for update blocks.
Update blocks run incrementally unless reset.
Maintain trail of introduced and dropped occurrences.
Generation iterates over nodes in the last (final) layer incrementally,
unless reset.
Introduce new test.
All test data (code) to be contained in j.m.blExt.test module.
The samples are to be kept separately with generation on to demonstrate fewer false negatives
of the new typechecker.
for historical documentation purposes, why it was needed:
It was used for specifying and handling rules which must be simply removed
without all related incremental invalidation machinery.
Long time ago some of these dependencies weren't captured
and such "simply dependent" rules must have been handled separately.
Currently it's handled by general invalidation case.
Grigorii Kirgizov