1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177

//! Defines everything necessary for avro serialization
//!
//! # For advanced usage
//!
//! You typically want to use top-level functions such as
//! [`to_datum`](crate::to_datum) but access to this may be
//! necessary for more advanced usage.
//!
//! This gives manual access to the type that implements
//! [`serde::Serializer`]
//!
//! Such usage would go as follows:
//! ```
//! let schema: serde_avro_fast::Schema = r#"
//! {
//! 	"namespace": "test",
//! 	"type": "record",
//! 	"name": "Test",
//! 	"fields": [
//! 		{
//! 			"type": {
//! 				"type": "string"
//! 			},
//! 			"name": "field"
//! 		}
//! 	]
//! }
//! "#
//! .parse()
//! .expect("Failed to parse schema");
//!
//! #[derive(serde_derive::Serialize, Debug, PartialEq)]
//! struct Test<'a> {
//! 	field: &'a str,
//! }
//!
//! // Build the struct that will generally serve through serialization
//! let serializer_config = &mut serde_avro_fast::ser::SerializerConfig::new(&schema);
//! let mut serializer_state =
//! 	serde_avro_fast::ser::SerializerState::from_writer(Vec::new(), serializer_config);
//!
//! // It's not the `&mut SerializerState` that implements `serde::Serializer` directly, instead
//! // it is `DatumSerializer` (which is essentially an `&mut SerializerState` but not exactly
//! // because it also keeps track of the current schema node)
//! // We build it through `SerializerState::serializer`
//! serde::Serialize::serialize(&Test { field: "foo" }, serializer_state.serializer())
//! 	.expect("Failed to serialize");
//! let serialized = serializer_state.into_writer();
//!
//! assert_eq!(serialized, &[6, 102, 111, 111]);
//!
//! // reuse config & output buffer across serializations for ideal performance (~40% perf gain)
//! let mut serializer_state = serde_avro_fast::ser::SerializerState::from_writer(
//! 	{
//! 		let mut buf = serialized;
//! 		buf.clear();
//! 		buf
//! 	},
//! 	serializer_config,
//! );
//!
//! serde::Serialize::serialize(&Test { field: "bar" }, serializer_state.serializer())
//! 	.expect("Failed to serialize");
//! let serialized = serializer_state.into_writer();
//!
//! assert_eq!(serialized, &[6, b'b', b'a', b'r']);
//! ```

mod error;
mod serializer;

pub use {error::SerError, serializer::*};

use crate::schema::{
	DecimalRepr, Enum, Fixed, RecordField, Schema, SchemaNode, Union, UnionVariantLookupKey,
};

use {integer_encoding::VarIntWriter, serde::ser::*, std::io::Write};

/// All configuration and state necessary for the serialization to run
///
/// Notably holds the writer and a `&mut` [`SerializerConfig`].
///
/// Does not implement [`Serializer`] directly (use
/// [`.serializer`](Self::serializer) to obtain that).
pub struct SerializerState<'c, 's, W> {
	pub(crate) writer: W,
	/// Storing these here for reuse so that we can bypass the allocation,
	/// and statistically obtain buffers that are already the proper length
	/// (since we have used them for previous records)
	config: &'c mut SerializerConfig<'s>,
}

/// Schema + serialization buffers
///
/// It can be built from a schema:
/// ```
/// # use serde_avro_fast::{ser, Schema};
/// let schema: Schema = r#""int""#.parse().unwrap();
/// let serializer_config = &mut ser::SerializerConfig::new(&schema);
///
/// let mut serialized: Vec<u8> = serde_avro_fast::to_datum_vec(&3, serializer_config).unwrap();
/// assert_eq!(serialized, &[6]);
///
/// // reuse config & output buffer across serializations for ideal performance (~40% perf gain)
/// serialized.clear();
/// let serialized = serde_avro_fast::to_datum(&4, serialized, serializer_config).unwrap();
/// assert_eq!(serialized, &[8]);
/// ```
pub struct SerializerConfig<'s> {
	buffers: Buffers,
	allow_slow_sequence_to_bytes: bool,
	schema: &'s Schema,
}

impl<'s> SerializerConfig<'s> {
	pub fn new(schema: &'s Schema) -> Self {
		Self {
			schema,
			allow_slow_sequence_to_bytes: false,
			buffers: Buffers::default(),
		}
	}

	/// For when you can't use `serde_bytes` and really need to serialize a
	/// sequence as bytes.
	///
	/// If you need to serialize a `Vec<u8>` or `&[u8]` as `Bytes`/`Fixed`,
	/// [`serde_bytes`] is the way to go. If however you can't use it because
	/// e.g. you are transcoding... then you may enable this instead.
	///
	/// It will be slow, because the bytes are processed one by one.
	pub fn allow_slow_sequence_to_bytes(&mut self) -> &mut Self {
		self.allow_slow_sequence_to_bytes = true;
		self
	}

	pub fn schema(&self) -> &'s Schema {
		self.schema
	}
}

impl<'c, 's, W: std::io::Write> SerializerState<'c, 's, W> {
	pub fn from_writer(writer: W, serializer_config: &'c mut SerializerConfig<'s>) -> Self {
		Self {
			writer,
			config: serializer_config,
		}
	}

	pub fn serializer<'r>(&'r mut self) -> DatumSerializer<'r, 'c, 's, W> {
		DatumSerializer {
			schema_node: self.config.schema.root(),
			state: self,
		}
	}
}

impl<W> SerializerState<'_, '_, W> {
	/// Get writer back
	pub fn into_writer(self) -> W {
		self.writer
	}
}

/// Buffers used during serialization, for reuse across serializations
///
/// In order to avoid allocating even when field reordering is necessary we can
/// preserve the necessary allocations from one record to another (even across
/// deserializations).
///
/// This brings ~40% perf improvement
#[derive(Default)]
struct Buffers {
	field_reordering_buffers: Vec<Vec<u8>>,
	field_reordering_super_buffers: Vec<Vec<Option<Vec<u8>>>>,
}