1//! Heap-allocated counterpart to core `wtf8` module.
2#![unstable(
3 feature = "wtf8_internals",
4 issue = "none",
5 reason = "this is internal code for representing OsStr on some platforms and not a public API"
6)]
7// rustdoc bug: doc(hidden) on the module won't stop types in the module from showing up in trait
8// implementations, so, we'll have to add more doc(hidden)s anyway
9#![doc(hidden)]
10
11// Note: This module is also included in the alloctests crate using #[path] to
12// run the tests. See the comment there for an explanation why this is the case.
13
14#[cfg(test)]
15mod tests;
16
17use core::char::encode_utf8_raw;
18use core::hash::{Hash, Hasher};
19pub use core::wtf8::{CodePoint, Wtf8};
20#[cfg(not(test))]
21pub use core::wtf8::{EncodeWide, Wtf8CodePoints};
22use core::{fmt, mem, ops, str};
23
24use crate::borrow::{Cow, ToOwned};
25use crate::boxed::Box;
26use crate::collections::TryReserveError;
27#[cfg(not(test))]
28use crate::rc::Rc;
29use crate::string::String;
30#[cfg(all(not(test), target_has_atomic = "ptr"))]
31use crate::sync::Arc;
32use crate::vec::Vec;
33
34/// An owned, growable string of well-formed WTF-8 data.
35///
36/// Similar to `String`, but can additionally contain surrogate code points
37/// if they’re not in a surrogate pair.
38#[derive(Eq, PartialEq, Ord, PartialOrd, Clone)]
39#[doc(hidden)]
40pub struct Wtf8Buf {
41 bytes: Vec<u8>,
42
43 /// Do we know that `bytes` holds a valid UTF-8 encoding? We can easily
44 /// know this if we're constructed from a `String` or `&str`.
45 ///
46 /// It is possible for `bytes` to have valid UTF-8 without this being
47 /// set, such as when we're concatenating `&Wtf8`'s and surrogates become
48 /// paired, as we don't bother to rescan the entire string.
49 is_known_utf8: bool,
50}
51
52impl ops::Deref for Wtf8Buf {
53 type Target = Wtf8;
54
55 fn deref(&self) -> &Wtf8 {
56 self.as_slice()
57 }
58}
59
60impl ops::DerefMut for Wtf8Buf {
61 fn deref_mut(&mut self) -> &mut Wtf8 {
62 self.as_mut_slice()
63 }
64}
65
66/// Formats the string in double quotes, with characters escaped according to
67/// [`char::escape_debug`] and unpaired surrogates represented as `\u{xxxx}`,
68/// where each `x` is a hexadecimal digit.
69///
70/// For example, the code units [U+0061, U+D800, U+000A] are formatted as
71/// `"a\u{D800}\n"`.
72impl fmt::Debug for Wtf8Buf {
73 #[inline]
74 fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
75 fmt::Debug::fmt(&**self, formatter)
76 }
77}
78
79/// Formats the string with unpaired surrogates substituted with the replacement
80/// character, U+FFFD.
81impl fmt::Display for Wtf8Buf {
82 fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
83 if let Some(s) = self.as_known_utf8() {
84 fmt::Display::fmt(s, formatter)
85 } else {
86 fmt::Display::fmt(&**self, formatter)
87 }
88 }
89}
90
91#[cfg_attr(test, allow(dead_code))]
92impl Wtf8Buf {
93 /// Creates a new, empty WTF-8 string.
94 #[inline]
95 pub fn new() -> Wtf8Buf {
96 Wtf8Buf { bytes: Vec::new(), is_known_utf8: true }
97 }
98
99 /// Creates a new, empty WTF-8 string with pre-allocated capacity for `capacity` bytes.
100 #[inline]
101 pub fn with_capacity(capacity: usize) -> Wtf8Buf {
102 Wtf8Buf { bytes: Vec::with_capacity(capacity), is_known_utf8: true }
103 }
104
105 /// Creates a WTF-8 string from a WTF-8 byte vec.
106 ///
107 /// Since the byte vec is not checked for valid WTF-8, this function is
108 /// marked unsafe.
109 #[inline]
110 pub unsafe fn from_bytes_unchecked(value: Vec<u8>) -> Wtf8Buf {
111 Wtf8Buf { bytes: value, is_known_utf8: false }
112 }
113
114 /// Creates a WTF-8 string from a UTF-8 `String`.
115 ///
116 /// This takes ownership of the `String` and does not copy.
117 ///
118 /// Since WTF-8 is a superset of UTF-8, this always succeeds.
119 #[inline]
120 pub const fn from_string(string: String) -> Wtf8Buf {
121 Wtf8Buf { bytes: string.into_bytes(), is_known_utf8: true }
122 }
123
124 /// Creates a WTF-8 string from a UTF-8 `&str` slice.
125 ///
126 /// This copies the content of the slice.
127 ///
128 /// Since WTF-8 is a superset of UTF-8, this always succeeds.
129 #[inline]
130 pub fn from_str(s: &str) -> Wtf8Buf {
131 Wtf8Buf { bytes: s.as_bytes().to_vec(), is_known_utf8: true }
132 }
133
134 pub fn clear(&mut self) {
135 self.bytes.clear();
136 self.is_known_utf8 = true;
137 }
138
139 /// Creates a WTF-8 string from a potentially ill-formed UTF-16 slice of 16-bit code units.
140 ///
141 /// This is lossless: calling `.encode_wide()` on the resulting string
142 /// will always return the original code units.
143 pub fn from_wide(v: &[u16]) -> Wtf8Buf {
144 let mut string = Wtf8Buf::with_capacity(v.len());
145 for item in char::decode_utf16(v.iter().cloned()) {
146 match item {
147 Ok(ch) => string.push_char(ch),
148 Err(surrogate) => {
149 let surrogate = surrogate.unpaired_surrogate();
150 // Surrogates are known to be in the code point range.
151 let code_point = unsafe { CodePoint::from_u32_unchecked(surrogate as u32) };
152 // The string will now contain an unpaired surrogate.
153 string.is_known_utf8 = false;
154 // Skip the WTF-8 concatenation check,
155 // surrogate pairs are already decoded by decode_utf16
156 unsafe {
157 string.push_code_point_unchecked(code_point);
158 }
159 }
160 }
161 }
162 string
163 }
164
165 /// Appends the given `char` to the end of this string.
166 /// This does **not** include the WTF-8 concatenation check or `is_known_utf8` check.
167 /// Copied from String::push.
168 unsafe fn push_code_point_unchecked(&mut self, code_point: CodePoint) {
169 let mut bytes = [0; char::MAX_LEN_UTF8];
170 let bytes = encode_utf8_raw(code_point.to_u32(), &mut bytes);
171 self.bytes.extend_from_slice(bytes)
172 }
173
174 #[inline]
175 pub fn as_slice(&self) -> &Wtf8 {
176 unsafe { Wtf8::from_bytes_unchecked(&self.bytes) }
177 }
178
179 #[inline]
180 pub fn as_mut_slice(&mut self) -> &mut Wtf8 {
181 // Safety: `Wtf8` doesn't expose any way to mutate the bytes that would
182 // cause them to change from well-formed UTF-8 to ill-formed UTF-8,
183 // which would break the assumptions of the `is_known_utf8` field.
184 unsafe { Wtf8::from_mut_bytes_unchecked(&mut self.bytes) }
185 }
186
187 /// Converts the string to UTF-8 without validation, if it was created from
188 /// valid UTF-8.
189 #[inline]
190 fn as_known_utf8(&self) -> Option<&str> {
191 if self.is_known_utf8 {
192 // SAFETY: The buffer is known to be valid UTF-8.
193 Some(unsafe { str::from_utf8_unchecked(self.as_bytes()) })
194 } else {
195 None
196 }
197 }
198
199 /// Reserves capacity for at least `additional` more bytes to be inserted
200 /// in the given `Wtf8Buf`.
201 /// The collection may reserve more space to avoid frequent reallocations.
202 ///
203 /// # Panics
204 ///
205 /// Panics if the new capacity exceeds `isize::MAX` bytes.
206 #[inline]
207 pub fn reserve(&mut self, additional: usize) {
208 self.bytes.reserve(additional)
209 }
210
211 /// Tries to reserve capacity for at least `additional` more bytes to be
212 /// inserted in the given `Wtf8Buf`. The `Wtf8Buf` may reserve more space to
213 /// avoid frequent reallocations. After calling `try_reserve`, capacity will
214 /// be greater than or equal to `self.len() + additional`. Does nothing if
215 /// capacity is already sufficient. This method preserves the contents even
216 /// if an error occurs.
217 ///
218 /// # Errors
219 ///
220 /// If the capacity overflows, or the allocator reports a failure, then an error
221 /// is returned.
222 #[inline]
223 pub fn try_reserve(&mut self, additional: usize) -> Result<(), TryReserveError> {
224 self.bytes.try_reserve(additional)
225 }
226
227 #[inline]
228 pub fn reserve_exact(&mut self, additional: usize) {
229 self.bytes.reserve_exact(additional)
230 }
231
232 /// Tries to reserve the minimum capacity for exactly `additional` more
233 /// bytes to be inserted in the given `Wtf8Buf`. After calling
234 /// `try_reserve_exact`, capacity will be greater than or equal to
235 /// `self.len() + additional` if it returns `Ok(())`.
236 /// Does nothing if the capacity is already sufficient.
237 ///
238 /// Note that the allocator may give the `Wtf8Buf` more space than it
239 /// requests. Therefore, capacity can not be relied upon to be precisely
240 /// minimal. Prefer [`try_reserve`] if future insertions are expected.
241 ///
242 /// [`try_reserve`]: Wtf8Buf::try_reserve
243 ///
244 /// # Errors
245 ///
246 /// If the capacity overflows, or the allocator reports a failure, then an error
247 /// is returned.
248 #[inline]
249 pub fn try_reserve_exact(&mut self, additional: usize) -> Result<(), TryReserveError> {
250 self.bytes.try_reserve_exact(additional)
251 }
252
253 #[inline]
254 pub fn shrink_to_fit(&mut self) {
255 self.bytes.shrink_to_fit()
256 }
257
258 #[inline]
259 pub fn shrink_to(&mut self, min_capacity: usize) {
260 self.bytes.shrink_to(min_capacity)
261 }
262
263 #[inline]
264 pub fn leak<'a>(self) -> &'a mut Wtf8 {
265 unsafe { Wtf8::from_mut_bytes_unchecked(self.bytes.leak()) }
266 }
267
268 /// Returns the number of bytes that this string buffer can hold without reallocating.
269 #[inline]
270 pub fn capacity(&self) -> usize {
271 self.bytes.capacity()
272 }
273
274 /// Append a UTF-8 slice at the end of the string.
275 #[inline]
276 pub fn push_str(&mut self, other: &str) {
277 self.bytes.extend_from_slice(other.as_bytes())
278 }
279
280 /// Append a WTF-8 slice at the end of the string.
281 ///
282 /// This replaces newly paired surrogates at the boundary
283 /// with a supplementary code point,
284 /// like concatenating ill-formed UTF-16 strings effectively would.
285 #[inline]
286 pub fn push_wtf8(&mut self, other: &Wtf8) {
287 match ((&*self).final_lead_surrogate(), other.initial_trail_surrogate()) {
288 // Replace newly paired surrogates by a supplementary code point.
289 (Some(lead), Some(trail)) => {
290 let len_without_lead_surrogate = self.len() - 3;
291 self.bytes.truncate(len_without_lead_surrogate);
292 let other_without_trail_surrogate = &other.as_bytes()[3..];
293 // 4 bytes for the supplementary code point
294 self.bytes.reserve(4 + other_without_trail_surrogate.len());
295 self.push_char(decode_surrogate_pair(lead, trail));
296 self.bytes.extend_from_slice(other_without_trail_surrogate);
297 }
298 _ => {
299 // If we'll be pushing a string containing a surrogate, we may
300 // no longer have UTF-8.
301 if self.is_known_utf8 && other.next_surrogate(0).is_some() {
302 self.is_known_utf8 = false;
303 }
304
305 self.bytes.extend_from_slice(other.as_bytes());
306 }
307 }
308 }
309
310 /// Append a Unicode scalar value at the end of the string.
311 #[inline]
312 pub fn push_char(&mut self, c: char) {
313 // SAFETY: It's always safe to push a char.
314 unsafe { self.push_code_point_unchecked(CodePoint::from_char(c)) }
315 }
316
317 /// Append a code point at the end of the string.
318 ///
319 /// This replaces newly paired surrogates at the boundary
320 /// with a supplementary code point,
321 /// like concatenating ill-formed UTF-16 strings effectively would.
322 #[inline]
323 pub fn push(&mut self, code_point: CodePoint) {
324 if let Some(trail) = code_point.to_trail_surrogate() {
325 if let Some(lead) = (&*self).final_lead_surrogate() {
326 let len_without_lead_surrogate = self.len() - 3;
327 self.bytes.truncate(len_without_lead_surrogate);
328 self.push_char(decode_surrogate_pair(lead, trail));
329 return;
330 }
331
332 // We're pushing a trailing surrogate.
333 self.is_known_utf8 = false;
334 } else if code_point.to_lead_surrogate().is_some() {
335 // We're pushing a leading surrogate.
336 self.is_known_utf8 = false;
337 }
338
339 // No newly paired surrogates at the boundary.
340 unsafe { self.push_code_point_unchecked(code_point) }
341 }
342
343 /// Shortens a string to the specified length.
344 ///
345 /// # Panics
346 ///
347 /// Panics if `new_len` > current length,
348 /// or if `new_len` is not a code point boundary.
349 #[inline]
350 pub fn truncate(&mut self, new_len: usize) {
351 assert!(self.is_code_point_boundary(new_len));
352 self.bytes.truncate(new_len)
353 }
354
355 /// Consumes the WTF-8 string and tries to convert it to a vec of bytes.
356 #[inline]
357 pub fn into_bytes(self) -> Vec<u8> {
358 self.bytes
359 }
360
361 /// Consumes the WTF-8 string and tries to convert it to UTF-8.
362 ///
363 /// This does not copy the data.
364 ///
365 /// If the contents are not well-formed UTF-8
366 /// (that is, if the string contains surrogates),
367 /// the original WTF-8 string is returned instead.
368 pub fn into_string(self) -> Result<String, Wtf8Buf> {
369 if self.is_known_utf8 || self.next_surrogate(0).is_none() {
370 Ok(unsafe { String::from_utf8_unchecked(self.bytes) })
371 } else {
372 Err(self)
373 }
374 }
375
376 /// Consumes the WTF-8 string and converts it lossily to UTF-8.
377 ///
378 /// This does not copy the data (but may overwrite parts of it in place).
379 ///
380 /// Surrogates are replaced with `"\u{FFFD}"` (the replacement character “�”)
381 pub fn into_string_lossy(mut self) -> String {
382 if !self.is_known_utf8 {
383 let mut pos = 0;
384 while let Some((surrogate_pos, _)) = self.next_surrogate(pos) {
385 pos = surrogate_pos + 3;
386 // Surrogates and the replacement character are all 3 bytes, so
387 // they can substituted in-place.
388 self.bytes[surrogate_pos..pos].copy_from_slice("\u{FFFD}".as_bytes());
389 }
390 }
391 unsafe { String::from_utf8_unchecked(self.bytes) }
392 }
393
394 /// Converts this `Wtf8Buf` into a boxed `Wtf8`.
395 #[inline]
396 pub fn into_box(self) -> Box<Wtf8> {
397 // SAFETY: relies on `Wtf8` being `repr(transparent)`.
398 unsafe { mem::transmute(self.bytes.into_boxed_slice()) }
399 }
400
401 /// Converts a `Box<Wtf8>` into a `Wtf8Buf`.
402 pub fn from_box(boxed: Box<Wtf8>) -> Wtf8Buf {
403 let bytes: Box<[u8]> = unsafe { mem::transmute(boxed) };
404 Wtf8Buf { bytes: bytes.into_vec(), is_known_utf8: false }
405 }
406
407 /// Provides plumbing to core `Vec::extend_from_slice`.
408 /// More well behaving alternative to allowing outer types
409 /// full mutable access to the core `Vec`.
410 #[inline]
411 pub unsafe fn extend_from_slice_unchecked(&mut self, other: &[u8]) {
412 self.bytes.extend_from_slice(other);
413 self.is_known_utf8 = false;
414 }
415}
416
417/// Creates a new WTF-8 string from an iterator of code points.
418///
419/// This replaces surrogate code point pairs with supplementary code points,
420/// like concatenating ill-formed UTF-16 strings effectively would.
421impl FromIterator<CodePoint> for Wtf8Buf {
422 fn from_iter<T: IntoIterator<Item = CodePoint>>(iter: T) -> Wtf8Buf {
423 let mut string: Wtf8Buf = Wtf8Buf::new();
424 string.extend(iter);
425 string
426 }
427}
428
429/// Append code points from an iterator to the string.
430///
431/// This replaces surrogate code point pairs with supplementary code points,
432/// like concatenating ill-formed UTF-16 strings effectively would.
433impl Extend<CodePoint> for Wtf8Buf {
434 fn extend<T: IntoIterator<Item = CodePoint>>(&mut self, iter: T) {
435 let iterator = iter.into_iter();
436 let (low: usize, _high) = iterator.size_hint();
437 // Lower bound of one byte per code point (ASCII only)
438 self.bytes.reserve(additional:low);
439 iterator.for_each(move |code_point| self.push(code_point));
440 }
441
442 #[inline]
443 fn extend_one(&mut self, code_point: CodePoint) {
444 self.push(code_point);
445 }
446
447 #[inline]
448 fn extend_reserve(&mut self, additional: usize) {
449 // Lower bound of one byte per code point (ASCII only)
450 self.bytes.reserve(additional);
451 }
452}
453
454/// Creates an owned `Wtf8Buf` from a borrowed `Wtf8`.
455pub(super) fn to_owned(slice: &Wtf8) -> Wtf8Buf {
456 Wtf8Buf { bytes: slice.as_bytes().to_vec(), is_known_utf8: false }
457}
458
459/// Lossily converts the string to UTF-8.
460/// Returns a UTF-8 `&str` slice if the contents are well-formed in UTF-8.
461///
462/// Surrogates are replaced with `"\u{FFFD}"` (the replacement character “�”).
463///
464/// This only copies the data if necessary (if it contains any surrogate).
465pub(super) fn to_string_lossy(slice: &Wtf8) -> Cow<'_, str> {
466 let Some((surrogate_pos, _)) = slice.next_surrogate(0) else {
467 return Cow::Borrowed(unsafe { str::from_utf8_unchecked(slice.as_bytes()) });
468 };
469 let wtf8_bytes = slice.as_bytes();
470 let mut utf8_bytes: Vec = Vec::with_capacity(slice.len());
471 utf8_bytes.extend_from_slice(&wtf8_bytes[..surrogate_pos]);
472 utf8_bytes.extend_from_slice("\u{FFFD}".as_bytes());
473 let mut pos = surrogate_pos + 3;
474 loop {
475 match slice.next_surrogate(pos) {
476 Some((surrogate_pos, _)) => {
477 utf8_bytes.extend_from_slice(&wtf8_bytes[pos..surrogate_pos]);
478 utf8_bytes.extend_from_slice("\u{FFFD}".as_bytes());
479 pos = surrogate_pos + 3;
480 }
481 None => {
482 utf8_bytes.extend_from_slice(&wtf8_bytes[pos..]);
483 return Cow::Owned(unsafe { String::from_utf8_unchecked(utf8_bytes) });
484 }
485 }
486 }
487}
488
489#[inline]
490pub(super) fn clone_into(slice: &Wtf8, buf: &mut Wtf8Buf) {
491 buf.is_known_utf8 = false;
492 slice.as_bytes().clone_into(&mut buf.bytes);
493}
494
495#[cfg(not(test))]
496impl Wtf8 {
497 #[rustc_allow_incoherent_impl]
498 pub fn to_owned(&self) -> Wtf8Buf {
499 to_owned(self)
500 }
501
502 #[rustc_allow_incoherent_impl]
503 pub fn clone_into(&self, buf: &mut Wtf8Buf) {
504 clone_into(self, buf)
505 }
506
507 #[rustc_allow_incoherent_impl]
508 pub fn to_string_lossy(&self) -> Cow<'_, str> {
509 to_string_lossy(self)
510 }
511
512 #[rustc_allow_incoherent_impl]
513 pub fn into_box(&self) -> Box<Wtf8> {
514 let boxed: Box<[u8]> = self.as_bytes().into();
515 unsafe { mem::transmute(boxed) }
516 }
517
518 #[rustc_allow_incoherent_impl]
519 pub fn empty_box() -> Box<Wtf8> {
520 let boxed: Box<[u8]> = Default::default();
521 unsafe { mem::transmute(boxed) }
522 }
523
524 #[cfg(target_has_atomic = "ptr")]
525 #[rustc_allow_incoherent_impl]
526 pub fn into_arc(&self) -> Arc<Wtf8> {
527 let arc: Arc<[u8]> = Arc::from(self.as_bytes());
528 unsafe { Arc::from_raw(Arc::into_raw(arc) as *const Wtf8) }
529 }
530
531 #[rustc_allow_incoherent_impl]
532 pub fn into_rc(&self) -> Rc<Wtf8> {
533 let rc: Rc<[u8]> = Rc::from(self.as_bytes());
534 unsafe { Rc::from_raw(Rc::into_raw(rc) as *const Wtf8) }
535 }
536
537 #[inline]
538 #[rustc_allow_incoherent_impl]
539 pub fn to_ascii_lowercase(&self) -> Wtf8Buf {
540 Wtf8Buf { bytes: self.as_bytes().to_ascii_lowercase(), is_known_utf8: false }
541 }
542
543 #[inline]
544 #[rustc_allow_incoherent_impl]
545 pub fn to_ascii_uppercase(&self) -> Wtf8Buf {
546 Wtf8Buf { bytes: self.as_bytes().to_ascii_uppercase(), is_known_utf8: false }
547 }
548}
549
550#[inline]
551fn decode_surrogate_pair(lead: u16, trail: u16) -> char {
552 let code_point: u32 = 0x10000 + ((((lead - 0xD800) as u32) << 10) | (trail - 0xDC00) as u32);
553 unsafe { char::from_u32_unchecked(code_point) }
554}
555
556impl Hash for Wtf8Buf {
557 #[inline]
558 fn hash<H: Hasher>(&self, state: &mut H) {
559 state.write(&self.bytes);
560 0xfeu8.hash(state)
561 }
562}
563