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
//! CL100K pretokenizer — single-pass, zero allocation.
//!
//! Pattern: `(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]|\s+(?!\S)|\s+`
//! Contractions are case-insensitive and standalone.
//! Any non-newline/non-alnum char can prefix letters (not just space).
//! Digits chunked to max 3 per piece.
use crate::util::{decode_utf8, is_ascii_letter, is_digit};
pub struct Cl100k<'a> {
bytes: &'a [u8],
pos: usize,
len: usize,
}
impl<'a> Cl100k<'a> {
pub fn new(text: &'a str) -> Self {
let bytes = text.as_bytes();
Self { bytes, pos: 0, len: bytes.len() }
}
#[inline(always)]
fn at(&self, pos: usize) -> u8 {
unsafe { *self.bytes.get_unchecked(pos) }
}
#[inline(always)]
fn scan_letters(&mut self) {
while self.pos < self.len {
let b = self.at(self.pos);
if is_ascii_letter(b) {
self.pos += 1;
} else if b >= 0x80 {
let (ch, cl) = decode_utf8(&self.bytes[self.pos..]);
if ch.is_alphabetic() { self.pos += cl; } else { return; }
} else {
return;
}
}
}
/// Scan up to 3 digits.
#[inline(always)]
fn scan_digits_chunked(&mut self) {
let mut count = 0u8;
while self.pos < self.len && count < 3 && is_digit(self.at(self.pos)) {
self.pos += 1;
count += 1;
}
}
/// Check for case-insensitive contraction ('s, 'T, 'Re, etc.)
#[inline(always)]
fn check_contraction(&self) -> usize {
if self.pos >= self.len || self.bytes[self.pos] != b'\'' { return 0; }
let rem = self.len - self.pos;
if rem < 2 { return 0; }
let b1 = self.bytes[self.pos + 1] | 0x20; // lowercase
if matches!(b1, b's' | b't' | b'd' | b'm') {
if rem == 2 || !is_ascii_letter(self.bytes[self.pos + 2]) {
return 2;
}
}
if rem < 3 { return 0; }
let b2 = self.bytes[self.pos + 2] | 0x20;
if (b1 == b'l' && b2 == b'l')
|| (b1 == b'v' && b2 == b'e')
|| (b1 == b'r' && b2 == b'e')
{
return 3;
}
0
}
/// Is current byte a non-newline, non-letter, non-number character?
#[inline(always)]
fn is_prefix_char(b: u8) -> bool {
b != b'\n' && b != b'\r' && !is_ascii_letter(b) && !is_digit(b) && b < 0x80
}
/// Scan non-whitespace, non-letter, non-digit punctuation + optional trailing newlines.
#[inline(always)]
fn scan_punct_with_newlines(&mut self) {
// Punct chars
while self.pos < self.len {
let b = self.at(self.pos);
if Self::is_prefix_char(b) {
self.pos += 1;
} else if b >= 0x80 {
let (ch, _cl) = decode_utf8(&self.bytes[self.pos..]);
if !ch.is_alphabetic() && !ch.is_numeric() && !ch.is_whitespace() {
self.pos += _cl;
} else { break; }
} else {
break;
}
}
// Trailing newlines
while self.pos < self.len {
let b = self.at(self.pos);
if b == b'\n' || b == b'\r' { self.pos += 1; }
else { break; }
}
}
#[inline(always)]
fn emit(&self, start: usize) -> &'a str {
unsafe { std::str::from_utf8_unchecked(&self.bytes[start..self.pos]) }
}
}
impl<'a> Iterator for Cl100k<'a> {
type Item = &'a str;
#[inline(always)]
fn next(&mut self) -> Option<&'a str> {
if self.pos >= self.len { return None; }
let start = self.pos;
let b = self.at(self.pos);
if is_ascii_letter(b) {
// Letter run
self.pos += 1;
self.scan_letters();
} else if b == b'\'' {
// Check contraction first (case-insensitive)
let clen = self.check_contraction();
if clen > 0 {
self.pos += clen;
} else {
// Apostrophe as prefix for letters
if self.pos + 1 < self.len {
let next = self.at(self.pos + 1);
if is_ascii_letter(next) {
self.pos += 1; // consume apostrophe as prefix
self.pos += 1;
self.scan_letters();
} else if next >= 0x80 {
let (ch, _) = decode_utf8(&self.bytes[self.pos + 1..]);
if ch.is_alphabetic() {
self.pos += 1;
self.scan_letters();
} else {
// Punct run
self.pos += 1;
self.scan_punct_with_newlines();
}
} else {
self.pos += 1;
self.scan_punct_with_newlines();
}
} else {
self.pos += 1;
}
}
} else if is_digit(b) {
// Digit chunk (max 3)
self.scan_digits_chunked();
} else if b == b' ' {
// Space can prefix letters
if self.pos + 1 < self.len {
let next = self.at(self.pos + 1);
if is_ascii_letter(next) {
self.pos += 1; // consume space as prefix
self.pos += 1;
self.scan_letters();
} else if next >= 0x80 {
let (ch, _) = decode_utf8(&self.bytes[self.pos + 1..]);
if ch.is_alphabetic() {
self.pos += 1;
self.scan_letters();
} else {
// Space as punct prefix
self.pos += 1;
self.scan_punct_with_newlines();
}
} else if Self::is_prefix_char(next) || next == b'\'' {
// Space + punct run
self.pos += 1;
self.scan_punct_with_newlines();
} else {
// Bare space (or space before digit/newline)
self.pos += 1;
// Group consecutive spaces
while self.pos < self.len && self.at(self.pos) == b' ' {
self.pos += 1;
}
}
} else {
self.pos += 1;
}
} else if b == b'\n' || b == b'\r' {
// Whitespace with newline: `\s*[\r\n]` — consume preceding spaces and the newline
self.pos += 1;
while self.pos < self.len {
let c = self.at(self.pos);
if c == b'\n' || c == b'\r' || c == b' ' { self.pos += 1; }
else { break; }
}
} else if b >= 0x80 {
let (ch, cl) = decode_utf8(&self.bytes[self.pos..]);
if ch.is_alphabetic() {
self.pos += cl;
self.scan_letters();
} else if ch.is_numeric() {
self.pos += cl;
self.scan_digits_chunked();
} else if ch.is_whitespace() {
self.pos += cl;
while self.pos < self.len {
let c = self.at(self.pos);
if c == b' ' || c == b'\n' || c == b'\r' { self.pos += 1; }
else if c >= 0x80 {
let (ch2, cl2) = decode_utf8(&self.bytes[self.pos..]);
if ch2.is_whitespace() { self.pos += cl2; } else { break; }
} else { break; }
}
} else {
// Non-ASCII symbol — prefix for letters?
if self.pos + cl < self.len {
let next = self.at(self.pos + cl);
if is_ascii_letter(next) {
self.pos += cl; // consume as prefix
self.pos += 1;
self.scan_letters();
} else {
self.pos += cl;
}
} else {
self.pos += cl;
}
}
} else {
// Other ASCII punct — can prefix letters only
if self.pos + 1 < self.len {
let next = self.at(self.pos + 1);
if is_ascii_letter(next) {
self.pos += 1; // consume as prefix
self.pos += 1;
self.scan_letters();
} else if next >= 0x80 {
let (ch, _) = decode_utf8(&self.bytes[self.pos + 1..]);
if ch.is_alphabetic() {
self.pos += 1;
self.scan_letters();
} else {
// Single char — don't group non-letter prefixes
self.pos += 1;
}
} else {
// Single char — next is not a letter
self.pos += 1;
}
} else {
self.pos += 1;
}
}
debug_assert!(self.pos > start, "no progress at pos {start}");
Some(self.emit(start))
}
}
#[cfg(test)]
mod tests {
use super::*;
fn split(text: &str) -> Vec<&str> {
Cl100k::new(text).collect()
}
#[test]
fn basic_words() {
assert_eq!(split("Hello world"), vec!["Hello", " world"]);
}
#[test]
fn digit_chunking() {
assert_eq!(split("12345"), vec!["123", "45"]);
}
#[test]
fn digit_exact_3() {
assert_eq!(split("123"), vec!["123"]);
}
#[test]
fn case_insensitive_contraction() {
assert_eq!(split("DON'T"), vec!["DON", "'T"]);
}
#[test]
fn punct_prefix_letter() {
assert_eq!(split("$hello"), vec!["$hello"]);
}
#[test]
fn space_prefix_letter() {
assert_eq!(split("a b"), vec!["a", " b"]);
}
}