rkchang / mlidk Goto Github PK

Currently the Lexer generates a LexerError from a std::exception, and the Parser generates a Parser::Error from std::runtime_error.

We should probably unify the error hierarchy under some "LettuceError" with common operations (e.g. source location, etc.) and stick with one way of generating errors (exception or runtime_error)

After source code is parsed, we need to emit MLIR operations for each Lettuce expression

Depends on operations being created: #12

Running build.sh without third-party throws error:

Python is listed as a dependency and also referred to by some CMake files, but we don't plan to use/support it.
Can be removed maybe?

Currently the build system is not aware of changes to flags for llvm-project. This can lead to subtle errors. Fully integrating llvm-project into the build system would prevent this.

Solutions:

• consider switching to Bazel? would be alot of work though
• use add_subdirectory()?
• probably add llvm-project as a git submodule

• currently we have no way of automatically testing the output of our compiler. llvm-lit should provide a way of doing that

Romaine dialect represents the high-level Lettuce language, we need to lower it to LLVM dialect (potentially using other built-in dialects) and eventually emit LLVM IR

Need to add MLIR Ops to Romaine dialect for:

• let-expressions
• binary expressions
• integer cosntants
• variable expressions