squonk 2.0.0

Extensible, fast, multi-dialect SQL tokenizer and parser for Rust
Documentation 
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

// SPDX-License-Identifier: MIT
// Copyright (c) 2026 Moderately AI Inc.

//! Byte-offset source cursor (rustc_lexer style).
//!
//! `Cursor` is the low-level navigation layer: it peeks and advances over the
//! source's *bytes* and tracks a `u32` byte offset. It carries no notion of
//! tokens; the scanner ([`super::scan`]) drives it to recognize lexemes.

/// A byte-offset cursor over UTF-8 source text.
///
/// `pos` is a byte offset, so it lines up directly with [`Span`] offsets and the
/// `&source[..]` slicing used to recover token text. The cursor stops on byte
/// boundaries that are always char boundaries, because every scanner consumes a
/// multi-byte UTF-8 sequence whole — identifiers char-by-char ([`bump_char`]),
/// quoted bodies by running to their ASCII delimiter — so slicing the source at a
/// cursor offset never splits a `char`.
///
/// [`bump_char`]: Cursor::bump_char
///
/// # Source length invariant
///
/// Offsets are `u32`. The cursor therefore assumes `src.len() <= u32::MAX`;
/// [`tokenize`] enforces this before constructing a cursor. Constructing a
/// cursor directly over a longer source and advancing past `u32::MAX` would
/// overflow `pos`.
///
/// [`Span`]: crate::ast::Span
/// [`tokenize`]: crate::tokenizer::tokenize
#[derive(Clone, Copy, Debug)]
pub struct Cursor<'a> {
    src: &'a str,
    pos: u32,
}

impl<'a> Cursor<'a> {
    /// Create a cursor at the start of `src`.
    pub fn new(src: &'a str) -> Self {
        debug_assert!(
            u32::try_from(src.len()).is_ok(),
            "Cursor source length must fit in u32; tokenize() guards this",
        );
        Self { src, pos: 0 }
    }

    /// The source this cursor walks.
    ///
    /// Returns the `&'a str` (a `Copy` reborrow), so callers can hold the source
    /// and derive `&'a` slices from it while still mutating the cursor.
    pub fn src(&self) -> &'a str {
        self.src
    }

    /// The current byte offset.
    pub fn pos(&self) -> u32 {
        self.pos
    }

    /// Rewind or jump the cursor to an earlier (or equal) absolute byte offset.
    ///
    /// Used by speculative lexers that must abandon a look-ahead (e.g. string
    /// continuation that does not actually continue). `pos` must stay within the
    /// source and on a UTF-8 boundary the caller already landed on via prior bumps.
    pub fn set_pos(&mut self, pos: u32) {
        debug_assert!(
            (pos as usize) <= self.src.len(),
            "set_pos must stay within the source",
        );
        self.pos = pos;
    }

    /// True once the cursor has reached the end of the source.
    pub fn is_eof(&self) -> bool {
        self.pos as usize >= self.src.len()
    }

    /// The byte at the cursor, or `None` at end of input.
    pub fn peek(&self) -> Option<u8> {
        self.src.as_bytes().get(self.pos as usize).copied()
    }

    /// The byte `n` positions ahead of the cursor, without advancing.
    ///
    /// The index is computed in `usize` so a large `n` cannot overflow the `u32`
    /// offset; out-of-range lookups return `None`.
    pub fn peek_nth(&self, n: u32) -> Option<u8> {
        let index = self.pos as usize + n as usize;
        self.src.as_bytes().get(index).copied()
    }

    /// The not-yet-consumed bytes, `&source[pos..]`.
    ///
    /// Borrows the source for `'a`, not the cursor, so the caller may keep the
    /// slice while continuing to advance the cursor.
    pub fn rest(&self) -> &'a [u8] {
        &self.src.as_bytes()[self.pos as usize..]
    }

    /// Advance one byte, returning the byte moved past, or `None` at end of input.
    pub fn bump(&mut self) -> Option<u8> {
        let byte = self.peek()?;
        // `peek` returned `Some`, so `pos < len <= u32::MAX`; the increment
        // cannot overflow under the source-length invariant.
        self.pos += 1;
        Some(byte)
    }

    /// Advance the cursor by `n` bytes in one step.
    ///
    /// A bulk [`bump`](Self::bump) for a run the caller has already confirmed is
    /// present — a fixed operator width, a measured prefix, a whole multi-byte
    /// character — so the per-byte bounds check `bump` repeats is skipped on warm
    /// paths. The caller owes the in-bounds guarantee (`pos + n <= len`), which the
    /// `debug_assert!` verifies in debug builds; the direct `pos` write then mirrors
    /// [`bump_char`](Self::bump_char), staying within the source-length invariant.
    pub fn advance_bytes(&mut self, n: u32) {
        debug_assert!(
            self.pos as usize + n as usize <= self.src.len(),
            "advance_bytes must stay within the source",
        );
        self.pos += n;
    }

    /// Advance while `pred` holds for the byte at the cursor.
    ///
    /// Stops at the first non-matching byte or at end of input.
    pub fn eat_while(&mut self, mut pred: impl FnMut(u8) -> bool) {
        while let Some(byte) = self.peek() {
            if !pred(byte) {
                break;
            }
            self.pos += 1;
        }
    }

    /// The whole UTF-8 character beginning `n` bytes ahead of the cursor, or `None`
    /// at or past end of input.
    ///
    /// Used where a lexeme's class is a per-*character* Unicode property rather than
    /// a per-byte one (non-ASCII identifier characters). `pos + n` must fall on a
    /// char boundary; every caller offsets by whole ASCII bytes (the cursor itself,
    /// or one past an ASCII sigil), so it always does — and a non-boundary offset
    /// yields `None` rather than splitting a character, because the slice is taken
    /// with [`str::get`].
    pub fn char_at(&self, n: u32) -> Option<char> {
        self.src
            .get(self.pos as usize + n as usize..)?
            .chars()
            .next()
    }

    /// Advance past one whole UTF-8 character, returning it, or `None` at end of
    /// input.
    ///
    /// Unlike [`bump`](Self::bump), which moves a single byte, this consumes every
    /// byte of a multi-byte character, so the cursor always stops on a char boundary
    /// — the invariant that lets a token span slice back to valid UTF-8.
    pub fn bump_char(&mut self) -> Option<char> {
        let ch = self.char_at(0)?;
        // `pos < len <= u32::MAX` and `pos + ch.len_utf8() <= len`, so this stays
        // within the source-length invariant and cannot overflow.
        self.pos += ch.len_utf8() as u32;
        Some(ch)
    }
}

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

    #[test]
    fn peek_and_bump_walk_bytes_and_track_offset() {
        let mut cursor = Cursor::new("ab");

        assert_eq!(cursor.pos(), 0);
        assert!(!cursor.is_eof());
        assert_eq!(cursor.peek(), Some(b'a'));
        assert_eq!(cursor.peek_nth(1), Some(b'b'));
        assert_eq!(cursor.peek_nth(2), None);

        assert_eq!(cursor.bump(), Some(b'a'));
        assert_eq!(cursor.pos(), 1);
        assert_eq!(cursor.bump(), Some(b'b'));
        assert_eq!(cursor.pos(), 2);
        assert!(cursor.is_eof());
        assert_eq!(cursor.bump(), None);
        assert_eq!(cursor.pos(), 2);
    }

    #[test]
    fn eat_while_stops_at_first_non_match_and_at_eof() {
        let mut cursor = Cursor::new("123abc");
        cursor.eat_while(|b| b.is_ascii_digit());
        assert_eq!(cursor.pos(), 3);
        assert_eq!(cursor.peek(), Some(b'a'));

        cursor.eat_while(|b| b.is_ascii_alphabetic());
        assert_eq!(cursor.pos(), 6);
        assert!(cursor.is_eof());
    }

    #[test]
    fn rest_returns_remaining_bytes_from_offset() {
        let mut cursor = Cursor::new("a$tag$");
        cursor.bump();
        assert_eq!(cursor.rest(), b"$tag$");
    }

    #[test]
    fn char_at_and_bump_char_walk_whole_characters() {
        // `é` is two UTF-8 bytes, `δ` is two, `🎉` is four: `bump_char` must advance
        // by the character's byte length so the offset stays on a char boundary.
        let mut cursor = Cursor::new("aé🎉");
        assert_eq!(cursor.char_at(0), Some('a'));
        assert_eq!(cursor.char_at(1), Some('é'));

        assert_eq!(cursor.bump_char(), Some('a'));
        assert_eq!(cursor.pos(), 1);
        assert_eq!(cursor.bump_char(), Some('é'));
        assert_eq!(cursor.pos(), 3); // skipped both bytes of `é`
        assert_eq!(cursor.bump_char(), Some('🎉'));
        assert_eq!(cursor.pos(), 7); // skipped all four bytes of `🎉`
        assert_eq!(cursor.bump_char(), None);

        // An offset that lands mid-character yields `None` rather than panicking.
        let cursor = Cursor::new("é");
        assert_eq!(cursor.char_at(1), None);
    }
}