RLMeta

RLMeta is a programming language in which you write grammars. Grammars have rules that specify how to match objects from an input stream and specify what should happen when objects are matched. It can be used to write lexers, parsers, tree transformers, code generators, and similar tools. The RLMeta compiler is implemented in RLMeta itself which makes it a metacompiler.

This repository will host the base implementation of RLMeta as well as modifications to it.

There are multiple other versions of RLMeta that all live in blog posts. Find them in category #rlmeta

Ideals

• How small, understandable can we make the core?

• Clarity: How does it affect understandability/learnability/readability?

• Size: Lines of code.

• Flexibility: How easy is it to modify RLMeta to be what you need?

• Performance: How fast does it compile?

Ideas

• LISP style semantic actions

• Compile semantic actions to more efficient code

  • Should not "Lookup" match variables more than once

• Memoization as add-on

• Left associate operators as add-on

• Smaller diff in modification -> large flexibility

• Remove indent/format syntax from RLMeta?

• Call ast something else in assembler? Tree? Code?

• Use LISP style function calls and support operators and function names (would make QOI more readable)

• Let LISP syntax instead of multiple -> in semantic actions?

TODO

• params compile_chain

• Can fail_pos be handled better?

  • MatchError probably has wrong stream if error occurs deep down

• Can support library (and new Runtime) become smaller?

• Substream matching should return substream as action

• What should an empty ["And"] return? Undefined or default None?

• ? operator should return empty list (in case of no match) or list with one item (in case of match).

• Why not better error message when action wrong? Why index wrong?

  Action        = .:xs
  Action        = .*:xs
  

  • Wrong pos is reported for "Not" instruction.

  • "Not" messes up latest_fail_pos in general. It should perhaps be disabled during a "Not"?

• Poster with intermediate versions (output from parser, etc.) shown.

  • Interactive on the web. (Requires JS version.)

  • Add DEBUG flag that outputs source between passes.

• Try to port to JS to see how flexible it is?

• Rename ast to tree?

• Lookup concat/splice/join/indent?