1 files changed, 70 insertions, 0 deletions
diff --git a/src/type_tree.rs b/src/type_tree.rs
new file mode 100644
index 0000000..8e8098b
--- /dev/null
+++ b/src/type_tree.rs
@@ -0,0 +1,70 @@
+// Copyright (C) 2021-2022 Soni L.
+// SPDX-License-Identifier: MIT OR Apache-2.0
+
+//! Type Tree support.
+//!
+//! Type Trees are a Datafu feature for extracting types from a `serde`-based
+//! `Deserialize` in such a way that it can be used with Datafu patterns.
+//!
+//! They work by matching the `Deserialize` against some data, with the help of
+//! `serde_transmute`. Datafu then collects the relevant `Deserialize` calls,
+//! and uses them to infer an appropriate type tree for dynamic
+//! deserialization.
+//!
+//! When introspecting the `Deserialize`, all matching parts are extracted, and
+//! non-matching parts are ignored. Even if an error occurs, Datafu will gladly
+//! infer a type tree for what it could match.
+//!
+//! For example, given a struct and the corresponding data:
+//!
+//! ```
+//! struct Foo {
+//!   bar: i32,
+//! }
+//!
+//! let data = Foo { bar: 0 };
+//! ```
+//!
+//! Building a type tree will first inspect the struct like so:
+//!
+//! 1. call `deserialize()` on `Foo`.
+//! 2. inspect the `deserialize_struct` from `Foo`, storing the name and
+//!     fields.
+//! 3. give `Foo` the appropriate visitor (from `data`), through
+//!     `serde_transmute`.
+//! 4. inspect the `deserialize_i32` etc, also storing those.
+//!
+//! The resulting type tree can then be used in any pattern to effectively
+//! match a `Foo`, but more efficiently than with a predicate. Another big
+//! difference between predicates and type trees is how predicates are eager,
+//! and can consume values that would otherwise be matched by the rest of a
+//! pattern.
+//!
+//! Type trees are pretty flexible. Consider the following example:
+//!
+//! ```
+//! struct Foo {
+//!   bar: Vec<u32>,
+//! }
+//! 
+//! let data = Foo { bar: vec![1, 2, 3] };
+//! ```
+//!
+//! This will actually produce a type tree which checks that the first 3 items
+//! are `u32`! Further, when using different types for the predicate and the
+//! data, you can get even more flexiblity. For example, with the following
+//! struct and data:
+//!
+//! ```
+//! struct Foo {
+//!   bar: Vec<u32>,
+//! }
+//!
+//! let data = ();
+//! ```
+//!
+//! Datafu will actually inspect the `deserialize_struct`, and then the
+//! struct visitor will error. But despite the error, it'll still create a type
+//! tree for the `deserialize_struct`!
+
+// TODO