regex_automata/util/
utf8.rs

1/*!
2Utilities for dealing with UTF-8.
3
4This module provides some UTF-8 related helper routines, including an
5incremental decoder.
6*/
7
8/// Returns true if and only if the given byte is considered a word character.
9/// This only applies to ASCII.
10///
11/// This was copied from regex-syntax so that we can use it to determine the
12/// starting DFA state while searching without depending on regex-syntax. The
13/// definition is never going to change, so there's no maintenance/bit-rot
14/// hazard here.
15#[cfg_attr(feature = "perf-inline", inline(always))]
16pub(crate) fn is_word_byte(b: u8) -> bool {
17    const fn mkwordset() -> [bool; 256] {
18        // FIXME: Use as_usize() once const functions in traits are stable.
19        let mut set = [false; 256];
20        set[b'_' as usize] = true;
21
22        let mut byte = b'0';
23        while byte <= b'9' {
24            set[byte as usize] = true;
25            byte += 1;
26        }
27        byte = b'A';
28        while byte <= b'Z' {
29            set[byte as usize] = true;
30            byte += 1;
31        }
32        byte = b'a';
33        while byte <= b'z' {
34            set[byte as usize] = true;
35            byte += 1;
36        }
37        set
38    }
39    const WORD: [bool; 256] = mkwordset();
40    WORD[b as usize]
41}
42
43/// Decodes the next UTF-8 encoded codepoint from the given byte slice.
44///
45/// If no valid encoding of a codepoint exists at the beginning of the given
46/// byte slice, then the first byte is returned instead.
47///
48/// This returns `None` if and only if `bytes` is empty.
49///
50/// This never panics.
51///
52/// *WARNING*: This is not designed for performance. If you're looking for a
53/// fast UTF-8 decoder, this is not it. If you feel like you need one in this
54/// crate, then please file an issue and discuss your use case.
55#[cfg_attr(feature = "perf-inline", inline(always))]
56pub(crate) fn decode(bytes: &[u8]) -> Option<Result<char, u8>> {
57    if bytes.is_empty() {
58        return None;
59    }
60    let len = match len(bytes[0]) {
61        None => return Some(Err(bytes[0])),
62        Some(len) if len > bytes.len() => return Some(Err(bytes[0])),
63        Some(1) => return Some(Ok(char::from(bytes[0]))),
64        Some(len) => len,
65    };
66    match core::str::from_utf8(&bytes[..len]) {
67        Ok(s) => Some(Ok(s.chars().next().unwrap())),
68        Err(_) => Some(Err(bytes[0])),
69    }
70}
71
72/// Decodes the last UTF-8 encoded codepoint from the given byte slice.
73///
74/// If no valid encoding of a codepoint exists at the end of the given byte
75/// slice, then the last byte is returned instead.
76///
77/// This returns `None` if and only if `bytes` is empty.
78#[cfg_attr(feature = "perf-inline", inline(always))]
79pub(crate) fn decode_last(bytes: &[u8]) -> Option<Result<char, u8>> {
80    if bytes.is_empty() {
81        return None;
82    }
83    let mut start = bytes.len() - 1;
84    let limit = bytes.len().saturating_sub(4);
85    while start > limit && !is_leading_or_invalid_byte(bytes[start]) {
86        start -= 1;
87    }
88    match decode(&bytes[start..]) {
89        None => None,
90        Some(Ok(ch)) => Some(Ok(ch)),
91        Some(Err(_)) => Some(Err(bytes[bytes.len() - 1])),
92    }
93}
94
95/// Given a UTF-8 leading byte, this returns the total number of code units
96/// in the following encoded codepoint.
97///
98/// If the given byte is not a valid UTF-8 leading byte, then this returns
99/// `None`.
100#[cfg_attr(feature = "perf-inline", inline(always))]
101fn len(byte: u8) -> Option<usize> {
102    if byte <= 0x7F {
103        return Some(1);
104    } else if byte & 0b1100_0000 == 0b1000_0000 {
105        return None;
106    } else if byte <= 0b1101_1111 {
107        Some(2)
108    } else if byte <= 0b1110_1111 {
109        Some(3)
110    } else if byte <= 0b1111_0111 {
111        Some(4)
112    } else {
113        None
114    }
115}
116
117/// Returns true if and only if the given offset in the given bytes falls on a
118/// valid UTF-8 encoded codepoint boundary.
119///
120/// If `bytes` is not valid UTF-8, then the behavior of this routine is
121/// unspecified.
122#[cfg_attr(feature = "perf-inline", inline(always))]
123pub(crate) fn is_boundary(bytes: &[u8], i: usize) -> bool {
124    match bytes.get(i) {
125        // The position at the end of the bytes always represents an empty
126        // string, which is a valid boundary. But anything after that doesn't
127        // make much sense to call valid a boundary.
128        None => i == bytes.len(),
129        // Other than ASCII (where the most significant bit is never set),
130        // valid starting bytes always have their most significant two bits
131        // set, where as continuation bytes never have their second most
132        // significant bit set. Therefore, this only returns true when bytes[i]
133        // corresponds to a byte that begins a valid UTF-8 encoding of a
134        // Unicode scalar value.
135        Some(&b) => b <= 0b0111_1111 || b >= 0b1100_0000,
136    }
137}
138
139/// Returns true if and only if the given byte is either a valid leading UTF-8
140/// byte, or is otherwise an invalid byte that can never appear anywhere in a
141/// valid UTF-8 sequence.
142#[cfg_attr(feature = "perf-inline", inline(always))]
143fn is_leading_or_invalid_byte(b: u8) -> bool {
144    // In the ASCII case, the most significant bit is never set. The leading
145    // byte of a 2/3/4-byte sequence always has the top two most significant
146    // bits set. For bytes that can never appear anywhere in valid UTF-8, this
147    // also returns true, since every such byte has its two most significant
148    // bits set:
149    //
150    //     \xC0 :: 11000000
151    //     \xC1 :: 11000001
152    //     \xF5 :: 11110101
153    //     \xF6 :: 11110110
154    //     \xF7 :: 11110111
155    //     \xF8 :: 11111000
156    //     \xF9 :: 11111001
157    //     \xFA :: 11111010
158    //     \xFB :: 11111011
159    //     \xFC :: 11111100
160    //     \xFD :: 11111101
161    //     \xFE :: 11111110
162    //     \xFF :: 11111111
163    (b & 0b1100_0000) != 0b1000_0000
164}
165
166/*
167/// Returns the smallest possible index of the next valid UTF-8 sequence
168/// starting after `i`.
169///
170/// For all inputs, including invalid UTF-8 and any value of `i`, the return
171/// value is guaranteed to be greater than `i`. (If there is no value greater
172/// than `i` that fits in `usize`, then this panics.)
173///
174/// Generally speaking, this should only be called on `text` when it is
175/// permitted to assume that it is valid UTF-8 and where either `i >=
176/// text.len()` or where `text[i]` is a leading byte of a UTF-8 sequence.
177///
178/// NOTE: This method was used in a previous conception of iterators where we
179/// specifically tried to skip over empty matches that split a codepoint by
180/// simply requiring that our next search begin at the beginning of codepoint.
181/// But we ended up changing that technique to always advance by 1 byte and
182/// then filter out matches that split a codepoint after-the-fact. Thus, we no
183/// longer use this method. But I've kept it around in case we want to switch
184/// back to this approach. Its guarantees are a little subtle, so I'd prefer
185/// not to rebuild it from whole cloth.
186pub(crate) fn next(text: &[u8], i: usize) -> usize {
187    let b = match text.get(i) {
188        None => return i.checked_add(1).unwrap(),
189        Some(&b) => b,
190    };
191    // For cases where we see an invalid UTF-8 byte, there isn't much we can do
192    // other than just start at the next byte.
193    let inc = len(b).unwrap_or(1);
194    i.checked_add(inc).unwrap()
195}
196*/