Crate miniconf

source ·
Expand description

§miniconf: serialize/deserialize/access reflection for trees docs QUARTIQ Matrix Chat Continuous Integration

miniconf enables lightweight (no_std/no alloc) serialization, deserialization, and access within a tree of heretogeneous types by keys.


See below for an example showing some of the features of the Tree* traits. See also the documentation and doctests of the TreeKey trait for a detailed description.

Note that the example below focuses on JSON and slash-separated paths while in fact any serde backend (or dyn Any trait objects) and many different Keys/Transcode providers are supported.

use serde::{Deserialize, Serialize};
use miniconf::{Error, JsonCoreSlash, JsonPath, Traversal, Tree, TreeKey, Path, Packed, Node};

#[derive(Deserialize, Serialize, Default)]
enum Either {

#[derive(Deserialize, Serialize, Default, Tree)]
struct Inner {
    a: i32,
    b: i32,

#[derive(Tree, Default)]
struct Settings {
    foo: bool,
    enum_: Either,
    struct_: Inner,
    array: [i32; 2],
    option: Option<i32>,

    // Skipping (`()` is not Deserialize/Serialize)
    skipped: (),

    struct_tree: Inner,
    array_tree: [i32; 2],
    array_tree2: [Inner; 2],

    option_tree: Option<i32>,
    option_tree2: Option<Inner>,
    array_option_tree: [Option<Inner>; 2],

let mut settings = Settings::default();

// Atomic updates by field name
settings.set_json("/foo", b"true")?;
assert_eq!(, true);
settings.set_json("/enum_", br#""Good""#)?;
settings.set_json("/struct_", br#"{"a": 3, "b": 3}"#)?;
settings.set_json("/array", b"[6, 6]")?;
settings.set_json("/option", b"12")?;
settings.set_json("/option", b"null")?;

// Exposing nodes of containers
// ... by field name in a struct
settings.set_json("/struct_tree/a", b"4")?;
// ... or by index in an array
settings.set_json("/array_tree/0", b"7")?;
// ... or by index and then struct field name
settings.set_json("/array_tree2/0/a", b"11")?;
// ... or by hierarchical index
settings.set_json_by_key([7, 0, 1], b"8")?;
// ... or by packed index
let (packed, node) = Settings::transcode::<Packed, _>([7, 1, 0]).unwrap();
assert_eq!(packed.into_lsb().get(), 0b1_0111_1_0);
assert_eq!(node, Node::leaf(3));
settings.set_json_by_key(packed, b"9")?;
// ... or by JSON path
settings.set_json_by_key(&JsonPath(".array_tree2[1].b"), b"10")?;

// Hiding paths by setting an Option to `None` at runtime
assert_eq!(settings.set_json("/option_tree", b"13"), Err(Traversal::Absent(1).into()));
settings.option_tree = Some(0);
settings.set_json("/option_tree", b"13")?;
// Hiding a path and descending into the inner `Tree`
settings.option_tree2 = Some(Inner::default());
settings.set_json("/option_tree2/a", b"14")?;
// Hiding items of an array of `Tree`s
settings.array_option_tree[1] = Some(Inner::default());
settings.set_json("/array_option_tree/1/a", b"15")?;

let mut buf = [0; 16];

// Serializing nodes by path
let len = settings.get_json("/struct_", &mut buf).unwrap();
assert_eq!(&buf[..len], br#"{"a":3,"b":3}"#);

// Iterating over all paths
for path in Settings::nodes::<Path<String, '/'>>() {
    let (path, node) = path.unwrap();
    // Serialize each
    match settings.get_json(&path, &mut buf) {
        // Full round-trip: deserialize and set again
        Ok(len) => { settings.set_json(&path, &buf[..len])?; }
        // Some settings are still `None` and thus their paths are expected to be absent
        Err(Error::Traversal(Traversal::Absent(_))) => {}
        e => { e.unwrap(); }

§Settings management

One possible use of miniconf is a backend for run-time settings management in embedded devices.

It was originally designed to work with JSON (serde_json_core) payloads over MQTT (minimq) and provides a MQTT settings management client in the miniconf_mqtt crate and a Python reference implementation to interact with it. Miniconf is agnostic of the serde backend/format, hierarchy separator, and transport/protocol.


miniconf can be used with any serde::Serializer/serde::Deserializer backend, and key format.

Support for / as the path hierarchy separator and JSON (serde_json_core) is implemented through the JsonCoreSlash subtrait.

The Postcard subtrait supports the postcard wire format with any postcard flavor and any Keys type. Combined with the Packed key representation, this is a very space-efficient serde-by-key API.

Blanket implementations are provided for all TreeSerialize+TreeDeserialize types for all formats.


miniconf is also protocol-agnostic. Any means that can receive or emit serialized key-value data can be used to access nodes by path.

The MqttClient in the miniconf_mqtt crate implements settings management over the MQTT protocol with JSON payloads. A Python reference library is provided that interfaces with it. This example discovers the unique prefix of an application listening to messages under the topic quartiq/application/12345 and set its /foo setting to true.

python -m miniconf -d quartiq/application/+ /foo=true

§Derive macros

For structs miniconf offers derive macros for TreeKey, TreeSerialize, and TreeDeserialize. The macros implements the TreeKey, TreeSerialize, and TreeDeserialize traits. Fields/items that form internal nodes (non-leaf) need to implement the respective Tree{Key,Serialize,Deserialize} trait. Leaf fields/items need to support the respective serde trait (and the desired serde::Serializer/serde::Deserializer backend).

Structs, arrays, and Options can then be cascaded to construct more complex trees. When using the derive macro, the behavior and tree recursion depth can be configured for each struct field using the #[tree(depth(Y))] attribute.

See also the TreeKey trait documentation for details.

§Keys and paths

Lookup into the tree is done using a Keys implementation. A blanket implementation through IntoKeys is provided for IntoIterators over Key items. The Key lookup capability is implemented for usize indices and &str names.

Path iteration is supported with arbitrary separator between names.

Very compact hierarchical indices encodings can be obtained from the Packed structure. It implements Keys.


Access to inner fields of some types is not yet supported, e.g. enums other than Option. These are still however usable in their atomic serde form as leaf nodes.

Many std smart pointers are not supported or handled in any special way: Box, Rc, Arc.


  • json-core: Enable the JsonCoreSlash implementation of serializing from and into json slices (using the serde_json_core crate).
  • postcard: Enable the Postcard implementation of serializing from and into the postcard compact binary format (using the postcard crate).
  • derive: Enable the derive macros in miniconf_derive. Enabled by default.


miniconf enables certain kinds of reflective access to heterogeneous trees. Let’s compare it to bevy_reflect which is a comprehensive and mature reflection crate:

bevy_reflect is thoroughly std while miniconf aims at no_std. bevy_reflect uses its Reflect trait to operate on and pass nodes as trait objects. miniconf uses serialized data or Any to access leaf nodes and pure “code” to traverse through internal nodes. The Tree* traits like Reflect thus give access to nodes but unlike Reflect they are all decidedly not object-safe and can not be used as trait objects. This allows miniconf to support non-'static borrowed data (only for TreeAny the leaf nodes need to be 'static) while bevy_reflect requires 'static for Reflect types.

miniconfsupports at least the following reflection features mentioned in the bevy_reflect README:

  • ➕ Derive the traits: miniconf has Tree* derive macros and blanket implementations for arrays and Options. Leaf nodes just need some impls of Serialize/Deserialize/Any where desired.
  • ➕ Interact with fields using their names
  • ➖ “Patch” your types with new values: miniconf only supports limited changes to the tree structure at runtime (Option and custom accessors) while bevy_reflect has powerful dynamic typing tools.
  • ➕ Look up nested fields using “path strings”: In addition to a superset of JSON path style “path strings” miniconf supports hierarchical indices and bit-packed ordered keys.
  • ➕ Iterate over struct fields: miniconf Supports recursive iteration over node keys.
  • ➕ Automatically serialize and deserialize via Serde without explicit serde impls: miniconf supports automatic serializing/deserializing into key-value pairs without an explicit container serde impls.
  • ➕ Trait “reflection”: Together with crosstrait supports building the type registry and enables casting from dyn Any returned by TreeAny to other desired trait objects. Together with erased-serde it can be used to implement node serialization/deserialization using miniconf’s TreeAny without using TreeSerialize/TreeDeserialize similar to bevy_reflect.

Some tangential crates:

  • serde-reflection: extract schemata from serde impls
  • typetag: “derive serde for trait objects” (local traits and impls)
  • deflect: reflection on trait objects using adjacent DWARF debug info as the type registry
  • intertrait: inspiration and source of ideas for crosstrait

§Functional Programming, Polymorphism

The type-heterogeneity of miniconf also borders on functional programming features. For that crates like the following may also be relevant:


  • Concatenate two Keys of different types
  • Counting wrapper for iterators with known exact size
  • Indices of usize to identify a node in a TreeKey
  • JSON style path notation
  • JSON style path notation
  • Keys/IntoKeys for Iterators of Key
  • Metadata about a TreeKey namespace.
  • Type and depth of a node in a TreeKey
  • Node iterator
  • A bit-packed representation of multiple indices.
  • Path with named keys separated by a separator char
  • String split/skip wrapper, smaller/simpler than .split(S).skip(1)


  • Compound errors
  • Type of a node: leaf or internal
  • Errors that can occur when using the Tree traits.


  • Be converted into a Keys
  • TreeSerialize/TreeDeserialize with “JSON and /”.
  • Convert a &str key into a node index on a TreeKey
  • Look up top level field names and convert to indices
  • Capability to yield and look up Keys
  • TreeSerialize/TreeDeserialize with postcard.
  • Look up an IntoKeys in a TreeKey and transcode it.
  • Access any node by keys.
  • Deserialize a leaf node by its keys.
  • Traversal, iteration of keys in a tree.
  • Serialize a leaf node by its keys.

Derive Macros§

  • Shorthand to derive the TreeKey, TreeAny, TreeSerialize, and TreeDeserialize traits for a struct.
  • Derive the TreeAny trait for a struct.
  • Derive the TreeDeserialize trait for a struct.
  • Derive the TreeKey trait for a struct.
  • Derive the TreeSerialize trait for a struct.