The language-ninja package

[Tags:apache, library, program, test]

A Haskell library for parsing, pretty-printing, and compiling the Ninja build language. The best place to start reading this documentation is the Language.Ninja module.

[Skip to Readme]


Versions 0.1.0, 0.2.0
Change log
Dependencies aeson (>=1.0 && <1.3), aeson-pretty (==0.8.*), base (>=4.8 && <5.0), bytestring (==0.10.*), containers (==0.5.*), deepseq (==1.4.*), flow (==1.0.*), hashable (==1.2.*), intern (==0.9.*), language-ninja, lens (==4.15.*), megaparsec (==5.2.*), mtl (==2.2.*), optparse-generic (==1.1.*), QuickCheck (==2.9.*), semigroups (==0.18.*), smallcheck (==1.1.*), system-filepath (==0.4.*), text (==1.2.*), transformers (>=0.4 && <0.6), unordered-containers (==0.2.*), versions (>=3.1.1 && <3.2) [details]
License Apache-2.0
Copyright Copyright 2017 Awake Security
Author Awake Security
Category Build, Language
Home page
Bug tracker
Source repository head: git clone -b master
this: git clone release-0.2.0)
Uploaded Wed Aug 2 00:16:29 UTC 2017 by taktoa
Distributions NixOS:0.2.0
Downloads 72 total (47 in the last 30 days)
0 []
Status Docs uploaded by user
Build status unknown [no reports yet]
Hackage Matrix CI




Maintainer's Corner

For package maintainers and hackage trustees

Readme for language-ninja

Readme for language-ninja-0.2.0


Hackage Stackage License

language-ninja is a Haskell library for parsing, pretty-printing, and compiling the Ninja build language.


language-ninja was written as a part of Awake Security's efforts in creating incremental build infrastructure for the Nix package manager, and in particular for Haskell packages. This library is the basis for a tool, ninja2nix, that, given a Ninja file, will output a JSON file containing information that a Nix function can use to compute a derivation representing an incremental build using one derivation per build edge in the parsed Ninja build graph.

In conjunction with an as-of-yet unwritten tool, cabal2ninja, that will generate a Ninja file based on information from cabal and ghc -M, we will have incremental builds for any Haskell package that uses a default Setup.hs file.

Using Ninja as an intermediate representation has advantages here, since it means that ninja2nix could be useful for building other projects that use a build system that can output Ninja, like the Linux kernel (with kninja), Chromium, any CMake project, any Bazel project, or some Make-based projects (with kati).


Originally, language-ninja used the Ninja lexer/parser from Neil Mitchell's shake project. However, we now use a megaparsec-based lexer and a highly modified version of the old parser. There is still work to be done on improving the diagnostic data and pretty-printability of the AST from this parser. In my view, it is very important that a parser output an AST that can be pretty-printed exactly to same sequence of bytes that were in the parsed file, as this makes tests and diagnostics much easier to write.


Currently there is a rudimentary pretty-printer in Language.Ninja.Pretty. It simply outputs text such that if that text is parsed and the parsed data is pretty-printed again, the resulting text will be identical to the original text (this is tested on a variety of Ninja files in the test suite).


The Ninja type from Language.Ninja.IR.Ninja contains precisely the data that must be acted on dynamically in a Ninja. In converting from the parsed AST to the intermediate representation using Language.Ninja.Compile.compile, you are eliminating all statically-dischargeable language features in Ninja, like variables. It also "monomorphizes" Ninja rules, since rule-level $out references are a kind of parametric polymorphism. This IR is thus far more suitable for processing than the original parsed AST.


The staging branch is used for active development; i.e.: force pushes may happen on staging but will not happen on master.

To build this, I recommend installing the Nix package manager and running nix-build release.nix -A language-ninja.