arrow_json/
lib.rs

1// Licensed to the Apache Software Foundation (ASF) under one
2// or more contributor license agreements.  See the NOTICE file
3// distributed with this work for additional information
4// regarding copyright ownership.  The ASF licenses this file
5// to you under the Apache License, Version 2.0 (the
6// "License"); you may not use this file except in compliance
7// with the License.  You may obtain a copy of the License at
8//
9//   http://www.apache.org/licenses/LICENSE-2.0
10//
11// Unless required by applicable law or agreed to in writing,
12// software distributed under the License is distributed on an
13// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
14// KIND, either express or implied.  See the License for the
15// specific language governing permissions and limitations
16// under the License.
17
18//! Transfer data between the Arrow memory format and JSON line-delimited records.
19//!
20//! See the module level documentation for the
21//! [`reader`] and [`writer`] for usage examples.
22//!
23//! # Binary Data uses `Base16` Encoding
24//!
25//! As per [RFC7159] JSON cannot encode arbitrary binary data. This crate works around that
26//! limitation by encoding/decoding binary data as a [hexadecimal] string (i.e.
27//! [`Base16` encoding]).
28//!
29//! Note that `Base16` only has 50% space efficiency (i.e., the encoded data is twice as large
30//! as the original). If that is an issue, we recommend to convert binary data to/from a different
31//! encoding format such as `Base64` instead. See the following example for details.
32//!
33//! ## `Base64` Encoding Example
34//!
35//! [`Base64`] is a common [binary-to-text encoding] scheme with a space efficiency of 75%. The
36//! following example shows how to use the [`arrow_cast`] crate to encode binary data to `Base64`
37//! before converting it to JSON and how to decode it back.
38//!
39//! ```
40//! # use std::io::Cursor;
41//! # use std::sync::Arc;
42//! # use arrow_array::{BinaryArray, RecordBatch, StringArray};
43//! # use arrow_array::cast::AsArray;
44//! use arrow_cast::base64::{b64_decode, b64_encode, BASE64_STANDARD};
45//! # use arrow_json::{LineDelimitedWriter, ReaderBuilder};
46//! #
47//! // The data we want to write
48//! let input = BinaryArray::from(vec![b"\xDE\x00\xFF".as_ref()]);
49//!
50//! // Base64 encode it to a string
51//! let encoded: StringArray = b64_encode(&BASE64_STANDARD, &input);
52//!
53//! // Write the StringArray to JSON
54//! let batch = RecordBatch::try_from_iter([("col", Arc::new(encoded) as _)]).unwrap();
55//! let mut buf = Vec::with_capacity(1024);
56//! let mut writer = LineDelimitedWriter::new(&mut buf);
57//! writer.write(&batch).unwrap();
58//! writer.finish().unwrap();
59//!
60//! // Read the JSON data
61//! let cursor = Cursor::new(buf);
62//! let mut reader = ReaderBuilder::new(batch.schema()).build(cursor).unwrap();
63//! let batch = reader.next().unwrap().unwrap();
64//!
65//! // Reverse the base64 encoding
66//! let col: BinaryArray = batch.column(0).as_string::<i32>().clone().into();
67//! let output = b64_decode(&BASE64_STANDARD, &col).unwrap();
68//!
69//! assert_eq!(input, output);
70//! ```
71//!
72//! [RFC7159]: https://datatracker.ietf.org/doc/html/rfc7159#section-8.1
73//! [binary-to-text encoding]: https://en.wikipedia.org/wiki/Binary-to-text_encoding
74//! [hexadecimal]: https://en.wikipedia.org/wiki/Hexadecimal
75//! [`Base16` encoding]: https://en.wikipedia.org/wiki/Base16#Base16
76//! [`Base64`]: https://en.wikipedia.org/wiki/Base64
77
78#![doc(
79    html_logo_url = "https://arrow.apache.org/img/arrow-logo_chevrons_black-txt_white-bg.svg",
80    html_favicon_url = "https://arrow.apache.org/img/arrow-logo_chevrons_black-txt_transparent-bg.svg"
81)]
82#![cfg_attr(docsrs, feature(doc_cfg))]
83#![deny(clippy::allow_attributes)]
84#![deny(rustdoc::broken_intra_doc_links)]
85#![warn(missing_docs)]
86
87pub mod reader;
88pub mod writer;
89
90pub use self::reader::{Reader, ReaderBuilder};
91pub use self::writer::{
92    ArrayWriter, Encoder, EncoderFactory, EncoderOptions, LineDelimitedWriter, Writer,
93    WriterBuilder,
94};
95use half::f16;
96use serde_json::{Number, Value};
97
98/// Specifies what is considered valid JSON when reading or writing
99/// RecordBatches or StructArrays.
100///
101/// This enum controls which form(s) the Reader will accept and which form the
102/// Writer will produce. For example, if the RecordBatch Schema is
103/// `[("a", Int32), ("r", Struct("b": Boolean, "c" Utf8))]`
104/// then a Reader with [`StructMode::ObjectOnly`] would read rows of the form
105/// `{"a": 1, "r": {"b": true, "c": "cat"}}` while with [`StructMode::ListOnly`]
106/// would read rows of the form `[1, [true, "cat"]]`. A Writer would produce
107/// rows formatted similarly.
108///
109/// The list encoding is more compact if the schema is known, and is used by
110/// tools such as [Presto] and [Trino].
111///
112/// When reading objects, the order of the key does not matter. When reading
113/// lists, the entries must be the same number and in the same order as the
114/// struct fields. Map columns are not affected by this option.
115///
116/// [Presto]: https://prestodb.io/docs/current/develop/client-protocol.html#important-queryresults-attributes
117/// [Trino]: https://trino.io/docs/current/develop/client-protocol.html#important-queryresults-attributes
118#[derive(Copy, Clone, Debug, Default, PartialEq, Eq)]
119pub enum StructMode {
120    #[default]
121    /// Encode/decode structs as objects (e.g., {"a": 1, "b": "c"})
122    ObjectOnly,
123    /// Encode/decode structs as lists (e.g., [1, "c"])
124    ListOnly,
125}
126
127/// Trait declaring any type that is serializable to JSON. This includes all primitive types (bool, i32, etc.).
128pub trait JsonSerializable: 'static {
129    /// Converts self into json value if its possible
130    fn into_json_value(self) -> Option<Value>;
131}
132
133macro_rules! json_serializable {
134    ($t:ty) => {
135        impl JsonSerializable for $t {
136            fn into_json_value(self) -> Option<Value> {
137                Some(self.into())
138            }
139        }
140    };
141}
142
143json_serializable!(bool);
144json_serializable!(u8);
145json_serializable!(u16);
146json_serializable!(u32);
147json_serializable!(u64);
148json_serializable!(i8);
149json_serializable!(i16);
150json_serializable!(i32);
151json_serializable!(i64);
152
153impl JsonSerializable for i128 {
154    fn into_json_value(self) -> Option<Value> {
155        // Serialize as string to avoid issues with arbitrary_precision serde_json feature
156        // - https://github.com/serde-rs/json/issues/559
157        // - https://github.com/serde-rs/json/issues/845
158        // - https://github.com/serde-rs/json/issues/846
159        Some(self.to_string().into())
160    }
161}
162
163impl JsonSerializable for f16 {
164    fn into_json_value(self) -> Option<Value> {
165        Number::from_f64(f64::round(f64::from(self) * 1000.0) / 1000.0).map(Value::Number)
166    }
167}
168
169impl JsonSerializable for f32 {
170    fn into_json_value(self) -> Option<Value> {
171        Number::from_f64(f64::round(self as f64 * 1000.0) / 1000.0).map(Value::Number)
172    }
173}
174
175impl JsonSerializable for f64 {
176    fn into_json_value(self) -> Option<Value> {
177        Number::from_f64(self).map(Value::Number)
178    }
179}
180
181#[cfg(test)]
182mod tests {
183    use super::*;
184    use crate::writer::JsonArray;
185    use crate::writer::LineDelimited;
186    use arrow_array::{
187        ArrayRef, GenericBinaryArray, GenericByteViewArray, GenericListViewArray, RecordBatch,
188        RecordBatchWriter, builder::FixedSizeBinaryBuilder, types::BinaryViewType,
189    };
190    use arrow_schema::{DataType, Field, Fields, Schema};
191    use serde_json::Value::{Bool, Number as VNumber, String as VString};
192    use std::io::Cursor;
193    use std::sync::Arc;
194
195    #[test]
196    fn test_arrow_native_type_to_json() {
197        assert_eq!(Some(Bool(true)), true.into_json_value());
198        assert_eq!(Some(VNumber(Number::from(1))), 1i8.into_json_value());
199        assert_eq!(Some(VNumber(Number::from(1))), 1i16.into_json_value());
200        assert_eq!(Some(VNumber(Number::from(1))), 1i32.into_json_value());
201        assert_eq!(Some(VNumber(Number::from(1))), 1i64.into_json_value());
202        assert_eq!(Some(VString("1".to_string())), 1i128.into_json_value());
203        assert_eq!(Some(VNumber(Number::from(1))), 1u8.into_json_value());
204        assert_eq!(Some(VNumber(Number::from(1))), 1u16.into_json_value());
205        assert_eq!(Some(VNumber(Number::from(1))), 1u32.into_json_value());
206        assert_eq!(Some(VNumber(Number::from(1))), 1u64.into_json_value());
207        assert_eq!(
208            Some(VNumber(Number::from_f64(0.01f64).unwrap())),
209            0.01.into_json_value()
210        );
211        assert_eq!(
212            Some(VNumber(Number::from_f64(0.01f64).unwrap())),
213            0.01f64.into_json_value()
214        );
215        assert_eq!(None, f32::NAN.into_json_value());
216    }
217
218    #[test]
219    fn test_json_roundtrip_structs() {
220        let schema = Arc::new(Schema::new(vec![
221            Field::new(
222                "c1",
223                DataType::Struct(Fields::from(vec![
224                    Field::new("c11", DataType::Int32, true),
225                    Field::new(
226                        "c12",
227                        DataType::Struct(vec![Field::new("c121", DataType::Utf8, false)].into()),
228                        false,
229                    ),
230                ])),
231                false,
232            ),
233            Field::new("c2", DataType::Utf8, false),
234        ]));
235
236        {
237            let object_input = r#"{"c1":{"c11":1,"c12":{"c121":"e"}},"c2":"a"}
238{"c1":{"c12":{"c121":"f"}},"c2":"b"}
239{"c1":{"c11":5,"c12":{"c121":"g"}},"c2":"c"}
240"#
241            .as_bytes();
242            let object_reader = ReaderBuilder::new(schema.clone())
243                .with_struct_mode(StructMode::ObjectOnly)
244                .build(object_input)
245                .unwrap();
246
247            let mut object_output: Vec<u8> = Vec::new();
248            let mut object_writer = WriterBuilder::new()
249                .with_struct_mode(StructMode::ObjectOnly)
250                .build::<_, LineDelimited>(&mut object_output);
251            for batch_res in object_reader {
252                object_writer.write(&batch_res.unwrap()).unwrap();
253            }
254            assert_eq!(object_input, &object_output);
255        }
256
257        {
258            let list_input = r#"[[1,["e"]],"a"]
259[[null,["f"]],"b"]
260[[5,["g"]],"c"]
261"#
262            .as_bytes();
263            let list_reader = ReaderBuilder::new(schema.clone())
264                .with_struct_mode(StructMode::ListOnly)
265                .build(list_input)
266                .unwrap();
267
268            let mut list_output: Vec<u8> = Vec::new();
269            let mut list_writer = WriterBuilder::new()
270                .with_struct_mode(StructMode::ListOnly)
271                .build::<_, LineDelimited>(&mut list_output);
272            for batch_res in list_reader {
273                list_writer.write(&batch_res.unwrap()).unwrap();
274            }
275            assert_eq!(list_input, &list_output);
276        }
277    }
278
279    #[test]
280    #[expect(invalid_from_utf8)]
281    fn test_json_roundtrip_binary() {
282        let not_utf8: &[u8] = b"Not UTF8 \xa0\xa1!";
283        assert!(str::from_utf8(not_utf8).is_err());
284
285        let values: &[Option<&[u8]>] = &[
286            Some(b"Ned Flanders" as &[u8]),
287            None,
288            Some(b"Troy McClure" as &[u8]),
289            Some(not_utf8),
290        ];
291        // Binary:
292        assert_binary_json(Arc::new(GenericBinaryArray::<i32>::from_iter(values)));
293
294        // LargeBinary:
295        assert_binary_json(Arc::new(GenericBinaryArray::<i64>::from_iter(values)));
296
297        // FixedSizeBinary:
298        assert_binary_json(build_array_fixed_size_binary(12, values));
299
300        // BinaryView:
301        assert_binary_json(Arc::new(GenericByteViewArray::<BinaryViewType>::from_iter(
302            values,
303        )));
304    }
305
306    fn build_array_fixed_size_binary(byte_width: i32, values: &[Option<&[u8]>]) -> ArrayRef {
307        let mut builder = FixedSizeBinaryBuilder::new(byte_width);
308        for value in values {
309            match value {
310                Some(v) => builder.append_value(v).unwrap(),
311                None => builder.append_null(),
312            }
313        }
314        Arc::new(builder.finish())
315    }
316
317    fn assert_binary_json(array: ArrayRef) {
318        // encode and check JSON with and without explicit nulls
319        assert_binary_json_with_writer(
320            array.clone(),
321            WriterBuilder::new().with_explicit_nulls(true),
322        );
323        assert_binary_json_with_writer(array, WriterBuilder::new().with_explicit_nulls(false));
324    }
325
326    fn assert_binary_json_with_writer(array: ArrayRef, builder: WriterBuilder) {
327        let batch = RecordBatch::try_from_iter([("bytes", array)]).unwrap();
328
329        let mut buf = Vec::new();
330        let json_value: Value = {
331            let mut writer = builder.build::<_, JsonArray>(&mut buf);
332            writer.write(&batch).unwrap();
333            writer.close().unwrap();
334            serde_json::from_slice(&buf).unwrap()
335        };
336
337        let json_array = json_value.as_array().unwrap();
338
339        let decoded = {
340            let mut decoder = ReaderBuilder::new(batch.schema().clone())
341                .build_decoder()
342                .unwrap();
343            decoder.serialize(json_array).unwrap();
344            decoder.flush().unwrap().unwrap()
345        };
346
347        assert_eq!(batch, decoded);
348    }
349
350    fn assert_list_view_roundtrip<O: arrow_array::OffsetSizeTrait>() {
351        let flat_field = Arc::new(Field::new("item", DataType::Int32, true));
352        let flat_dt = GenericListViewArray::<O>::DATA_TYPE_CONSTRUCTOR(flat_field);
353
354        let nested_inner = Arc::new(Field::new("item", DataType::Int32, false));
355        let nested_inner_dt = GenericListViewArray::<O>::DATA_TYPE_CONSTRUCTOR(nested_inner);
356        let nested_outer = Arc::new(Field::new("item", nested_inner_dt, true));
357        let nested_dt = GenericListViewArray::<O>::DATA_TYPE_CONSTRUCTOR(nested_outer);
358
359        let schema = Arc::new(Schema::new(vec![
360            Field::new("flat", flat_dt, true),
361            Field::new("nested", nested_dt, true),
362        ]));
363
364        let input = r#"{"flat":[1,2,3],"nested":[[1,2],[3]]}
365{"flat":[4,null]}
366{}
367{"flat":[6],"nested":[[4,5,6]]}
368{"flat":[]}
369"#
370        .as_bytes();
371
372        let batches: Vec<RecordBatch> = ReaderBuilder::new(schema.clone())
373            .with_batch_size(1024)
374            .build(Cursor::new(input))
375            .unwrap()
376            .collect::<Result<Vec<_>, _>>()
377            .unwrap();
378
379        let mut output = Vec::new();
380        let mut writer = WriterBuilder::new().build::<_, LineDelimited>(&mut output);
381        for batch in &batches {
382            writer.write(batch).unwrap();
383        }
384        writer.finish().unwrap();
385
386        assert_eq!(input, &output);
387    }
388
389    #[test]
390    fn test_json_roundtrip_list_view() {
391        assert_list_view_roundtrip::<i32>();
392        assert_list_view_roundtrip::<i64>();
393    }
394
395    #[test]
396    fn test_json_roundtrip_fixed_size_list() {
397        let inner = Arc::new(Field::new("item", DataType::Int32, true));
398        let schema = Arc::new(Schema::new(vec![
399            Field::new("flat", DataType::FixedSizeList(inner.clone(), 3), true),
400            Field::new(
401                "nested",
402                DataType::FixedSizeList(
403                    Arc::new(Field::new("item", DataType::FixedSizeList(inner, 2), true)),
404                    2,
405                ),
406                true,
407            ),
408        ]));
409
410        let input = r#"{"flat":[1,2,3],"nested":[[1,2],[3,4]]}
411{"flat":[4,null,5]}
412{"flat":[6,7,8],"nested":[[null,5],[6,null]]}
413"#
414        .as_bytes();
415
416        let batches: Vec<RecordBatch> = ReaderBuilder::new(schema.clone())
417            .with_batch_size(1024)
418            .build(Cursor::new(input))
419            .unwrap()
420            .collect::<Result<Vec<_>, _>>()
421            .unwrap();
422
423        let mut output = Vec::new();
424        let mut writer = WriterBuilder::new().build::<_, LineDelimited>(&mut output);
425        for batch in &batches {
426            writer.write(batch).unwrap();
427        }
428        writer.finish().unwrap();
429
430        assert_eq!(input, &output);
431    }
432}
arrow_json/lib.rs

arrow_json/
lib.rs
