line_index/
lib.rs

1//! See [`LineIndex`].
2
3#![deny(missing_debug_implementations, missing_docs, rust_2018_idioms)]
4
5#[cfg(test)]
6mod tests;
7
8use nohash_hasher::IntMap;
9
10pub use text_size::{TextRange, TextSize};
11
12/// `(line, column)` information in the native, UTF-8 encoding.
13#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
14pub struct LineCol {
15    /// Zero-based.
16    pub line: u32,
17    /// Zero-based UTF-8 offset.
18    pub col: u32,
19}
20
21/// A kind of wide character encoding.
22#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
23#[non_exhaustive]
24pub enum WideEncoding {
25    /// UTF-16.
26    Utf16,
27    /// UTF-32.
28    Utf32,
29}
30
31impl WideEncoding {
32    /// Returns the number of code units it takes to encode `text` in this encoding.
33    pub fn measure(&self, text: &str) -> usize {
34        match self {
35            WideEncoding::Utf16 => text.encode_utf16().count(),
36            WideEncoding::Utf32 => text.chars().count(),
37        }
38    }
39}
40
41/// `(line, column)` information in wide encodings.
42///
43/// See [`WideEncoding`] for the kinds of wide encodings available.
44//
45// Deliberately not a generic type and different from `LineCol`.
46#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
47pub struct WideLineCol {
48    /// Zero-based.
49    pub line: u32,
50    /// Zero-based.
51    pub col: u32,
52}
53
54#[derive(Debug, Clone, Copy, PartialEq, Eq)]
55struct WideChar {
56    /// Start offset of a character inside a line, zero-based.
57    start: TextSize,
58    /// End offset of a character inside a line, zero-based.
59    end: TextSize,
60}
61
62impl WideChar {
63    /// Returns the length in 8-bit UTF-8 code units.
64    fn len(&self) -> TextSize {
65        self.end - self.start
66    }
67
68    /// Returns the length in UTF-16 or UTF-32 code units.
69    fn wide_len(&self, enc: WideEncoding) -> u32 {
70        match enc {
71            WideEncoding::Utf16 => {
72                if self.len() == TextSize::from(4) {
73                    2
74                } else {
75                    1
76                }
77            }
78            WideEncoding::Utf32 => 1,
79        }
80    }
81}
82
83/// Maps flat [`TextSize`] offsets to/from `(line, column)` representation.
84#[derive(Debug, Clone, PartialEq, Eq)]
85pub struct LineIndex {
86    /// Offset the beginning of each line (except the first, which always has offset 0).
87    newlines: Box<[TextSize]>,
88    /// List of non-ASCII characters on each line.
89    line_wide_chars: IntMap<u32, Box<[WideChar]>>,
90    /// The length of the entire text.
91    len: TextSize,
92}
93
94impl LineIndex {
95    /// Returns a `LineIndex` for the `text`.
96    pub fn new(text: &str) -> LineIndex {
97        let (newlines, line_wide_chars) = analyze_source_file(text);
98        LineIndex {
99            newlines: newlines.into_boxed_slice(),
100            line_wide_chars,
101            len: TextSize::of(text),
102        }
103    }
104
105    /// Transforms the `TextSize` into a `LineCol`.
106    ///
107    /// # Panics
108    ///
109    /// If the offset is invalid. See [`Self::try_line_col`].
110    pub fn line_col(&self, offset: TextSize) -> LineCol {
111        self.try_line_col(offset).expect("invalid offset")
112    }
113
114    /// Transforms the `TextSize` into a `LineCol`.
115    ///
116    /// Returns `None` if the `offset` was invalid, e.g. if it extends past the end of the text or
117    /// points to the middle of a multi-byte character.
118    pub fn try_line_col(&self, offset: TextSize) -> Option<LineCol> {
119        if offset > self.len {
120            return None;
121        }
122        let line = self.newlines.partition_point(|&it| it <= offset);
123        let start = self.start_offset(line)?;
124        let col = offset - start;
125        let ret = LineCol { line: line as u32, col: col.into() };
126        self.line_wide_chars
127            .get(&ret.line)
128            .into_iter()
129            .flat_map(|it| it.iter())
130            .all(|it| col <= it.start || it.end <= col)
131            .then_some(ret)
132    }
133
134    /// Transforms the `LineCol` into a `TextSize`.
135    pub fn offset(&self, line_col: LineCol) -> Option<TextSize> {
136        self.start_offset(line_col.line as usize).map(|start| start + TextSize::from(line_col.col))
137    }
138
139    fn start_offset(&self, line: usize) -> Option<TextSize> {
140        match line.checked_sub(1) {
141            None => Some(TextSize::from(0)),
142            Some(it) => self.newlines.get(it).copied(),
143        }
144    }
145
146    /// Transforms the `LineCol` with the given `WideEncoding` into a `WideLineCol`.
147    pub fn to_wide(&self, enc: WideEncoding, line_col: LineCol) -> Option<WideLineCol> {
148        let mut col = line_col.col;
149        if let Some(wide_chars) = self.line_wide_chars.get(&line_col.line) {
150            for c in wide_chars.iter() {
151                if u32::from(c.end) <= line_col.col {
152                    col = col.checked_sub(u32::from(c.len()) - c.wide_len(enc))?;
153                } else {
154                    // From here on, all utf16 characters come *after* the character we are mapping,
155                    // so we don't need to take them into account
156                    break;
157                }
158            }
159        }
160        Some(WideLineCol { line: line_col.line, col })
161    }
162
163    /// Transforms the `WideLineCol` with the given `WideEncoding` into a `LineCol`.
164    pub fn to_utf8(&self, enc: WideEncoding, line_col: WideLineCol) -> Option<LineCol> {
165        let mut col = line_col.col;
166        if let Some(wide_chars) = self.line_wide_chars.get(&line_col.line) {
167            for c in wide_chars.iter() {
168                if col > u32::from(c.start) {
169                    col = col.checked_add(u32::from(c.len()) - c.wide_len(enc))?;
170                } else {
171                    // From here on, all utf16 characters come *after* the character we are mapping,
172                    // so we don't need to take them into account
173                    break;
174                }
175            }
176        }
177        Some(LineCol { line: line_col.line, col })
178    }
179
180    /// Returns the given line's range.
181    pub fn line(&self, line: u32) -> Option<TextRange> {
182        let start = self.start_offset(line as usize)?;
183        let next_newline = self.newlines.get(line as usize).copied().unwrap_or(self.len);
184        let line_length = next_newline - start;
185        Some(TextRange::new(start, start + line_length))
186    }
187
188    /// Given a range [start, end), returns a sorted iterator of non-empty ranges [start, x1), [x1,
189    /// x2), ..., [xn, end) where all the xi, which are positions of newlines, are inside the range
190    /// [start, end).
191    pub fn lines(&self, range: TextRange) -> impl Iterator<Item = TextRange> + '_ {
192        let lo = self.newlines.partition_point(|&it| it < range.start());
193        let hi = self.newlines.partition_point(|&it| it <= range.end());
194        let all = std::iter::once(range.start())
195            .chain(self.newlines[lo..hi].iter().copied())
196            .chain(std::iter::once(range.end()));
197
198        all.clone()
199            .zip(all.skip(1))
200            .map(|(lo, hi)| TextRange::new(lo, hi))
201            .filter(|it| !it.is_empty())
202    }
203
204    /// Returns the length of the original text.
205    pub fn len(&self) -> TextSize {
206        self.len
207    }
208}
209
210/// This is adapted from the rustc_span crate, https://github.com/rust-lang/rust/blob/de59844c98f7925242a798a72c59dc3610dd0e2c/compiler/rustc_span/src/analyze_source_file.rs
211fn analyze_source_file(src: &str) -> (Vec<TextSize>, IntMap<u32, Box<[WideChar]>>) {
212    assert!(src.len() < !0u32 as usize);
213    let mut lines = vec![];
214    let mut line_wide_chars = IntMap::<u32, Vec<WideChar>>::default();
215
216    // Calls the right implementation, depending on hardware support available.
217    analyze_source_file_dispatch(src, &mut lines, &mut line_wide_chars);
218
219    (lines, line_wide_chars.into_iter().map(|(k, v)| (k, v.into_boxed_slice())).collect())
220}
221
222#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
223fn analyze_source_file_dispatch(
224    src: &str,
225    lines: &mut Vec<TextSize>,
226    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
227) {
228    if is_x86_feature_detected!("sse2") {
229        // SAFETY: SSE2 support was checked
230        unsafe {
231            analyze_source_file_sse2(src, lines, multi_byte_chars);
232        }
233    } else {
234        analyze_source_file_generic(src, src.len(), TextSize::from(0), lines, multi_byte_chars);
235    }
236}
237
238#[cfg(target_arch = "aarch64")]
239fn analyze_source_file_dispatch(
240    src: &str,
241    lines: &mut Vec<TextSize>,
242    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
243) {
244    if std::arch::is_aarch64_feature_detected!("neon") {
245        // SAFETY: NEON support was checked
246        unsafe {
247            analyze_source_file_neon(src, lines, multi_byte_chars);
248        }
249    } else {
250        analyze_source_file_generic(src, src.len(), TextSize::from(0), lines, multi_byte_chars);
251    }
252}
253
254/// Checks 16 byte chunks of text at a time. If the chunk contains
255/// something other than printable ASCII characters and newlines, the
256/// function falls back to the generic implementation. Otherwise it uses
257/// SSE2 intrinsics to quickly find all newlines.
258#[target_feature(enable = "sse2")]
259#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
260unsafe fn analyze_source_file_sse2(
261    src: &str,
262    lines: &mut Vec<TextSize>,
263    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
264) {
265    #[cfg(target_arch = "x86")]
266    use std::arch::x86::*;
267    #[cfg(target_arch = "x86_64")]
268    use std::arch::x86_64::*;
269
270    const CHUNK_SIZE: usize = 16;
271
272    let src_bytes = src.as_bytes();
273
274    let chunk_count = src.len() / CHUNK_SIZE;
275
276    // This variable keeps track of where we should start decoding a
277    // chunk. If a multi-byte character spans across chunk boundaries,
278    // we need to skip that part in the next chunk because we already
279    // handled it.
280    let mut intra_chunk_offset = 0;
281
282    for chunk_index in 0..chunk_count {
283        let ptr = src_bytes.as_ptr() as *const __m128i;
284        // We don't know if the pointer is aligned to 16 bytes, so we
285        // use `loadu`, which supports unaligned loading.
286        let chunk = unsafe { _mm_loadu_si128(ptr.add(chunk_index)) };
287
288        // For character in the chunk, see if its byte value is < 0, which
289        // indicates that it's part of a UTF-8 char.
290        let multibyte_test = unsafe { _mm_cmplt_epi8(chunk, _mm_set1_epi8(0)) };
291        // Create a bit mask from the comparison results.
292        let multibyte_mask = unsafe { _mm_movemask_epi8(multibyte_test) };
293
294        // If the bit mask is all zero, we only have ASCII chars here:
295        if multibyte_mask == 0 {
296            assert!(intra_chunk_offset == 0);
297
298            // Check for newlines in the chunk
299            let newlines_test = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8)) };
300            let newlines_mask = unsafe { _mm_movemask_epi8(newlines_test) };
301
302            if newlines_mask != 0 {
303                // All control characters are newlines, record them
304                let mut newlines_mask = 0xFFFF0000 | newlines_mask as u32;
305                let output_offset = TextSize::from((chunk_index * CHUNK_SIZE + 1) as u32);
306
307                loop {
308                    let index = newlines_mask.trailing_zeros();
309
310                    if index >= CHUNK_SIZE as u32 {
311                        // We have arrived at the end of the chunk.
312                        break;
313                    }
314
315                    lines.push(TextSize::from(index) + output_offset);
316
317                    // Clear the bit, so we can find the next one.
318                    newlines_mask &= (!1) << index;
319                }
320            }
321            continue;
322        }
323
324        // The slow path.
325        // There are control chars in here, fallback to generic decoding.
326        let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
327        intra_chunk_offset = analyze_source_file_generic(
328            &src[scan_start..],
329            CHUNK_SIZE - intra_chunk_offset,
330            TextSize::from(scan_start as u32),
331            lines,
332            multi_byte_chars,
333        );
334    }
335
336    // There might still be a tail left to analyze
337    let tail_start = chunk_count * CHUNK_SIZE + intra_chunk_offset;
338    if tail_start < src.len() {
339        analyze_source_file_generic(
340            &src[tail_start..],
341            src.len() - tail_start,
342            TextSize::from(tail_start as u32),
343            lines,
344            multi_byte_chars,
345        );
346    }
347}
348
349#[target_feature(enable = "neon")]
350#[cfg(target_arch = "aarch64")]
351#[inline]
352// See https://community.arm.com/arm-community-blogs/b/infrastructure-solutions-blog/posts/porting-x86-vector-bitmask-optimizations-to-arm-neon
353//
354// The mask is a 64-bit integer, where each 4-bit corresponds to a u8 in the
355// input vector. The least significant 4 bits correspond to the first byte in
356// the vector.
357unsafe fn move_mask(v: std::arch::aarch64::uint8x16_t) -> u64 {
358    use std::arch::aarch64::*;
359
360    let nibble_mask = unsafe { vshrn_n_u16(vreinterpretq_u16_u8(v), 4) };
361    unsafe { vget_lane_u64(vreinterpret_u64_u8(nibble_mask), 0) }
362}
363
364#[target_feature(enable = "neon")]
365#[cfg(target_arch = "aarch64")]
366unsafe fn analyze_source_file_neon(
367    src: &str,
368    lines: &mut Vec<TextSize>,
369    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
370) {
371    use std::arch::aarch64::*;
372
373    const CHUNK_SIZE: usize = 16;
374
375    let src_bytes = src.as_bytes();
376
377    let chunk_count = src.len() / CHUNK_SIZE;
378
379    let newline = unsafe { vdupq_n_s8(b'\n' as i8) };
380
381    // This variable keeps track of where we should start decoding a
382    // chunk. If a multi-byte character spans across chunk boundaries,
383    // we need to skip that part in the next chunk because we already
384    // handled it.
385    let mut intra_chunk_offset = 0;
386
387    for chunk_index in 0..chunk_count {
388        let ptr = src_bytes.as_ptr() as *const i8;
389        let chunk = unsafe { vld1q_s8(ptr.add(chunk_index * CHUNK_SIZE)) };
390
391        // For character in the chunk, see if its byte value is < 0, which
392        // indicates that it's part of a UTF-8 char.
393        let multibyte_test = unsafe { vcltzq_s8(chunk) };
394        // Create a bit mask from the comparison results.
395        let multibyte_mask = unsafe { move_mask(multibyte_test) };
396
397        // If the bit mask is all zero, we only have ASCII chars here:
398        if multibyte_mask == 0 {
399            assert!(intra_chunk_offset == 0);
400
401            // Check for newlines in the chunk
402            let newlines_test = unsafe { vceqq_s8(chunk, newline) };
403            let mut newlines_mask = unsafe { move_mask(newlines_test) };
404
405            // If the bit mask is not all zero, there are newlines in this chunk.
406            if newlines_mask != 0 {
407                let output_offset = TextSize::from((chunk_index * CHUNK_SIZE + 1) as u32);
408
409                while newlines_mask != 0 {
410                    let trailing_zeros = newlines_mask.trailing_zeros();
411                    let index = trailing_zeros / 4;
412
413                    lines.push(TextSize::from(index) + output_offset);
414
415                    // Clear the current 4-bit, so we can find the next one.
416                    newlines_mask &= (!0xF) << trailing_zeros;
417                }
418            }
419            continue;
420        }
421
422        let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
423        intra_chunk_offset = analyze_source_file_generic(
424            &src[scan_start..],
425            CHUNK_SIZE - intra_chunk_offset,
426            TextSize::from(scan_start as u32),
427            lines,
428            multi_byte_chars,
429        );
430    }
431
432    let tail_start = chunk_count * CHUNK_SIZE + intra_chunk_offset;
433    if tail_start < src.len() {
434        analyze_source_file_generic(
435            &src[tail_start..],
436            src.len() - tail_start,
437            TextSize::from(tail_start as u32),
438            lines,
439            multi_byte_chars,
440        );
441    }
442}
443
444#[cfg(not(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64")))]
445// The target (or compiler version) does not support SSE2 ...
446fn analyze_source_file_dispatch(
447    src: &str,
448    lines: &mut Vec<TextSize>,
449    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
450) {
451    analyze_source_file_generic(src, src.len(), TextSize::from(0), lines, multi_byte_chars);
452}
453
454// `scan_len` determines the number of bytes in `src` to scan. Note that the
455// function can read past `scan_len` if a multi-byte character start within the
456// range but extends past it. The overflow is returned by the function.
457fn analyze_source_file_generic(
458    src: &str,
459    scan_len: usize,
460    output_offset: TextSize,
461    lines: &mut Vec<TextSize>,
462    multi_byte_chars: &mut IntMap<u32, Vec<WideChar>>,
463) -> usize {
464    assert!(src.len() >= scan_len);
465    let mut i = 0;
466    let src_bytes = src.as_bytes();
467
468    while i < scan_len {
469        let byte = unsafe {
470            // We verified that i < scan_len <= src.len()
471            *src_bytes.get_unchecked(i)
472        };
473
474        // How much to advance in order to get to the next UTF-8 char in the
475        // string.
476        let mut char_len = 1;
477
478        if byte == b'\n' {
479            lines.push(TextSize::from(i as u32 + 1) + output_offset);
480        } else if byte >= 127 {
481            // The slow path: Just decode to `char`.
482            let c = src[i..].chars().next().unwrap();
483            char_len = c.len_utf8();
484
485            // The last element of `lines` represents the offset of the start of
486            // current line. To get the offset inside the line, we subtract it.
487            let pos = TextSize::from(i as u32) + output_offset
488                - lines.last().unwrap_or(&TextSize::default());
489
490            if char_len > 1 {
491                assert!((2..=4).contains(&char_len));
492                let mbc = WideChar { start: pos, end: pos + TextSize::from(char_len as u32) };
493                multi_byte_chars.entry(lines.len() as u32).or_default().push(mbc);
494            }
495        }
496
497        i += char_len;
498    }
499
500    i - scan_len
501}
line_index/lib.rs

line_index/
lib.rs
