Add derive for Schema (#13)

Fixes #3

Introduces two new crates:
- `serde_avro_derive` the user-facing crate that contains the necessary traits/structs and reexports the macro
- `serde_avro_derive_macros` the proc-macro crate that contains the actual derive (not user-facing)

The new traits and construction structs are not in the main crate for several reasons:
- `serde_avro_fast` can be built before `serde_avro_derive`, which improves compilation time
- The API of `serde_avro_derive` is not deemed stable and/or private enough for the 1.0 release of `serde_avro_fast`

.gitignore

@@ -1,2 +1,2 @@
 /target
-/Cargo.lock
+Cargo.lock

Cargo.toml

@@ -1,51 +1,7 @@
-[package]
-	authors = ["Thomas BESSOU <[email protected]>"]
-	description = "An idiomatic implementation of serde/avro (de)serialization"
-	edition = "2021"
-	license = "LGPL-3.0-only"
-	name = "serde_avro_fast"
-	repository = "https://github.com/Ten0/serde_avro_fast"
-	version = "1.0.0-rc.4"
-
-[features]
-	default = ["deflate"]
-	deflate = ["flate2"]
-	snappy = ["snap", "crc32fast"]
-	xz = ["xz2"]
-	zstandard = ["zstd"]
-
-[dependencies]
-	bzip2 = { version = "0.4", optional = true }
-	crc32fast = { version = "1", optional = true }
-	flate2 = { version = "1", optional = true }
-	integer-encoding = { default-features = false, version = "4" }
-	num-traits = "0.2"
-	rand = "0.8"
-	rust_decimal = { version = "1", default-features = false, features = ["std", "serde-with-str"] }
-	serde = "1"
-	serde-transcode = "1"
-	serde_derive = "1"
-	serde_json = "1"
-	serde_serializer_quick_unsupported = "0.1"
-	snap = { version = "1", optional = true }
-	thiserror = "1"
-	xz2 = { version = "0.1", optional = true }
-	zstd = { version = "0.13", optional = true }
-
-[dev-dependencies]
-	anyhow = "1"
-	apache-avro = { version = "0.14", features = ["bzip", "snappy", "xz", "zstandard"] }
-	criterion = "0.5"
-	lazy_static = "1"
-	paste = "1"
-	pretty_assertions = "1"
-	serde-tuple-vec-map = "1"
-	serde_bytes = "0.11"
-
-[[bench]]
-	harness = false
-	name = "single"
-
-[[bench]]
-	harness = false
-	name = "object_container_file_encoding"
+[workspace]
+	members = [
+		"serde_avro_derive",
+		"serde_avro_derive_macros",
+		"serde_avro_fast",
+	]
+	resolver = "2"

serde_avro_derive/Cargo.toml

@@ -0,0 +1,19 @@
+[package]
+	authors = ["Thomas BESSOU <[email protected]>"]
+	description = "Derive avro schema for Rust structs for serde_avro_fast"
+	edition = "2021"
+	license = "LGPL-3.0-only"
+	name = "serde_avro_derive"
+	repository = "https://github.com/Ten0/serde_avro_fast"
+	version = "0.1.0"
+	workspace = ".."
+
+[dependencies]
+	serde_avro_derive_macros = { path = "../serde_avro_derive_macros", version = "0.1" }
+	serde_avro_fast = { path = "../serde_avro_fast", version = "1.0.0-rc.4" }
+
+[dev-dependencies]
+	lazy_static = "1"
+	pretty_assertions = "1"
+	regex = "1"
+	serde_json = "1"

serde_avro_derive/src/lib.rs

@@ -0,0 +1,257 @@
+//! Bring automatic Avro Schema generation to [`serde_avro_fast`]
+//!
+//! See the [`#[derive(Schema)]`](derive@Schema) documentation for more
+//! information
+
+pub use serde_avro_fast;
+
+pub use serde_avro_derive_macros::*;
+
+use std::{any::TypeId, collections::HashMap};
+
+use serde_avro_fast::schema::*;
+
+/// We can automatically build a schema for this type (can be `derive`d)
+///
+/// This trait can be derived using [`#[derive(Schema)]`](derive@Schema)
+pub trait BuildSchema {
+	/// Build a [`struct@Schema`] for this type
+	fn schema() -> Result<Schema, SchemaError> {
+		Self::schema_mut().try_into()
+	}
+	/// Build a [`SchemaMut`] for this type
+	fn schema_mut() -> SchemaMut {
+		let mut builder = SchemaBuilder::default();
+		Self::append_schema(&mut builder);
+		SchemaMut::from_nodes(builder.nodes)
+	}
+
+	/// Largely internal method to build the schema. Registers the schema within
+	/// the builder.
+	///
+	/// This does not check if this type already exists in the builder, so it
+	/// should never be called directly (instead, use
+	/// [`SchemaBuilder::find_or_build`])
+	///
+	/// The [`SchemaNode`] for this type should be put at the current end of the
+	/// `nodes` array, and its non-already-built dependencies should be put
+	/// after in the array.
+	fn append_schema(builder: &mut SchemaBuilder);
+
+	/// Largely internal type used by [`#[derive(Schema)]`](derive@Schema)
+	///
+	/// The TypeId of this type will be used to lookup whether the
+	/// [`SchemaNode`] for this type has already been built in the
+	/// [`SchemaBuilder`].
+	///
+	/// This indirection is required to allow non-static types to implement
+	/// [`BuildSchema`], and also enables using the same node for types that we
+	/// know map to the same schema.
+	type TypeLookup: std::any::Any;
+}
+
+/// Largely internal type used by [`#[derive(Schema)]`](derive@Schema)
+///
+/// You should typically not use this directly
+#[derive(Default)]
+pub struct SchemaBuilder {
+	pub nodes: Vec<SchemaNode>,
+	pub already_built_types: HashMap<TypeId, SchemaKey>,
+	_private: (),
+}
+
+impl SchemaBuilder {
+	/// Reserve a slot in the `nodes` array
+	///
+	/// After building the `SchemaNode`, it should be put at the corresponding
+	/// position in `nodes`.
+	pub fn reserve(&mut self) -> usize {
+		let idx = self.nodes.len();
+		self.nodes.push(SchemaNode::RegularType(RegularType::Null));
+		idx
+	}
+
+	pub fn find_or_build<T: BuildSchema + ?Sized>(&mut self) -> SchemaKey {
+		match self
+			.already_built_types
+			.entry(TypeId::of::<T::TypeLookup>())
+		{
+			std::collections::hash_map::Entry::Occupied(entry) => *entry.get(),
+			std::collections::hash_map::Entry::Vacant(entry) => {
+				let idx = SchemaKey::from_idx(self.nodes.len());
+				entry.insert(idx);
+				T::append_schema(self);
+				assert!(
+					self.nodes.len() > idx.idx(),
+					"append_schema should always insert at least a node \
+					(and its dependencies below itself)"
+				);
+				idx
+			}
+		}
+	}
+
+	pub fn build_logical_type<T: BuildSchema + ?Sized>(
+		&mut self,
+		logical_type: LogicalType,
+	) -> SchemaKey {
+		let reserved_schema_key = self.reserve();
+		let new_node = SchemaNode::LogicalType {
+			logical_type,
+			inner: self.find_or_build::<T>(),
+		};
+		self.nodes[reserved_schema_key] = new_node;
+		SchemaKey::from_idx(reserved_schema_key)
+	}
+}
+
+macro_rules! impl_primitive {
+	($($ty:ty, $variant:ident;)+) => {
+		$(
+			impl BuildSchema for $ty {
+				fn append_schema(builder: &mut SchemaBuilder) {
+					builder.nodes.push(SchemaNode::RegularType(RegularType::$variant));
+				}
+				type TypeLookup = Self;
+			}
+		)*
+	};
+}
+impl_primitive!(
+	(), Null;
+	bool, Boolean;
+	i32, Int;
+	i64, Long;
+	f32, Float;
+	f64, Double;
+	String, String;
+	Vec<u8>, Bytes;
+);
+
+macro_rules! impl_forward {
+	($($ty:ty, $to:ty;)+) => {
+		$(
+			impl BuildSchema for $ty {
+				fn append_schema(builder: &mut SchemaBuilder) {
+					<$to as BuildSchema>::append_schema(builder)
+				}
+				type TypeLookup = <$to as BuildSchema>::TypeLookup;
+			}
+		)*
+	};
+}
+impl_forward! {
+	str, String;
+	[u8], Vec<u8>;
+	u16, i32;
+	u32, i64;
+	u64, i64;
+	i8, i32;
+	i16, i32;
+	usize, i64;
+}
+
+macro_rules! impl_ptr {
+	($($($ty_path:ident)::+,)+) => {
+		$(
+			impl<T: BuildSchema + ?Sized> BuildSchema for $($ty_path)::+<T> {
+				fn append_schema(builder: &mut SchemaBuilder) {
+					<T as BuildSchema>::append_schema(builder)
+				}
+				type TypeLookup = T::TypeLookup;
+			}
+		)*
+	};
+}
+impl_ptr! {
+	Box,
+	std::sync::Arc,
+	std::rc::Rc,
+	std::cell::RefCell,
+	std::cell::Cell,
+}
+impl<T: BuildSchema + ?Sized> BuildSchema for &'_ T {
+	fn append_schema(builder: &mut SchemaBuilder) {
+		<T as BuildSchema>::append_schema(builder)
+	}
+	type TypeLookup = T::TypeLookup;
+}
+impl<T: BuildSchema + ?Sized> BuildSchema for &'_ mut T {
+	fn append_schema(builder: &mut SchemaBuilder) {
+		<T as BuildSchema>::append_schema(builder)
+	}
+	type TypeLookup = T::TypeLookup;
+}
+
+impl<T: BuildSchema> BuildSchema for Vec<T> {
+	fn append_schema(builder: &mut SchemaBuilder) {
+		let reserved_schema_key = builder.reserve();
+		let new_node =
+			SchemaNode::RegularType(RegularType::Array(Array::new(builder.find_or_build::<T>())));
+		builder.nodes[reserved_schema_key] = new_node;
+	}
+
+	type TypeLookup = Vec<T::TypeLookup>;
+}
+
+impl<T: BuildSchema> BuildSchema for [T] {
+	fn append_schema(builder: &mut SchemaBuilder) {
+		<Vec<T> as BuildSchema>::append_schema(builder)
+	}
+	type TypeLookup = <Vec<T> as BuildSchema>::TypeLookup;
+}
+
+impl<T: BuildSchema> BuildSchema for Option<T> {
+	fn append_schema(builder: &mut SchemaBuilder) {
+		let reserved_schema_key = builder.reserve();
+		let new_node = SchemaNode::RegularType(RegularType::Union(Union::new(vec![
+			builder.find_or_build::<()>(),
+			builder.find_or_build::<T>(),
+		])));
+		builder.nodes[reserved_schema_key] = new_node;
+	}
+
+	type TypeLookup = Option<T::TypeLookup>;
+}
+
+impl<const N: usize> BuildSchema for [u8; N] {
+	fn append_schema(builder: &mut SchemaBuilder) {
+		builder
+			.nodes
+			.push(SchemaNode::RegularType(RegularType::Fixed(Fixed::new(
+				Name::from_fully_qualified_name(format!("u8_array_{}", N)),
+				N,
+			))));
+	}
+	type TypeLookup = Self;
+}
+
+impl<S: std::ops::Deref<Target = str>, V: BuildSchema> BuildSchema for HashMap<S, V> {
+	fn append_schema(builder: &mut SchemaBuilder) {
+		let reserved_schema_key = builder.reserve();
+		let new_node =
+			SchemaNode::RegularType(RegularType::Map(Map::new(builder.find_or_build::<V>())));
+		builder.nodes[reserved_schema_key] = new_node;
+	}
+	type TypeLookup = HashMap<String, V::TypeLookup>;
+}
+impl<S: std::ops::Deref<Target = str>, V: BuildSchema> BuildSchema
+	for std::collections::BTreeMap<S, V>
+{
+	fn append_schema(builder: &mut SchemaBuilder) {
+		<HashMap<String, V> as BuildSchema>::append_schema(builder)
+	}
+	type TypeLookup = <HashMap<String, V> as BuildSchema>::TypeLookup;
+}
+
+#[doc(hidden)]
+pub fn hash_type_id(struct_name: &mut String, type_id: TypeId) {
+	use std::{
+		fmt::Write,
+		hash::{Hash as _, Hasher as _},
+	};
+	#[allow(deprecated)] // I actually want to not change hasher
+	let mut hasher = std::hash::SipHasher::new();
+	type_id.hash(&mut hasher);
+	write!(struct_name, "_{:016x?}", hasher.finish()).unwrap();
+}