A Haskell interface to the Nix store.
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
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,
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
Nix expression language in Haskell, this project is meant to be
generic and could be used for a number of other cases of interaction
Nix store (e.g. a
shake backend that emitted each build
action as a store derivation). Currently, there are three
mockstore which performs no IO whatsoever, for unit testing.
readonlystore, 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).
daemonstore, which implements the client side of the
Nixdaemon 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
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
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
and only at the very edges of the system would it be necessary to
bring in a specific implementation.