polars_io/csv/read/
schema_inference.rs

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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
use std::borrow::Cow;

use polars_core::config::verbose;
use polars_core::prelude::*;
#[cfg(feature = "polars-time")]
use polars_time::chunkedarray::string::infer as date_infer;
#[cfg(feature = "polars-time")]
use polars_time::prelude::string::Pattern;
use polars_utils::format_pl_smallstr;

use super::options::{CommentPrefix, CsvEncoding, NullValues};
use super::parser::{is_comment_line, skip_bom, skip_line_ending, SplitLines};
use super::splitfields::SplitFields;
use super::CsvReadOptions;
use crate::mmap::ReaderBytes;
use crate::utils::{BOOLEAN_RE, FLOAT_RE, FLOAT_RE_DECIMAL, INTEGER_RE};

#[derive(Clone, Debug, Default)]
pub struct SchemaInferenceResult {
    inferred_schema: SchemaRef,
    rows_read: usize,
    bytes_read: usize,
    bytes_total: usize,
    n_threads: Option<usize>,
}

impl SchemaInferenceResult {
    pub fn try_from_reader_bytes_and_options(
        reader_bytes: &ReaderBytes,
        options: &CsvReadOptions,
    ) -> PolarsResult<Self> {
        let parse_options = options.get_parse_options();

        let separator = parse_options.separator;
        let infer_schema_length = options.infer_schema_length;
        let has_header = options.has_header;
        let schema_overwrite_arc = options.schema_overwrite.clone();
        let schema_overwrite = schema_overwrite_arc.as_ref().map(|x| x.as_ref());
        let skip_rows = options.skip_rows;
        let skip_rows_after_header = options.skip_rows_after_header;
        let comment_prefix = parse_options.comment_prefix.as_ref();
        let quote_char = parse_options.quote_char;
        let eol_char = parse_options.eol_char;
        let null_values = parse_options.null_values.clone();
        let try_parse_dates = parse_options.try_parse_dates;
        let raise_if_empty = options.raise_if_empty;
        let mut n_threads = options.n_threads;
        let decimal_comma = parse_options.decimal_comma;

        let bytes_total = reader_bytes.len();

        let (inferred_schema, rows_read, bytes_read) = infer_file_schema(
            reader_bytes,
            separator,
            infer_schema_length,
            has_header,
            schema_overwrite,
            skip_rows,
            skip_rows_after_header,
            comment_prefix,
            quote_char,
            eol_char,
            null_values.as_ref(),
            try_parse_dates,
            raise_if_empty,
            &mut n_threads,
            decimal_comma,
        )?;

        let this = Self {
            inferred_schema: Arc::new(inferred_schema),
            rows_read,
            bytes_read,
            bytes_total,
            n_threads,
        };

        Ok(this)
    }

    pub fn with_inferred_schema(mut self, inferred_schema: SchemaRef) -> Self {
        self.inferred_schema = inferred_schema;
        self
    }

    pub fn get_inferred_schema(&self) -> SchemaRef {
        self.inferred_schema.clone()
    }

    pub fn get_estimated_n_rows(&self) -> usize {
        (self.rows_read as f64 / self.bytes_read as f64 * self.bytes_total as f64) as usize
    }
}

impl CsvReadOptions {
    /// Note: This does not update the schema from the inference result.
    pub fn update_with_inference_result(&mut self, si_result: &SchemaInferenceResult) {
        self.n_threads = si_result.n_threads;
    }
}

pub fn finish_infer_field_schema(possibilities: &PlHashSet<DataType>) -> DataType {
    // determine data type based on possible types
    // if there are incompatible types, use DataType::String
    match possibilities.len() {
        1 => possibilities.iter().next().unwrap().clone(),
        2 if possibilities.contains(&DataType::Int64)
            && possibilities.contains(&DataType::Float64) =>
        {
            // we have an integer and double, fall down to double
            DataType::Float64
        },
        // default to String for conflicting datatypes (e.g bool and int)
        _ => DataType::String,
    }
}

/// Infer the data type of a record
pub fn infer_field_schema(string: &str, try_parse_dates: bool, decimal_comma: bool) -> DataType {
    // when quoting is enabled in the reader, these quotes aren't escaped, we default to
    // String for them
    if string.starts_with('"') {
        if try_parse_dates {
            #[cfg(feature = "polars-time")]
            {
                match date_infer::infer_pattern_single(&string[1..string.len() - 1]) {
                    Some(pattern_with_offset) => match pattern_with_offset {
                        Pattern::DatetimeYMD | Pattern::DatetimeDMY => {
                            DataType::Datetime(TimeUnit::Microseconds, None)
                        },
                        Pattern::DateYMD | Pattern::DateDMY => DataType::Date,
                        Pattern::DatetimeYMDZ => DataType::Datetime(
                            TimeUnit::Microseconds,
                            Some(PlSmallStr::from_static("UTC")),
                        ),
                        Pattern::Time => DataType::Time,
                    },
                    None => DataType::String,
                }
            }
            #[cfg(not(feature = "polars-time"))]
            {
                panic!("activate one of {{'dtype-date', 'dtype-datetime', dtype-time'}} features")
            }
        } else {
            DataType::String
        }
    }
    // match regex in a particular order
    else if BOOLEAN_RE.is_match(string) {
        DataType::Boolean
    } else if !decimal_comma && FLOAT_RE.is_match(string)
        || decimal_comma && FLOAT_RE_DECIMAL.is_match(string)
    {
        DataType::Float64
    } else if INTEGER_RE.is_match(string) {
        DataType::Int64
    } else if try_parse_dates {
        #[cfg(feature = "polars-time")]
        {
            match date_infer::infer_pattern_single(string) {
                Some(pattern_with_offset) => match pattern_with_offset {
                    Pattern::DatetimeYMD | Pattern::DatetimeDMY => {
                        DataType::Datetime(TimeUnit::Microseconds, None)
                    },
                    Pattern::DateYMD | Pattern::DateDMY => DataType::Date,
                    Pattern::DatetimeYMDZ => DataType::Datetime(
                        TimeUnit::Microseconds,
                        Some(PlSmallStr::from_static("UTC")),
                    ),
                    Pattern::Time => DataType::Time,
                },
                None => DataType::String,
            }
        }
        #[cfg(not(feature = "polars-time"))]
        {
            panic!("activate one of {{'dtype-date', 'dtype-datetime', dtype-time'}} features")
        }
    } else {
        DataType::String
    }
}

#[inline]
fn parse_bytes_with_encoding(bytes: &[u8], encoding: CsvEncoding) -> PolarsResult<Cow<str>> {
    Ok(match encoding {
        CsvEncoding::Utf8 => simdutf8::basic::from_utf8(bytes)
            .map_err(|_| polars_err!(ComputeError: "invalid utf-8 sequence"))?
            .into(),
        CsvEncoding::LossyUtf8 => String::from_utf8_lossy(bytes),
    })
}

#[allow(clippy::too_many_arguments)]
fn infer_file_schema_inner(
    reader_bytes: &ReaderBytes,
    separator: u8,
    max_read_rows: Option<usize>,
    has_header: bool,
    schema_overwrite: Option<&Schema>,
    // we take &mut because we maybe need to skip more rows dependent
    // on the schema inference
    mut skip_rows: usize,
    skip_rows_after_header: usize,
    comment_prefix: Option<&CommentPrefix>,
    quote_char: Option<u8>,
    eol_char: u8,
    null_values: Option<&NullValues>,
    try_parse_dates: bool,
    recursion_count: u8,
    raise_if_empty: bool,
    n_threads: &mut Option<usize>,
    decimal_comma: bool,
) -> PolarsResult<(Schema, usize, usize)> {
    // keep track so that we can determine the amount of bytes read
    let start_ptr = reader_bytes.as_ptr() as usize;

    // We use lossy utf8 here because we don't want the schema inference to fail on utf8.
    // It may later.
    let encoding = CsvEncoding::LossyUtf8;

    let bytes = skip_line_ending(skip_bom(reader_bytes), eol_char);
    if raise_if_empty {
        polars_ensure!(!bytes.is_empty(), NoData: "empty CSV");
    };
    let mut lines = SplitLines::new(bytes, quote_char, eol_char).skip(skip_rows);

    // get or create header names
    // when has_header is false, creates default column names with column_ prefix

    // skip lines that are comments
    let mut first_line = None;

    for (i, line) in (&mut lines).enumerate() {
        if !is_comment_line(line, comment_prefix) {
            first_line = Some(line);
            skip_rows += i;
            break;
        }
    }

    if first_line.is_none() {
        first_line = lines.next();
    }

    // now that we've found the first non-comment line we parse the headers, or we create a header
    let headers: Vec<PlSmallStr> = if let Some(mut header_line) = first_line {
        let len = header_line.len();
        if len > 1 {
            // remove carriage return
            let trailing_byte = header_line[len - 1];
            if trailing_byte == b'\r' {
                header_line = &header_line[..len - 1];
            }
        }

        let byterecord = SplitFields::new(header_line, separator, quote_char, eol_char);
        if has_header {
            let headers = byterecord
                .map(|(slice, needs_escaping)| {
                    let slice_escaped = if needs_escaping && (slice.len() >= 2) {
                        &slice[1..(slice.len() - 1)]
                    } else {
                        slice
                    };
                    let s = parse_bytes_with_encoding(slice_escaped, encoding)?;
                    Ok(s)
                })
                .collect::<PolarsResult<Vec<_>>>()?;

            let mut final_headers = Vec::with_capacity(headers.len());

            let mut header_names = PlHashMap::with_capacity(headers.len());

            for name in &headers {
                let count = header_names.entry(name.as_ref()).or_insert(0usize);
                if *count != 0 {
                    final_headers.push(format_pl_smallstr!("{}_duplicated_{}", name, *count - 1))
                } else {
                    final_headers.push(PlSmallStr::from_str(name))
                }
                *count += 1;
            }
            final_headers
        } else {
            byterecord
                .enumerate()
                .map(|(i, _s)| format_pl_smallstr!("column_{}", i + 1))
                .collect::<Vec<PlSmallStr>>()
        }
    } else if has_header && !bytes.is_empty() && recursion_count == 0 {
        // there was no new line char. So we copy the whole buf and add one
        // this is likely to be cheap as there are no rows.
        let mut buf = Vec::with_capacity(bytes.len() + 2);
        buf.extend_from_slice(bytes);
        buf.push(eol_char);

        return infer_file_schema_inner(
            &ReaderBytes::Owned(buf.into()),
            separator,
            max_read_rows,
            has_header,
            schema_overwrite,
            skip_rows,
            skip_rows_after_header,
            comment_prefix,
            quote_char,
            eol_char,
            null_values,
            try_parse_dates,
            recursion_count + 1,
            raise_if_empty,
            n_threads,
            decimal_comma,
        );
    } else if !raise_if_empty {
        return Ok((Schema::default(), 0, 0));
    } else {
        polars_bail!(NoData: "empty CSV");
    };
    if !has_header {
        // re-init lines so that the header is included in type inference.
        lines = SplitLines::new(bytes, quote_char, eol_char).skip(skip_rows);
    }

    let header_length = headers.len();
    // keep track of inferred field types
    let mut column_types: Vec<PlHashSet<DataType>> =
        vec![PlHashSet::with_capacity(4); header_length];
    // keep track of columns with nulls
    let mut nulls: Vec<bool> = vec![false; header_length];

    let mut rows_count = 0;
    let mut fields = Vec::with_capacity(header_length);

    // needed to prevent ownership going into the iterator loop
    let records_ref = &mut lines;

    let mut end_ptr = start_ptr;
    for mut line in records_ref
        .take(match max_read_rows {
            Some(max_read_rows) => {
                if max_read_rows <= (usize::MAX - skip_rows_after_header) {
                    // read skip_rows_after_header more rows for inferring
                    // the correct schema as the first skip_rows_after_header
                    // rows will be skipped
                    max_read_rows + skip_rows_after_header
                } else {
                    max_read_rows
                }
            },
            None => usize::MAX,
        })
        .skip(skip_rows_after_header)
    {
        rows_count += 1;
        // keep track so that we can determine the amount of bytes read
        end_ptr = line.as_ptr() as usize + line.len();

        if line.is_empty() {
            continue;
        }

        // line is a comment -> skip
        if is_comment_line(line, comment_prefix) {
            continue;
        }

        let len = line.len();
        if len > 1 {
            // remove carriage return
            let trailing_byte = line[len - 1];
            if trailing_byte == b'\r' {
                line = &line[..len - 1];
            }
        }

        let mut record = SplitFields::new(line, separator, quote_char, eol_char);

        for i in 0..header_length {
            if let Some((slice, needs_escaping)) = record.next() {
                if slice.is_empty() {
                    unsafe { *nulls.get_unchecked_mut(i) = true };
                } else {
                    let slice_escaped = if needs_escaping && (slice.len() >= 2) {
                        &slice[1..(slice.len() - 1)]
                    } else {
                        slice
                    };
                    let s = parse_bytes_with_encoding(slice_escaped, encoding)?;
                    let dtype = match &null_values {
                        None => Some(infer_field_schema(&s, try_parse_dates, decimal_comma)),
                        Some(NullValues::AllColumns(names)) => {
                            if !names.iter().any(|nv| nv == s.as_ref()) {
                                Some(infer_field_schema(&s, try_parse_dates, decimal_comma))
                            } else {
                                None
                            }
                        },
                        Some(NullValues::AllColumnsSingle(name)) => {
                            if s.as_ref() != name.as_str() {
                                Some(infer_field_schema(&s, try_parse_dates, decimal_comma))
                            } else {
                                None
                            }
                        },
                        Some(NullValues::Named(names)) => {
                            // SAFETY:
                            // we iterate over headers length.
                            let current_name = unsafe { headers.get_unchecked(i) };
                            let null_name = &names.iter().find(|name| name.0 == current_name);

                            if let Some(null_name) = null_name {
                                if null_name.1.as_str() != s.as_ref() {
                                    Some(infer_field_schema(&s, try_parse_dates, decimal_comma))
                                } else {
                                    None
                                }
                            } else {
                                Some(infer_field_schema(&s, try_parse_dates, decimal_comma))
                            }
                        },
                    };
                    if let Some(dtype) = dtype {
                        if matches!(&dtype, DataType::String)
                            && needs_escaping
                            && n_threads.unwrap_or(2) > 1
                        {
                            // The parser will chunk the file.
                            // However this will be increasingly unlikely to be correct if there are many
                            // new line characters in an escaped field. So we set a (somewhat arbitrary)
                            // upper bound to the number of escaped lines we accept.
                            // On the chunking side we also have logic to make this more robust.
                            if slice.iter().filter(|b| **b == eol_char).count() > 8 {
                                if verbose() {
                                    eprintln!("falling back to single core reading because of many escaped new line chars.")
                                }
                                *n_threads = Some(1);
                            }
                        }
                        unsafe { column_types.get_unchecked_mut(i).insert(dtype) };
                    }
                }
            }
        }
    }

    // build schema from inference results
    for i in 0..header_length {
        let field_name = &headers[i];

        if let Some(schema_overwrite) = schema_overwrite {
            if let Some((_, name, dtype)) = schema_overwrite.get_full(field_name) {
                fields.push(Field::new(name.clone(), dtype.clone()));
                continue;
            }

            // column might have been renamed
            // execute only if schema is complete
            if schema_overwrite.len() == header_length {
                if let Some((name, dtype)) = schema_overwrite.get_at_index(i) {
                    fields.push(Field::new(name.clone(), dtype.clone()));
                    continue;
                }
            }
        }

        let possibilities = &column_types[i];
        let dtype = finish_infer_field_schema(possibilities);
        fields.push(Field::new(field_name.clone(), dtype));
    }
    // if there is a single line after the header without an eol
    // we copy the bytes add an eol and rerun this function
    // so that the inference is consistent with and without eol char
    if rows_count == 0
        && !reader_bytes.is_empty()
        && reader_bytes[reader_bytes.len() - 1] != eol_char
        && recursion_count == 0
    {
        let mut rb = Vec::with_capacity(reader_bytes.len() + 1);
        rb.extend_from_slice(reader_bytes);
        rb.push(eol_char);
        return infer_file_schema_inner(
            &ReaderBytes::Owned(rb.into()),
            separator,
            max_read_rows,
            has_header,
            schema_overwrite,
            skip_rows,
            skip_rows_after_header,
            comment_prefix,
            quote_char,
            eol_char,
            null_values,
            try_parse_dates,
            recursion_count + 1,
            raise_if_empty,
            n_threads,
            decimal_comma,
        );
    }

    Ok((Schema::from_iter(fields), rows_count, end_ptr - start_ptr))
}

pub(super) fn check_decimal_comma(decimal_comma: bool, separator: u8) -> PolarsResult<()> {
    if decimal_comma {
        polars_ensure!(b',' != separator, InvalidOperation: "'decimal_comma' argument cannot be combined with ',' separator")
    }
    Ok(())
}

/// Infer the schema of a CSV file by reading through the first n rows of the file,
/// with `max_read_rows` controlling the maximum number of rows to read.
///
/// If `max_read_rows` is not set, the whole file is read to infer its schema.
///
/// Returns
///     - inferred schema
///     - number of rows used for inference.
///     - bytes read
#[allow(clippy::too_many_arguments)]
pub fn infer_file_schema(
    reader_bytes: &ReaderBytes,
    separator: u8,
    max_read_rows: Option<usize>,
    has_header: bool,
    schema_overwrite: Option<&Schema>,
    // we take &mut because we maybe need to skip more rows dependent
    // on the schema inference
    skip_rows: usize,
    skip_rows_after_header: usize,
    comment_prefix: Option<&CommentPrefix>,
    quote_char: Option<u8>,
    eol_char: u8,
    null_values: Option<&NullValues>,
    try_parse_dates: bool,
    raise_if_empty: bool,
    n_threads: &mut Option<usize>,
    decimal_comma: bool,
) -> PolarsResult<(Schema, usize, usize)> {
    check_decimal_comma(decimal_comma, separator)?;
    infer_file_schema_inner(
        reader_bytes,
        separator,
        max_read_rows,
        has_header,
        schema_overwrite,
        skip_rows,
        skip_rows_after_header,
        comment_prefix,
        quote_char,
        eol_char,
        null_values,
        try_parse_dates,
        0,
        raise_if_empty,
        n_threads,
        decimal_comma,
    )
}