Haskell implementation of the nix store API
Go to file
Anton Latukha 69174725ac
CI: GitHub: add Optional-Nix-dev-env-main: add mvp nix-build (#95)
Work towards having enough CI testing #67.

I currently can not create the Cachix binary cache current for the project, details in #96.

So for the time being added the simple GH cache to them.
2020-12-22 21:02:19 +02:00
.github/workflows CI: GitHub: add Optional-Nix-dev-env-main: add mvp nix-build (#95) 2020-12-22 21:02:19 +02:00
hnix-store-core readme: sync doc links to source 2020-12-19 13:47:40 +02:00
hnix-store-remote remote: cosmetic 2020-12-17 13:06:26 +02:00
.envrc Lorri + direnv. 2019-03-19 20:52:25 -04:00
.gitignore cabal new gitignores. 2019-03-19 20:44:29 -04:00
cabal.project add cabal.project 2018-07-16 09:12:34 +02:00
default.nix Use constant-space encoding and decoding for NARs 2020-08-05 21:00:52 -04:00
LICENSE Copyright 2018-04-24 09:34:16 -04:00
overlay.nix Use constant-space encoding and decoding for NARs 2020-08-05 21:00:52 -04:00
README.md README: Elaborate on project goals and layout. 2018-05-13 21:18:49 -04:00
shell.nix Build expressions. 2019-03-09 18:32:01 -05:00

hnix-store

A Haskell interface to the Nix store.

Rationale

Nix can conceptually be broken up into two layers, both (perhaps unfortunately) named "Nix": The expression language and the store. The semantics of the expression language fundamentally depend on the store, but the store is independent of the language. The store semantics provide the basic building blocks of Nix: content-addressed files and directories, the drv file format and the semantics for building drvs, tracking references of store paths, copying files between stores (or to/from caches), distributed builds, etc.

The goal of hnix-store is to provide a Haskell interface to the Nix store semantics, as well as various implementations of that interface. Though the current primary client is hnix, an effort to reimplement the Nix expression language in Haskell, this project is meant to be generic and could be used for a number of other cases of interaction with the Nix store (e.g. a shake backend that emitted each build action as a store derivation). Currently, there are three implementations planned:

  • A mock store which performs no IO whatsoever, for unit testing.
  • A readonly store, which defers to another implementation for readonly effects (such as querying whether some path is valid in the store, or reading a file) but performs mutating effects in-memory only (for example, computing the store path a given directory would live at if added to the store, without actually modifying anything).
  • A daemon store, which implements the client side of the Nix daemon Unix domain socket protocol, allowing full interaction with the store on a system with the C++ daemon installed.

While this project is in the early stages of development, the daemon store can be seen as something of a road map: We want to express and implement all of (and only) the useful functionality available to a client of the existing daemon protocol.

Note that there are currently no plans for hnix-store to include an implementation which directly mutates the filesystem and database of the Nix store.

Packages

In the interest of separating concerns, this project is split into several Haskell packages. The current expectation is at least one package, hnix-store-core, containing the core effect types and fundamental operations combining them, agnostic to any particular effectful implementation (e.g. in-memory, talking to the Nix daemon in IO, etc.), with the actual implementations in a different package. Whether each implementation gets its own package or not remains to be seen.

The intent is that core business logic for a project that needs to interact with the Nix store can simply depend on hnix-store-core, and only at the very edges of the system would it be necessary to bring in a specific implementation.