CR typechecking options are now application-level.
The solution j.m.c.ideaplugin contains the code responsible
for accessing and setting the options. This solution is loaded
as IDEA plugin on deploy and as regular module in source project.
Also merge code from two modules into j.m.c.typechecking.
The templates that typechecking relies upon are generated
into the user model as synthetic constructs. These later
need to be discovered by the generator to establish
correct references.
Request is no longer dependent on `TypecheckingService`.
Only single request can be processed at a time, which helps
prevent recursion.
All facade method in `TypecheckingService` now return instance
of `Response` instead.
Drop hacks that try to figure out the active project.
Instead introduce new ones that cache the instance of `ProjectRepository`
(unfortunately, there is no way around that now).
Stop accessing and modifying the singleton `Config` instance,
instead make an instance of `Config` a parameter to `TypecheckingService`.
Rule's unique_tag is important for update procedure
implemented by RulesPlanHandler. Only when we can guarantee
that an argument to constraint/predicate has not changed
we can save on re-building a rule from template.
Ignoring LateExpression was a mistake.
While check/checkTypeOf queries are required to
process a whole root, other types of queries
can be sensitive to what aspects to include in
the processing.
Aspects are sorted in strongly-connected components,
among which the relevant one is selected.
The deisgn decision to have only one instance of QueryTemplate
leads to the necessity to keep AspectClique separated for
every "typesystem". Whereas QueryTemplate are really perfectly
composable, and thus only one AspectClique is neeeded.
Also drop a rudimentary unused feature.
Fedor Isakov