xylem

Xylem is a stateful type conversion framework for Rust.

Concepts

Xylem provides the [Xylem] trait, which is similar to the [std::convert::TryFrom] trait, but with the following differences:

Stateful context

[Xylem::convert] passes a mutable [Context], enabling stateful operations throughout the conversion process. See [Context] documentation for details.

Fixed concersion source

Unlike [std::convert::TryFrom], [Xylem] takes the conversion source type as an associated type [Xylem::From] instead of a generic parameter. This means each type can only be converted from exactly one other specific type under a given [Schema].

Schemas

[Xylem] accepts a type parameter S ("schema"), which acts as a namespace defining the set of conversion rules. This allows different downstream crates to define their own conversion rules without conflicting with each other. For example, if crate foo wants to convert bool from String and crate bar wants to convert bool from i32, they can separately define schema types foo::Xylem and bar::Xylem, then separately implement

impl Xylem<foo::Schema> for bool {
    type From = String;
    // fn convert() omitted
}
impl Xylem<bar::Schema> for bool {
    type From = i32;
    // fn convert() omitted
}

Furthermore, since foo::Schema and bar::Schema are declared in their own crates, Xylem<S> is not considered a foreign trait, so implementing custom conversion rules for [std] types will not result in error E0220 or error E0119.

To use the default conversion rules defined by xylem, make the schema implement the [SchemaExt] trait. There is a convenience macro [declare_schema] for this:

xylem::declare_schema!(Schema: xylem::SchemaExt);

// we defined a schema type called `Schema`.

It is recommended to use Schema as the schema name and declare it at the crate level, because the [Xylem][xylem_codegen::Xylem] macro uses crate::Schema as the default schema type.

The Xylem macro

Xylem provides a [Xylem][xylem_codegen::Xylem] macro, which derives the corresponding [Xylem::From] type from a struct or enum by replacing each type with the corresponding [Xylem::From] type, as well as a [Xylem] implementation. See the [Xylem][xylem_codegen::Xylem] documentation for details.

Note that the order of fields matters because xylem type conversion is stateful, i.e. previous conversions may affect subsequent ones.

The id feature

With the id feature enabled, xylem provides the [Id] type, which is the motivational use case for xylem: Deserialize a config file that references other fields by string ID, replace each declaring ID with an integer storing its occurrence order, and replace each referencing ID with the occurrence order of the declaring ID.

The [Id] type takes two generic parameters, S and X. The type S is just the schema type, while the type X is the subject of identification. X must also implement the [Identifiable] trait, which has an associated type [Identifiable::Scope] used to provide a namespace for the ID. The declaring [Id] field must be declared under X, and X must occur as a (transitive) child of the scope. Further references to the ID of X must occur also as transitive children of the scope, because the scope is dropped when it completes parsing.

Declaring IDs are marked with the argument new = true. If the ID is to be cross-referenced after the scope drops, also mark track = true. Referencing IDs do not need to be marked, but if they serve to import a scope, they should be marked as import = true.

See tests/id.rs and tests/cross_id.rs for example usage.

Note that it is not a design goal for xylem to support lookahead IDs. Due to the stateful nature of xylem, IDs are only indexed when the declaration has been scanned. There is currently no plan to implement multiple passes to pre-index IDs.