polars_io/json/mod.rs
1//! # (De)serialize JSON files.
2//!
3//! ## Read JSON to a DataFrame
4//!
5//! ## Example
6//!
7//! ```
8//! use polars_core::prelude::*;
9//! use polars_io::prelude::*;
10//! use std::io::Cursor;
11//! use std::num::NonZeroUsize;
12//!
13//! let basic_json = r#"{"a":1, "b":2.0, "c":false, "d":"4"}
14//! {"a":-10, "b":-3.5, "c":true, "d":"4"}
15//! {"a":2, "b":0.6, "c":false, "d":"text"}
16//! {"a":1, "b":2.0, "c":false, "d":"4"}
17//! {"a":7, "b":-3.5, "c":true, "d":"4"}
18//! {"a":1, "b":0.6, "c":false, "d":"text"}
19//! {"a":1, "b":2.0, "c":false, "d":"4"}
20//! {"a":5, "b":-3.5, "c":true, "d":"4"}
21//! {"a":1, "b":0.6, "c":false, "d":"text"}
22//! {"a":1, "b":2.0, "c":false, "d":"4"}
23//! {"a":1, "b":-3.5, "c":true, "d":"4"}
24//! {"a":1, "b":0.6, "c":false, "d":"text"}"#;
25//! let file = Cursor::new(basic_json);
26//! let df = JsonReader::new(file)
27//! .with_json_format(JsonFormat::JsonLines)
28//! .infer_schema_len(NonZeroUsize::new(3))
29//! .with_batch_size(NonZeroUsize::new(3).unwrap())
30//! .finish()
31//! .unwrap();
32//!
33//! println!("{:?}", df);
34//! ```
35//! >>> Outputs:
36//!
37//! ```text
38//! +-----+--------+-------+--------+
39//! | a | b | c | d |
40//! | --- | --- | --- | --- |
41//! | i64 | f64 | bool | str |
42//! +=====+========+=======+========+
43//! | 1 | 2 | false | "4" |
44//! +-----+--------+-------+--------+
45//! | -10 | -3.5e0 | true | "4" |
46//! +-----+--------+-------+--------+
47//! | 2 | 0.6 | false | "text" |
48//! +-----+--------+-------+--------+
49//! | 1 | 2 | false | "4" |
50//! +-----+--------+-------+--------+
51//! | 7 | -3.5e0 | true | "4" |
52//! +-----+--------+-------+--------+
53//! | 1 | 0.6 | false | "text" |
54//! +-----+--------+-------+--------+
55//! | 1 | 2 | false | "4" |
56//! +-----+--------+-------+--------+
57//! | 5 | -3.5e0 | true | "4" |
58//! +-----+--------+-------+--------+
59//! | 1 | 0.6 | false | "text" |
60//! +-----+--------+-------+--------+
61//! | 1 | 2 | false | "4" |
62//! +-----+--------+-------+--------+
63//! ```
64//!
65pub(crate) mod infer;
66
67use std::io::Write;
68use std::num::NonZeroUsize;
69use std::ops::Deref;
70
71use arrow::array::LIST_VALUES_NAME;
72use arrow::legacy::conversion::chunk_to_struct;
73use polars_core::chunked_array::cast::CastOptions;
74use polars_core::error::to_compute_err;
75use polars_core::prelude::*;
76use polars_error::{PolarsResult, polars_bail};
77use polars_json::json::write::FallibleStreamingIterator;
78#[cfg(feature = "serde")]
79use serde::{Deserialize, Serialize};
80use simd_json::BorrowedValue;
81
82use crate::mmap::{MmapBytesReader, ReaderBytes};
83use crate::prelude::*;
84
85#[derive(Copy, Clone, Debug, PartialEq, Eq, Default, Hash)]
86#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
87#[cfg_attr(feature = "dsl-schema", derive(schemars::JsonSchema))]
88pub struct JsonWriterOptions {}
89
90/// The format to use to write the DataFrame to JSON: `Json` (a JSON array)
91/// or `JsonLines` (each row output on a separate line).
92///
93/// In either case, each row is serialized as a JSON object whose keys are the column names and
94/// whose values are the row's corresponding values.
95pub enum JsonFormat {
96 /// A single JSON array containing each DataFrame row as an object. The length of the array is the number of rows in
97 /// the DataFrame.
98 ///
99 /// Use this to create valid JSON that can be deserialized back into an array in one fell swoop.
100 Json,
101 /// Each DataFrame row is serialized as a JSON object on a separate line. The number of lines in the output is the
102 /// number of rows in the DataFrame.
103 ///
104 /// The [JSON Lines](https://jsonlines.org) format makes it easy to read records in a streaming fashion, one (line)
105 /// at a time. But the output in its entirety is not valid JSON; only the individual lines are.
106 ///
107 /// It is recommended to use the file extension `.jsonl` when saving as JSON Lines.
108 JsonLines,
109}
110
111/// Writes a DataFrame to JSON.
112///
113/// Under the hood, this uses [`arrow2::io::json`](https://docs.rs/arrow2/latest/arrow2/io/json/write/fn.write.html).
114/// `arrow2` generally serializes types that are not JSON primitives, such as Date and DateTime, as their
115/// `Display`-formatted versions. For instance, a (naive) DateTime column is formatted as the String `"yyyy-mm-dd
116/// HH:MM:SS"`. To control how non-primitive columns are serialized, convert them to String or another primitive type
117/// before serializing.
118#[must_use]
119pub struct JsonWriter<W: Write> {
120 /// File or Stream handler
121 buffer: W,
122 json_format: JsonFormat,
123}
124
125impl<W: Write> JsonWriter<W> {
126 pub fn with_json_format(mut self, format: JsonFormat) -> Self {
127 self.json_format = format;
128 self
129 }
130}
131
132impl<W> SerWriter<W> for JsonWriter<W>
133where
134 W: Write,
135{
136 /// Create a new `JsonWriter` writing to `buffer` with format `JsonFormat::JsonLines`. To specify a different
137 /// format, use e.g., [`JsonWriter::new(buffer).with_json_format(JsonFormat::Json)`](JsonWriter::with_json_format).
138 fn new(buffer: W) -> Self {
139 JsonWriter {
140 buffer,
141 json_format: JsonFormat::JsonLines,
142 }
143 }
144
145 fn finish(&mut self, df: &mut DataFrame) -> PolarsResult<()> {
146 df.align_chunks_par();
147 let fields = df
148 .iter()
149 .map(|s| {
150 #[cfg(feature = "object")]
151 polars_ensure!(!matches!(s.dtype(), DataType::Object(_)), ComputeError: "cannot write 'Object' datatype to json");
152 Ok(s.field().to_arrow(CompatLevel::newest()))
153 })
154 .collect::<PolarsResult<Vec<_>>>()?;
155 let batches = df
156 .iter_chunks(CompatLevel::newest(), false)
157 .map(|chunk| Ok(Box::new(chunk_to_struct(chunk, fields.clone())) as ArrayRef));
158
159 match self.json_format {
160 JsonFormat::JsonLines => {
161 let serializer = polars_json::ndjson::write::Serializer::new(batches, vec![]);
162 let writer =
163 polars_json::ndjson::write::FileWriter::new(&mut self.buffer, serializer);
164 writer.collect::<PolarsResult<()>>()?;
165 },
166 JsonFormat::Json => {
167 let serializer = polars_json::json::write::Serializer::new(batches, vec![]);
168 polars_json::json::write::write(&mut self.buffer, serializer)?;
169 },
170 }
171
172 Ok(())
173 }
174}
175
176pub struct BatchedWriter<W: Write> {
177 writer: W,
178}
179
180impl<W> BatchedWriter<W>
181where
182 W: Write,
183{
184 pub fn new(writer: W) -> Self {
185 BatchedWriter { writer }
186 }
187 /// Write a batch to the json writer.
188 ///
189 /// # Panics
190 /// The caller must ensure the chunks in the given [`DataFrame`] are aligned.
191 pub fn write_batch(&mut self, df: &DataFrame) -> PolarsResult<()> {
192 let fields = df
193 .iter()
194 .map(|s| {
195 #[cfg(feature = "object")]
196 polars_ensure!(!matches!(s.dtype(), DataType::Object(_)), ComputeError: "cannot write 'Object' datatype to json");
197 Ok(s.field().to_arrow(CompatLevel::newest()))
198 })
199 .collect::<PolarsResult<Vec<_>>>()?;
200 let chunks = df.iter_chunks(CompatLevel::newest(), false);
201 let batches =
202 chunks.map(|chunk| Ok(Box::new(chunk_to_struct(chunk, fields.clone())) as ArrayRef));
203 let mut serializer = polars_json::ndjson::write::Serializer::new(batches, vec![]);
204 while let Some(block) = serializer.next()? {
205 self.writer.write_all(block)?;
206 }
207 Ok(())
208 }
209}
210
211/// Reads JSON in one of the formats in [`JsonFormat`] into a DataFrame.
212#[must_use]
213pub struct JsonReader<'a, R>
214where
215 R: MmapBytesReader,
216{
217 reader: R,
218 rechunk: bool,
219 ignore_errors: bool,
220 infer_schema_len: Option<NonZeroUsize>,
221 batch_size: NonZeroUsize,
222 projection: Option<Vec<PlSmallStr>>,
223 schema: Option<SchemaRef>,
224 schema_overwrite: Option<&'a Schema>,
225 json_format: JsonFormat,
226}
227
228pub fn remove_bom(bytes: &[u8]) -> PolarsResult<&[u8]> {
229 if bytes.starts_with(&[0xEF, 0xBB, 0xBF]) {
230 // UTF-8 BOM
231 Ok(&bytes[3..])
232 } else if bytes.starts_with(&[0xFE, 0xFF]) || bytes.starts_with(&[0xFF, 0xFE]) {
233 // UTF-16 BOM
234 polars_bail!(ComputeError: "utf-16 not supported")
235 } else {
236 Ok(bytes)
237 }
238}
239impl<R> SerReader<R> for JsonReader<'_, R>
240where
241 R: MmapBytesReader,
242{
243 fn new(reader: R) -> Self {
244 JsonReader {
245 reader,
246 rechunk: true,
247 ignore_errors: false,
248 infer_schema_len: Some(NonZeroUsize::new(100).unwrap()),
249 batch_size: NonZeroUsize::new(8192).unwrap(),
250 projection: None,
251 schema: None,
252 schema_overwrite: None,
253 json_format: JsonFormat::Json,
254 }
255 }
256
257 fn set_rechunk(mut self, rechunk: bool) -> Self {
258 self.rechunk = rechunk;
259 self
260 }
261
262 /// Take the SerReader and return a parsed DataFrame.
263 ///
264 /// Because JSON values specify their types (number, string, etc), no upcasting or conversion is performed between
265 /// incompatible types in the input. In the event that a column contains mixed dtypes, is it unspecified whether an
266 /// error is returned or whether elements of incompatible dtypes are replaced with `null`.
267 fn finish(mut self) -> PolarsResult<DataFrame> {
268 let pre_rb: ReaderBytes = (&mut self.reader).into();
269 let bytes = remove_bom(pre_rb.deref())?;
270 let rb = ReaderBytes::Borrowed(bytes);
271 let out = match self.json_format {
272 JsonFormat::Json => {
273 polars_ensure!(!self.ignore_errors, InvalidOperation: "'ignore_errors' only supported in ndjson");
274 let mut bytes = rb.deref().to_vec();
275 let owned = &mut vec![];
276 compression::maybe_decompress_bytes(&bytes, owned)?;
277 // the easiest way to avoid ownership issues is by implicitly figuring out if
278 // decompression happened (owned is only populated on decompress), then pick which bytes to parse
279 let json_value = if owned.is_empty() {
280 simd_json::to_borrowed_value(&mut bytes).map_err(to_compute_err)?
281 } else {
282 simd_json::to_borrowed_value(owned).map_err(to_compute_err)?
283 };
284 if let BorrowedValue::Array(array) = &json_value {
285 if array.is_empty() & self.schema.is_none() & self.schema_overwrite.is_none() {
286 return Ok(DataFrame::empty());
287 }
288 }
289
290 let allow_extra_fields_in_struct = self.schema.is_some();
291
292 let mut schema = if let Some(schema) = self.schema {
293 Arc::unwrap_or_clone(schema)
294 } else {
295 // Infer.
296 let inner_dtype = if let BorrowedValue::Array(values) = &json_value {
297 infer::json_values_to_supertype(
298 values,
299 self.infer_schema_len
300 .unwrap_or(NonZeroUsize::new(usize::MAX).unwrap()),
301 )?
302 } else {
303 DataType::from_arrow_dtype(&polars_json::json::infer(&json_value)?)
304 };
305
306 let DataType::Struct(fields) = inner_dtype else {
307 polars_bail!(ComputeError: "can only deserialize json objects")
308 };
309
310 Schema::from_iter(fields)
311 };
312
313 if let Some(overwrite) = self.schema_overwrite {
314 overwrite_schema(&mut schema, overwrite)?;
315 }
316
317 let mut needs_cast = false;
318 let deserialize_schema = schema
319 .iter()
320 .map(|(name, dt)| {
321 Field::new(
322 name.clone(),
323 dt.clone().map_leaves(&mut |leaf_dt| {
324 // Deserialize enums and categoricals as strings first.
325 match leaf_dt {
326 #[cfg(feature = "dtype-categorical")]
327 DataType::Enum(..) | DataType::Categorical(..) => {
328 needs_cast = true;
329 DataType::String
330 },
331 leaf_dt => leaf_dt,
332 }
333 }),
334 )
335 })
336 .collect();
337
338 let arrow_dtype =
339 DataType::Struct(deserialize_schema).to_arrow(CompatLevel::newest());
340
341 let arrow_dtype = if let BorrowedValue::Array(_) = &json_value {
342 ArrowDataType::LargeList(Box::new(arrow::datatypes::Field::new(
343 LIST_VALUES_NAME,
344 arrow_dtype,
345 true,
346 )))
347 } else {
348 arrow_dtype
349 };
350
351 let arr = polars_json::json::deserialize(
352 &json_value,
353 arrow_dtype,
354 allow_extra_fields_in_struct,
355 )?;
356 let arr = arr.as_any().downcast_ref::<StructArray>().ok_or_else(
357 || polars_err!(ComputeError: "can only deserialize json objects"),
358 )?;
359
360 let mut df = DataFrame::try_from(arr.clone())?;
361 if needs_cast {
362 unsafe {
363 for (col, dt) in df.get_columns_mut().iter_mut().zip(schema.iter_values()) {
364 *col = col.cast_with_options(
365 dt,
366 if self.ignore_errors {
367 CastOptions::NonStrict
368 } else {
369 CastOptions::Strict
370 },
371 )?;
372 }
373 df.clear_schema();
374 }
375 }
376 PolarsResult::Ok(df)
377 },
378 JsonFormat::JsonLines => {
379 let mut json_reader = CoreJsonReader::new(
380 rb,
381 None,
382 self.schema,
383 self.schema_overwrite,
384 None,
385 1024, // sample size
386 NonZeroUsize::new(1 << 18).unwrap(),
387 false,
388 self.infer_schema_len,
389 self.ignore_errors,
390 None,
391 None,
392 None,
393 )?;
394 let mut df: DataFrame = json_reader.as_df()?;
395 if self.rechunk {
396 df.as_single_chunk_par();
397 }
398 Ok(df)
399 },
400 }?;
401
402 // TODO! Ensure we don't materialize the columns we don't need
403 if let Some(proj) = self.projection.as_deref() {
404 out.select(proj.iter().cloned())
405 } else {
406 Ok(out)
407 }
408 }
409}
410
411impl<'a, R> JsonReader<'a, R>
412where
413 R: MmapBytesReader,
414{
415 /// Set the JSON file's schema
416 pub fn with_schema(mut self, schema: SchemaRef) -> Self {
417 self.schema = Some(schema);
418 self
419 }
420
421 /// Overwrite parts of the inferred schema.
422 pub fn with_schema_overwrite(mut self, schema: &'a Schema) -> Self {
423 self.schema_overwrite = Some(schema);
424 self
425 }
426
427 /// Set the JSON reader to infer the schema of the file. Currently, this is only used when reading from
428 /// [`JsonFormat::JsonLines`], as [`JsonFormat::Json`] reads in the entire array anyway.
429 ///
430 /// When using [`JsonFormat::JsonLines`], `max_records = None` will read the entire buffer in order to infer the
431 /// schema, `Some(1)` would look only at the first record, `Some(2)` the first two records, etc.
432 ///
433 /// It is an error to pass `max_records = Some(0)`, as a schema cannot be inferred from 0 records when deserializing
434 /// from JSON (unlike CSVs, there is no header row to inspect for column names).
435 pub fn infer_schema_len(mut self, max_records: Option<NonZeroUsize>) -> Self {
436 self.infer_schema_len = max_records;
437 self
438 }
439
440 /// Set the batch size (number of records to load at one time)
441 ///
442 /// This heavily influences loading time.
443 pub fn with_batch_size(mut self, batch_size: NonZeroUsize) -> Self {
444 self.batch_size = batch_size;
445 self
446 }
447
448 /// Set the reader's column projection: the names of the columns to keep after deserialization. If `None`, all
449 /// columns are kept.
450 ///
451 /// Setting `projection` to the columns you want to keep is more efficient than deserializing all of the columns and
452 /// then dropping the ones you don't want.
453 pub fn with_projection(mut self, projection: Option<Vec<PlSmallStr>>) -> Self {
454 self.projection = projection;
455 self
456 }
457
458 pub fn with_json_format(mut self, format: JsonFormat) -> Self {
459 self.json_format = format;
460 self
461 }
462
463 /// Return a `null` if an error occurs during parsing.
464 pub fn with_ignore_errors(mut self, ignore: bool) -> Self {
465 self.ignore_errors = ignore;
466 self
467 }
468}