Takeaways

We have now all the conceptual ingredients required for complex typed languages:

• management of declaration contexts and scoped variables

• type checking using a postfix intermediate language

In the Haskell example, the type inference was performed directly in the parser, but it is possible instead to call upon a language server in order to tackle more complex languages (like dependent types).

The intermediate postfix language may look like a trick, but I reckon there is a deeper meaning to it. Prefix and postfix forms are both depth-first walks of a same inference tree. The choice to use prefix form is only – I believe – a question of human-readability. So it may be possible to build a generic tool around these inference systems that transform grammars and languages back and forth between easy-to-read and easy-to-generate, thus avoiding the task of defining manually new grammars and tree transformation.

Having this kind of tools for existing languages also means being able to transform entire codebases, and thus fine-tune a LLM on them.

The features showcased here a very much a work in progress. Some features are lacking; for instance contextual terminal symbols could lead the generation in a deadend if not used carefully, so the grammar builder should detect and warn of such cases.