A Haskell package for writing Emacs modules

Why would anyone want to write Emacs modules in Haskell?

Emacs Lisp is not a young language and can go quite a long way, but it has a couple of issues that are not going to be solved any time soon:

  • It’s dynamically typed which makes refactoring large extensions a pain

  • It’s intepreted and is quite slow. It might be argued that editors don’t need much computing power, but from time to time computation-intensive tasks do occur. For example, fuzzy matching provided by the cool flx.el package and used by great ivy.el package to quickly find things.

  • Somewhat related to the previous point, there’s virtually no support for parallelising computations. There’re adavances on adding threads to Emacs lisp, but this only provides concurrency, but no parallelism.

    Haskell is well known for solving points 1 and 3 outlined above. For me it also solves point 2 by providing enough performance and adding parallelism on top of it.

    If you think this might be a good idea and would like to see what this package can do for you, you can look at part of my emacs config that uses this package to implement things like

    • Rewrite of flx.el that leverages parallelism
    • Fast search across filesystem
    • Concurrrent grep reimplementation (somewhat dubious since things like ripgrep exist)

FAQ

How do I start writing my own extensions?

See tutorial at https://github.com/sergv/emacs-module/blob/master/Tutorial.md.

Also check out this package’s tests.

What about Windows?

It works, Cabal can build a dll for you.

How it’s related to haskell-emacs?

The haskell-emacs aims to address the same problem - writing Emacs extensions in Haskell, but uses different approach. It seems to use some kind of marshalling scheme to make Emacs data available in Haskell with a caveat that not all Emacs types can be converted (e.g. buffers cannot be typically serialised). Presumably, an extension built with this project will look like an executable that reads sexps from stdin and produces output on stdout. Or, possibly, as a daemon process that communicates with Emacs over network.

This project is a bit different. It wraps Emacs C API for writing new extensions that can manipulate Emacs values directly, without marhsalling. In this approach, an extension will look like a shared library/dll that can be loaded by standard emacs with (load "/tmp/libmy-ext.so").

Supported GHC versions

Tested with GHC 9.2, 9.4, 9.6, 9.8.

Changes

0.2.1

  • Expose make_unibyte_string as makeBinaryString

0.2

  • Major rework of the package’s core

  • Bump minimum required Emacs version to 28 due to exposing process_input from API. It allows to check whether user wants to abort and control should be returned back to Emacs ASAP

  • Bump minimum required base version to 4.14 - minimum supported GHC is 8.10

  • Symbols can be either statically known or dynamically known. This is mostly an optimization that doesn’t affect symbol use but client code may require updating. Statically known symbol are just pointers to statically allocated (by GHC) bytes and the pointers are simply passed to Emacs to create symbols.

  • Some commonly used symbols are now cached on first use so they won’t be re-interned on subsequent uses. New symbols can be defined by users.

  • makeFunction now cleans up after itself and no longer has memory leak if called a lot of times (which shouldn’t have typically happened anyway but still nice to have just in case)

  • makeFunctionExtra and EmacsFunctionExtra are gone. They offered to pass extra pointer into subroutine exposed to Emacs but it was never needed since arbitrary closure can be exposed. Now extra pointer is used to clean up result of makeFunction.

  • Removed extractUnboxedVectorWith - now regular extractVector produces generic vectors

  • Removed extractVector - extractVectorWith and similar for other vector/array types are preferable since they’re guaranteed to not create intermediate vectors/arrays

  • Introduce dedicated Doc type for function documentation that can be constructed from unboxed string literals to just pass the pointer around

  • produceRef and EmacsReturn are gone - vanilla pure and EmacsRef are enough

  • Symbol names got IsString instance and can be defined as string constants

  • Removed UserPtrFinaliser and UserPtrFinaliserType. Use Foreign.ForeignPtr.FinalizerPtr instead

  • GlobalRef got removed, instead RawValue is indexed by whether it’s going to be GC’ed after returning control back to Emacs

  • funcall and funcallPrimitive now accept any emacs value function rather than symbols only

  • extractString removed, extractText is now part of MonadEmacs typeclass

0.1.1.1

  • Fix build with GHC 9.0+
  • Switch license from BSD 3 to Apache 2.0
  • Bump minimum required GHC to 8.10
  • Bump minimum prettyprinter to 1.7