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
// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Use of this software is governed by the Business Source License
// included in the LICENSE file.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0.

use std::mem;
use std::str::FromStr;

use derivative::Derivative;
use mz_lowertest::MzReflect;
use mz_ore::fmt::FormatBuffer;
use mz_proto::{ProtoType, RustType, TryFromProtoError};
use mz_repr::adt::regex::Regex;
use proptest::prelude::{Arbitrary, Strategy};
use serde::{Deserialize, Serialize};

use crate::scalar::EvalError;

include!(concat!(env!("OUT_DIR"), "/mz_expr.scalar.like_pattern.rs"));

/// The number of subpatterns after which using regexes would be more efficient.
const MAX_SUBPATTERNS: usize = 5;

/// The escape character to use by default in LIKE patterns.
const DEFAULT_ESCAPE: char = '\\';
const DOUBLED_ESCAPE: &str = "\\\\";

/// Specifies escape behavior for the LIKE pattern.
#[derive(Clone, Copy, Debug)]
pub enum EscapeBehavior {
    /// No escape character.
    Disabled,
    /// Use a custom escape character.
    Char(char),
}

impl Default for EscapeBehavior {
    fn default() -> EscapeBehavior {
        EscapeBehavior::Char(DEFAULT_ESCAPE)
    }
}

impl FromStr for EscapeBehavior {
    type Err = EvalError;

    fn from_str(s: &str) -> Result<EscapeBehavior, EvalError> {
        let mut chars = s.chars();
        match chars.next() {
            None => Ok(EscapeBehavior::Disabled),
            Some(c) => match chars.next() {
                None => Ok(EscapeBehavior::Char(c)),
                Some(_) => Err(EvalError::LikeEscapeTooLong),
            },
        }
    }
}

/// Converts a pattern string that uses a custom escape character to one that uses the default.
pub fn normalize_pattern(pattern: &str, escape: EscapeBehavior) -> Result<String, EvalError> {
    match escape {
        EscapeBehavior::Disabled => Ok(pattern.replace(DEFAULT_ESCAPE, DOUBLED_ESCAPE)),
        EscapeBehavior::Char(DEFAULT_ESCAPE) => Ok(pattern.into()),
        EscapeBehavior::Char(custom_escape_char) => {
            let mut p = String::with_capacity(2 * pattern.len());
            let mut cs = pattern.chars();
            while let Some(c) = cs.next() {
                if c == custom_escape_char {
                    match cs.next() {
                        Some(c2) => {
                            p.push(DEFAULT_ESCAPE);
                            p.push(c2);
                        }
                        None => return Err(EvalError::UnterminatedLikeEscapeSequence),
                    }
                } else if c == DEFAULT_ESCAPE {
                    p.push_str(DOUBLED_ESCAPE);
                } else {
                    p.push(c);
                }
            }
            p.shrink_to_fit();
            Ok(p)
        }
    }
}

// This implementation supports a couple of different methods of matching
// text against a SQL LIKE or ILIKE pattern.
//
// The most general approach is to convert the LIKE pattern into a
// regular expression and use the well-tested Regex library to perform the
// match. This works well with complex patterns and case-insensitive matches
// that are hard to get right.
//
// That said, regular expressions aren't that efficient. For most patterns
// we can do better using built-in string matching.

pub use matcher::Matcher;
use matcher::MatcherImpl;

// This lint interacts poorly with `derivative` here; we are confident it generates
// compatible `PartialOrd` and `Ord` impls. Unfortunately it also requires we introduce
// this module to allow it.
#[allow(clippy::non_canonical_partial_ord_impl)]
mod matcher {
    use super::*;

    /// An object that can test whether a string matches a LIKE or ILIKE pattern.
    #[derive(Debug, Clone, Deserialize, Serialize, Derivative, MzReflect)]
    #[derivative(Eq, PartialEq, Ord, PartialOrd, Hash)]
    pub struct Matcher {
        pub pattern: String,
        pub case_insensitive: bool,
        #[derivative(
            PartialEq = "ignore",
            Hash = "ignore",
            Ord = "ignore",
            PartialOrd = "ignore"
        )]
        pub(super) matcher_impl: MatcherImpl,
    }

    impl Matcher {
        pub fn is_match(&self, text: &str) -> bool {
            match &self.matcher_impl {
                MatcherImpl::String(subpatterns) => is_match_subpatterns(subpatterns, text),
                MatcherImpl::Regex(r) => r.is_match(text),
            }
        }
    }

    impl RustType<ProtoMatcher> for Matcher {
        fn into_proto(&self) -> ProtoMatcher {
            ProtoMatcher {
                pattern: self.pattern.clone(),
                case_insensitive: self.case_insensitive,
            }
        }

        fn from_proto(proto: ProtoMatcher) -> Result<Self, TryFromProtoError> {
            compile(proto.pattern.as_str(), proto.case_insensitive).map_err(|eval_err| {
                TryFromProtoError::LikePatternDeserializationError(eval_err.to_string())
            })
        }
    }

    #[derive(Debug, Clone, Deserialize, Serialize, MzReflect)]
    pub(super) enum MatcherImpl {
        String(Vec<Subpattern>),
        Regex(Regex),
    }
}

/// Builds a Matcher that matches a SQL LIKE pattern.
pub fn compile(pattern: &str, case_insensitive: bool) -> Result<Matcher, EvalError> {
    // We would like to have a consistent, documented limit to the size of
    // supported LIKE patterns. The real limiting factor is the number of states
    // that can be handled by the Regex library. In testing, I was able to
    // create an adversarial pattern "%a%b%c%d%e..." that started failing around
    // 9 KiB, so we chose 8 KiB as the limit. This is consistent with limits
    // set by other databases, like SQL Server.
    // On the other hand, PostgreSQL does not have a documented limit.
    if pattern.len() > 8 << 10 {
        return Err(EvalError::LikePatternTooLong);
    }
    let subpatterns = build_subpatterns(pattern)?;
    let matcher_impl = match case_insensitive || subpatterns.len() > MAX_SUBPATTERNS {
        false => MatcherImpl::String(subpatterns),
        true => MatcherImpl::Regex(build_regex(&subpatterns, case_insensitive)?),
    };
    Ok(Matcher {
        pattern: pattern.into(),
        case_insensitive,
        matcher_impl,
    })
}

pub fn any_matcher() -> impl Strategy<Value = Matcher> {
    // Generates a string out of a pool of characters. The pool has at least one
    // representative from the following classes of the characters (listed in
    // order of its appearance in the regex):
    // * Alphanumeric characters, both upper and lower-case.
    // * Control characters.
    // * Punctuation minus the escape character.
    // * Space characters.
    // * Multi-byte characters.
    // * _ and %, which are special characters for a like pattern.
    // * Escaped _ and %, plus the escape character itself. This implementation
    //   will have to be modified if we support choosing a different escape character.
    //
    // Syntax reference for LIKE here:
    // https://www.postgresql.org/docs/current/functions-matching.html#FUNCTIONS-LIKE
    (
        r"([[:alnum:]]|[[:cntrl:]]|([[[:punct:]]&&[^\\]])|[[:space:]]|华|_|%|(\\_)|(\\%)|(\\\\)){0, 50}",
        bool::arbitrary(),
    )
        .prop_map(|(pattern, case_insensitive)| compile(&pattern, case_insensitive).unwrap())
}

// The algorithm below is based on the observation that any LIKE pattern can be
// decomposed into multiple parts:
//     <PATTERN> := <SUB-PATTERN> (<SUB-PATTERN> ...)
//     <SUB-PATTERN> := <WILDCARDS> <SUFFIX>
//
// The sub-patterns start with zero or more wildcard characters, eventually
// followed by (non-wildcard) literal characters. The last sub-pattern may
// have an empty SUFFIX.
//
// Example: the PATTERN "n__dl%" can be broken into the following parts:
//   1. SUB-PATTERN = <WILDCARDS ""> <SUFFIX "n">
//   2. SUB-PATTERN = <WILDCARDS "__"> <SUFFIX "dl">
//   3. SUB-PATTERN = <WILDCARDS "%"> <SUFFIX "">
//
// The WILDCARDS can be any combination of '_', which matches exactly 1 char,
// and '%' which matches zero or more chars. These wildcards can be simplified
// down to the (min, max) of characters they might consume:
//     ""  = (0, 0)    // doesn't consume any characters
//     "_" = (1, 1)    // consumes exactly one
//     "%" = (0, many) // zero or more
// These are additive, so:
//     "_%"    = (1, many)
//     "__%__" = (4, many)
//     "%%%_"  = (1, many)

#[derive(Debug, Default, Clone, Deserialize, Serialize, MzReflect)]
struct Subpattern {
    /// The minimum number of characters that can be consumed by the wildcard expression.
    consume: usize,
    /// Whether the wildcard expression can consume an arbitrary number of characters.
    many: bool,
    /// A string literal that is expected after the wildcards.
    suffix: String,
}

impl Subpattern {
    /// Converts a Subpattern to an equivalent regular expression and writes it to a given string.
    fn write_regex_to(&self, r: &mut String) {
        match self.consume {
            0 => {
                if self.many {
                    r.push_str(".*");
                }
            }
            1 => {
                r.push('.');
                if self.many {
                    r.push('+');
                }
            }
            n => {
                r.push_str(".{");
                write!(r, "{}", n);
                if self.many {
                    r.push(',');
                }
                r.push('}');
            }
        }
        regex_syntax::escape_into(&self.suffix, r);
    }
}

impl RustType<ProtoSubpattern> for Subpattern {
    fn into_proto(&self) -> ProtoSubpattern {
        ProtoSubpattern {
            consume: self.consume.into_proto(),
            many: self.many,
            suffix: self.suffix.clone(),
        }
    }

    fn from_proto(proto: ProtoSubpattern) -> Result<Self, TryFromProtoError> {
        Ok(Subpattern {
            consume: proto.consume.into_rust()?,
            many: proto.many,
            suffix: proto.suffix,
        })
    }
}

fn is_match_subpatterns(subpatterns: &[Subpattern], mut text: &str) -> bool {
    let (subpattern, subpatterns) = match subpatterns {
        [] => return text.is_empty(),
        [subpattern, subpatterns @ ..] => (subpattern, subpatterns),
    };
    // Go ahead and skip the minimum number of characters the sub-pattern consumes:
    if subpattern.consume > 0 {
        let mut chars = text.chars();
        if chars.nth(subpattern.consume - 1).is_none() {
            return false;
        }
        text = chars.as_str();
    }
    if subpattern.many {
        // The sub-pattern might consume any number of characters, but we need to find
        // where it terminates so we can match any subsequent sub-patterns. We do this
        // by searching for the suffix string using str::find.
        //
        // We could investigate using a fancier substring search like Boyer-Moore:
        // https://en.wikipedia.org/wiki/Boyer%E2%80%93Moore_string-search_algorithm
        //
        // .. but it's likely not worth it. It's slower for small strings,
        // and doesn't really start outperforming the naive approach until
        // haystack sizes of 1KB or greater. See benchmarking results from:
        // https://github.com/killerswan/boyer-moore-search/blob/master/README.md
        //
        // Another approach that may be interesting to look at is a
        // hardware-optimized search:
        // http://0x80.pl/articles/simd-strfind.html
        if subpattern.suffix.len() == 0 {
            // Nothing to find... This should only happen in the last sub-pattern.
            assert!(
                subpatterns.is_empty(),
                "empty suffix in middle of a pattern"
            );
            return true;
        }
        // Use rfind so we perform a greedy capture, like Regex.
        let mut found = text.rfind(&subpattern.suffix);
        loop {
            match found {
                None => return false,
                Some(offset) => {
                    let mut end = offset + subpattern.suffix.len();
                    if is_match_subpatterns(subpatterns, &text[end..]) {
                        return true;
                    }
                    // Didn't match, look for the next rfind.
                    if offset == 0 {
                        return false;
                    }
                    // Find the previous valid char byte.
                    loop {
                        end -= 1;
                        if text.is_char_boundary(end) {
                            break;
                        }
                    }
                    found = text[..end].rfind(&subpattern.suffix);
                }
            }
        }
    }
    // No string search needed, we just use a prefix match on rest.
    if !text.starts_with(&subpattern.suffix) {
        return false;
    }
    is_match_subpatterns(subpatterns, &text[subpattern.suffix.len()..])
}

/// Breaks a LIKE pattern into a chain of sub-patterns.
fn build_subpatterns(pattern: &str) -> Result<Vec<Subpattern>, EvalError> {
    let mut subpatterns = Vec::with_capacity(MAX_SUBPATTERNS);
    let mut current = Subpattern::default();
    let mut in_wildcard = true;
    let mut in_escape = false;
    for c in pattern.chars() {
        match c {
            c if !in_escape && c == DEFAULT_ESCAPE => {
                in_escape = true;
                in_wildcard = false;
            }
            '_' if !in_escape => {
                if !in_wildcard {
                    current.suffix.shrink_to_fit();
                    subpatterns.push(mem::take(&mut current));
                    in_wildcard = true;
                }
                current.consume += 1;
            }
            '%' if !in_escape => {
                if !in_wildcard {
                    current.suffix.shrink_to_fit();
                    subpatterns.push(mem::take(&mut current));
                    in_wildcard = true;
                }
                current.many = true;
            }
            c => {
                current.suffix.push(c);
                in_escape = false;
                in_wildcard = false;
            }
        }
    }
    if in_escape {
        return Err(EvalError::UnterminatedLikeEscapeSequence);
    }
    current.suffix.shrink_to_fit();
    subpatterns.push(current);
    subpatterns.shrink_to_fit();
    Ok(subpatterns)
}

/// Builds a regular expression that matches some parsed Subpatterns.
fn build_regex(subpatterns: &[Subpattern], case_insensitive: bool) -> Result<Regex, EvalError> {
    let mut r = String::from("^");
    for sp in subpatterns {
        sp.write_regex_to(&mut r);
    }
    r.push('$');
    match Regex::new(&r, case_insensitive) {
        Ok(regex) => Ok(regex),
        Err(regex::Error::CompiledTooBig(_)) => Err(EvalError::LikePatternTooLong),
        Err(e) => Err(EvalError::Internal(
            format!("build_regex produced invalid regex: {}", e).into(),
        )),
    }
}

// Unit Tests
//
// Most of the unit tests for LIKE and ILIKE can be found in:
//    test/sqllogictest/cockroach/like.slt
// These tests are here as a convenient place to run quick tests while
// actively working on changes to the implementation. Make sure you
// run the full test suite before submitting any changes.

#[cfg(test)]
mod test {
    use super::*;

    #[mz_ore::test]
    fn test_normalize_pattern() {
        struct TestCase<'a> {
            pattern: &'a str,
            escape: EscapeBehavior,
            expected: &'a str,
        }
        let test_cases = vec![
            TestCase {
                pattern: "",
                escape: EscapeBehavior::Disabled,
                expected: "",
            },
            TestCase {
                pattern: "ban%na!",
                escape: EscapeBehavior::default(),
                expected: "ban%na!",
            },
            TestCase {
                pattern: "ban%%%na!",
                escape: EscapeBehavior::Char('%'),
                expected: "ban\\%\\na!",
            },
            TestCase {
                pattern: "ban%na\\!",
                escape: EscapeBehavior::Char('n'),
                expected: "ba\\%\\a\\\\!",
            },
            TestCase {
                pattern: "ban%na\\!",
                escape: EscapeBehavior::Disabled,
                expected: "ban%na\\\\!",
            },
            TestCase {
                pattern: "ban\\na!",
                escape: EscapeBehavior::Char('n'),
                expected: "ba\\\\\\a!",
            },
            TestCase {
                pattern: "ban\\\\na!",
                escape: EscapeBehavior::Char('n'),
                expected: "ba\\\\\\\\\\a!",
            },
            TestCase {
                pattern: "food",
                escape: EscapeBehavior::Char('o'),
                expected: "f\\od",
            },
            TestCase {
                pattern: "漢漢",
                escape: EscapeBehavior::Char('漢'),
                expected: "\\漢",
            },
        ];

        for input in test_cases {
            let actual = normalize_pattern(input.pattern, input.escape).unwrap();
            assert!(
                actual == input.expected,
                "normalize_pattern({:?}, {:?}):\n\tactual: {:?}\n\texpected: {:?}\n",
                input.pattern,
                input.escape,
                actual,
                input.expected,
            );
        }
    }

    #[mz_ore::test]
    fn test_escape_too_long() {
        match EscapeBehavior::from_str("foo") {
            Err(EvalError::LikeEscapeTooLong) => {}
            _ => {
                panic!("expected error when using escape string with >1 character");
            }
        }
    }

    #[mz_ore::test]
    fn test_like() {
        struct Input<'a> {
            haystack: &'a str,
            matches: bool,
        }
        let input = |haystack, matches| Input { haystack, matches };
        struct Pattern<'a> {
            needle: &'a str,
            case_insensitive: bool,
            inputs: Vec<Input<'a>>,
        }
        let test_cases = vec![
            Pattern {
                needle: "ban%na!",
                case_insensitive: false,
                inputs: vec![input("banana!", true)],
            },
            Pattern {
                needle: "foo",
                case_insensitive: true,
                inputs: vec![
                    input("", false),
                    input("f", false),
                    input("fo", false),
                    input("foo", true),
                    input("FOO", true),
                    input("Foo", true),
                    input("fOO", true),
                    input("food", false),
                ],
            },
        ];

        for tc in test_cases {
            let matcher = compile(tc.needle, tc.case_insensitive).unwrap();
            for input in tc.inputs {
                let actual = matcher.is_match(input.haystack);
                assert!(
                    actual == input.matches,
                    "{:?} {} {:?}:\n\tactual: {:?}\n\texpected: {:?}\n",
                    input.haystack,
                    match tc.case_insensitive {
                        true => "ILIKE",
                        false => "LIKE",
                    },
                    tc.needle,
                    actual,
                    input.matches,
                );
            }
        }
    }
}