This post answers why I created 'haskell-scope' even though there's already another library that addresses the same problem.
There are two libraries for resolving references in Haskell source code on Hackage: haskell-names and haskell-scope. Of the two, haskell-names is the oldest, the most feature complete, and the most ambitious. It uses a very innovative scheme that allows the scope to be inspected at any point in the syntax tree. You can read more about it in the linked article. Unfortunately, all this innovation comes at a price of complexity.
Here's the complete list of extensions used by haskell-names: CPP, ConstraintKinds, DefaultSignatures, DeriveDataTypeable, DeriveFoldable, DeriveFunctor, DeriveTraversable, FlexibleContexts, FlexibleInstances, FunctionalDependencies, GADTs, GeneralizedNewtypeDeriving, ImplicitParams, KindSignatures, MultiParamTypeClasses, NamedFieldPuns, OverlappingInstances, OverloadedStrings, RankNTypes, ScopedTypeVariables, StandaloneDeriving, TemplateHaskell, TupleSections, TypeFamilies, TypeOperators, UndecidableInstances, and ViewPatterns.
A total of 27 extensions and many of them will never be implemented by LHC. If LHC is to compile itself one day, this obviously won't do. Enter haskell-scope: a library more plain than bread without butter. Give it an AST and it will annotate all the references. Nothing more, nothing less.
No comments:
Post a Comment