IdrisExtSTGCodegen VS typelevel-rewrite-rules

In practice we will probably have an STG backend for Idris2 before a dependent Haskell, then at least you could link together Haskell and Idris code

I write an Idris2 backend which compiles to STG, yesterday we achieved to run the HelloWorld using functions from GHC.Base :) https://github.com/andorp/IdrisExtSTGCodegen/commit/0150510a7d6160806a85d799768d39f9acc65d30#diff-a15cf9e0e0625f8260bb7b91a4f4ca8e4f4558acf37d4cc834acd2e8a4cdf89aR6

There is a WIP Idris2 to Ext-STG compiler: https://github.com/andorp/IdrisExtSTGCodegen

Ah, the repo also has the slides of the talk, which are a nice first step to video-less content.

The github repo describes the project as "Pre-alpha, compiles a HelloWorld".

| The Idris language versions have always supported easy and modular code generation. I am working on the Idris-ExtSTG backend closely collaborating with Csaba. The progress can be followed here: https://github.com/andorp/IdrisExtSTGCodegen . I am going to give a talk about my experiences at BobKonf 2021: https://bobkonf.de/2021/penzes.html

Nice! Thanks for including a comparison with my package typelevel-rewrite-rules. Since my package indeed struggles with infinite loops introduced by self-triggering rewrite rules, I would like to better understand why your package doesn't suffer from that same problem.

When writing a typechecker plugin, you can eliminate constraints from the user's program by providing an implementation of the corresponding dictionary. The way you provide that dictionary to ghc is by providing a core expression; for example, this evCast futureDict co expression has type EvTerm, whose first constructor takes an EvExpr, which is a synonym for CoreExpr.

Until then, I guess we can simply poll. Here's a GitHub Action I wrote today which checks if the latest report for a given package includes a failure: https://github.com/gelisam/typelevel-rewrite-rules/blob/main/.github/workflows/check-hackage-matrix.yml