Functional Wrappers, Bridge Methods, Type descriptor fix
Functional interface wrappers should work now
Implementing a method with a signature differing from the actual one (e.g. when dealing with generics) now creates a proper bridge method
The Java Bytecde compiler used an incorrect method in checkCast
We now no longer use hardcoded 'accept's as method names but whatever the context tells us to.
- Fixed a number of bugs
- Upgrading build system to a more uniform system
- Improving error output (for IDE)
- Changed parser to generate line offsets differently (easier for IDE to process)
- Fix auto getters & setters for static fields not being static
- Fix getters for globals not having their corresponding fields registered correctly in the native implementation
- Added java sources for JavaScripting
- Fix crash when performing access checks between script and class
- Fix default constructor not working properly
- Fix integration classes without exposed constructor being instantiable if they have a default (non-exposed) constructor
- Fix extern methods failing validation
Revamped the compilation system; it now works fully modular. Also, a new ScriptingEngine class has been created to make it very easy to create a scripting environment.
Improved handling of functions, they now use the most generic variant available, using standard function types where possible. (TODO: there are still more to add)
- Fix missing shift operators in some integer types
- Fix some problems when comparing a long with an int or ulong with uint
- Add signed-unsigned implicit array conversion (sbyte[] <-> byte[], short[] <-> ushort[], int[] <-> uint[], long[] <-> ulong[])
- Fix bug when copying multiple lines from the source editor
- Added string <-> byte[] conversion operations
- Fix unsigned arrays, use compact representations
- Added suffixes & custom suffixes
- Added 0x 0b 0o for hex, binary and octal notation
- Ranges now only have a single base type; from and to must be the same type
- Adjusted builtins and stdlib to use usize where applicable
- Moved building a SemanticModule from ParsedFile to the parser; reduces dependencies
- Fixed ranges, all range types should be supported now
- Cleaned up unused code
- Refactored the JavaMethodInfo class to the shared JavaMethod one
- Use valueOf for boxing instead of the boxed type constructors
- Fixed numerous issues with undetermined types, type inference, type aliases, inner types and generic types
- Distinguished between type resolution and code compilation stages, both have a different context now
- Added new concepts: "compiling type" and "compiling package", which is able to precompile correctly now
- Rewrote the way precompilation works; precompilation no longer exists, instead, compilation order of methods is changed. This reduces the amount of code needed and improves quality of the system.
- Fixed members of base interfaces not being available to implementation code
- Changed the way type modifiers are coded
- Changed the way modified types are processed
- Added immutable keyword
- Added expression, statement and type normalization
- Improved naming of synthetic functions
- Fixed crash formatting scripts without script statements
- Fixed bug causing lambda headers not to be inferred anymore
- Refactored some IDE code, also fixes scrolling not working properly
- Added support for implementations and method overriding
- Added override method header inference: if a method overrides or implements another method, and types are missing from the header, these will be inherited from the base method. This works with any kind of method (method, operator, getter, setter, caller)
- Added method return type inference, which is automatically inferred from the return value of the contents of the method. This functionality is not yet fully implemented, and will fail with certain expressions (most notably, member expressions and calls). The goal is to make this system work as long as there is no recursion. This works with methods, operators, getters, calls as well as fields and constants with initializers.
- Removed the any type. With return type inference its usefulness has become even more limited, and proper implementation of the any type is expected to be complicated, thus not worth the effort. (this may be revisited later on, as "any" remains a reserved keyword)
- Fixed compilation of throw as panic expressions in lambda function bodies
- Moved name from FunctionalMember to MethodMember, where it makes more sense
- Each method (or functional) is now also aware of the member that it overrides
kindlich
Stan Hebben