cl-token:
- Minimize data redundancy by consolidating TokenKind::Literal; TokenData::{String, Identifier}
- Rename Op to Punct
cl-ast:
- Remove ExprKind::{Member, Call} in favor of making them
'binary' operators
- Consolidate boxes (TODO: consolidate more boxes)
- Remove repetition vecs in favor of boxes (this may come with performance tradeoffs!)
cl-lexer:
- Reflect changes from cl-token
cl-interpret, cl-repl/src/examples:
- Reflect changes from cl-ast
cl-parser:
- Switch to Pratt parsing for expressions
- TODO: Code cleanup
- TODO: Use total ordering for Precedence instead of binding powers (that's what the binding powers are there for anyway)
- Switch functional parsers to take Punct instead of TokenKind
- It's not like we need a `for`-separated list
- Remove `binary` macro. No longer needed with precedence climbing.
- Repurpose `operator` macro to produce both the operator and the respective Precedence
- Remove several of the smaller parser functions, since they've been consolidated into the larger `exprkind`
- cl-typeck: Add modules, intrinsic types, unify definition ids
- cl-ast: make attribute lists `Default`
- cl-structures: Add functions to iterate through a pool
- cl-repl: Create an example REPL for the type checker
- Refactor for cl-structures intern pool
- Create a list of types the compiler is supposed to care about/have implementations for (Intrinsic/primitive types)
- Add modules and projects (the sym equivalent of ast::File)
- Flesh out value definitions
- TODO: Create an IR for statements and expressions, and lower the AST into it
My editor's performance was tanking because of macro interpreter::builtins::builtin!
Temporary solution: move the interpreter into a separate crate
If I intended to keep the interpreter around, in the long-term, it might be an idea to make a proc-macro for builtin expansion.
However, the only reason I need the macros is because the interpreter's dynamic typing implementation is so half-baked. After I bang out the new type checker/inference engine, I'll have to rewrite the entire interpreter anyway!
- Allowed builtins to self-describe
- Broke builtins into their own module
- Created a macro to work with BuiltIns easier
- Uses macro 2.0 syntax, so it requires passing in ALL externally referenced identifiers
- Luckily, this is already a requirement of good, readable macro definitions!
- As a temporary hack, turn overloadable operators into function calls
- This is kind of pointless at the moment, since there can only be one definition of a function per name (no ADL/function overloading/traits/type namespaces yet!)
- This is also pretty slow, but benchmarking shows it's not as slow as I thought (~400x slower in release mode than a native Rust implementation when running `fib.cl`/`fib.rs`. Totally unacceptable for most work, but this is a tree walk interpreter.)
- TODO: Remove this when desugaring from operators to function calls is implemented
