Skip to content

Latest commit

 

History

History
59 lines (40 loc) · 2.22 KB

README.md

File metadata and controls

59 lines (40 loc) · 2.22 KB

hpython

Haskell-based language tools for Python

CSIRO's Data61 Logo

hpython provides an abstract syntax tree for Python 3.5, along with a parser, printer, and syntax checker. It also contains optics for working with the AST, and a DSL for writing Python programs directly in Haskell.

Features

  • Formatting-preserving syntax tree
  • Parser and printer, with a round-trip law: print ∘ parse ≡ id
  • Optics for manipulating the syntax tree
  • Indentation, syntax, and scope checking
  • The syntax tree is indexed by its level of validation, to distinguish between syntactically valid Python and unvalidated code
  • Convenient DSL for building Python programs

Examples

See the example directory

FAQ

Why not just use language-python?

There are two main reasons:

  1. We think the print-parse identity is important. language-python discards lexical information like indentation levels and spacing, which means there is big cost to using it to modify human-written code. hpython retains formatting information in a way that has minimal impact when using the library. This means that program transformations change as little formatting as possible.

  2. We want to use types to precisely model the domain. language-python unifies Python 2 and 3 into a single data structure. We disagree with this choice, because Python 2 and 3 have different, non-compatible features. In Haskell terms, they are different datatypes with a large amount of overlap. Our goal is to make this difference visible in the type system without increasing code repetition.

    There are other minor places where language-python has made similar concessions, like in the treatment of 'starred expressions' (which are not really expressions at all).

Development Pipeline

  • Support other versions of Python while re-using as much common code as possible
  • Style configs for the DSL
  • Human readable validation errors, with source spans

Contribution

Feel free to file an issue or pull request on Github, or contact us at:

IRC - #qfpl on Freenode Email - <oᴉ˙ldɟb@llǝʞsɐɥ>