5 self, BorrowedCursor, BufRead, IoSliceMut, Read, Seek, SeekFrom, SizeHint, DEFAULT_BUF_SIZE,
9 /// The `BufReader<R>` struct adds buffering to any reader.
11 /// It can be excessively inefficient to work directly with a [`Read`] instance.
12 /// For example, every call to [`read`][`TcpStream::read`] on [`TcpStream`]
13 /// results in a system call. A `BufReader<R>` performs large, infrequent reads on
14 /// the underlying [`Read`] and maintains an in-memory buffer of the results.
16 /// `BufReader<R>` can improve the speed of programs that make *small* and
17 /// *repeated* read calls to the same file or network socket. It does not
18 /// help when reading very large amounts at once, or reading just one or a few
19 /// times. It also provides no advantage when reading from a source that is
20 /// already in memory, like a <code>[Vec]\<u8></code>.
22 /// When the `BufReader<R>` is dropped, the contents of its buffer will be
23 /// discarded. Creating multiple instances of a `BufReader<R>` on the same
24 /// stream can cause data loss. Reading from the underlying reader after
25 /// unwrapping the `BufReader<R>` with [`BufReader::into_inner`] can also cause
28 // HACK(#78696): can't use `crate` for associated items
29 /// [`TcpStream::read`]: super::super::super::net::TcpStream::read
30 /// [`TcpStream`]: crate::net::TcpStream
35 /// use std::io::prelude::*;
36 /// use std::io::BufReader;
37 /// use std::fs::File;
39 /// fn main() -> std::io::Result<()> {
40 /// let f = File::open("log.txt")?;
41 /// let mut reader = BufReader::new(f);
43 /// let mut line = String::new();
44 /// let len = reader.read_line(&mut line)?;
45 /// println!("First line is {len} bytes long");
49 #[stable(feature = "rust1", since = "1.0.0")]
50 pub struct BufReader<R> {
55 impl<R: Read> BufReader<R> {
56 /// Creates a new `BufReader<R>` with a default buffer capacity. The default is currently 8 KB,
57 /// but may change in the future.
62 /// use std::io::BufReader;
63 /// use std::fs::File;
65 /// fn main() -> std::io::Result<()> {
66 /// let f = File::open("log.txt")?;
67 /// let reader = BufReader::new(f);
71 #[stable(feature = "rust1", since = "1.0.0")]
72 pub fn new(inner: R) -> BufReader<R> {
73 BufReader::with_capacity(DEFAULT_BUF_SIZE, inner)
76 /// Creates a new `BufReader<R>` with the specified buffer capacity.
80 /// Creating a buffer with ten bytes of capacity:
83 /// use std::io::BufReader;
84 /// use std::fs::File;
86 /// fn main() -> std::io::Result<()> {
87 /// let f = File::open("log.txt")?;
88 /// let reader = BufReader::with_capacity(10, f);
92 #[stable(feature = "rust1", since = "1.0.0")]
93 pub fn with_capacity(capacity: usize, inner: R) -> BufReader<R> {
94 BufReader { inner, buf: Buffer::with_capacity(capacity) }
98 impl<R> BufReader<R> {
99 /// Gets a reference to the underlying reader.
101 /// It is inadvisable to directly read from the underlying reader.
106 /// use std::io::BufReader;
107 /// use std::fs::File;
109 /// fn main() -> std::io::Result<()> {
110 /// let f1 = File::open("log.txt")?;
111 /// let reader = BufReader::new(f1);
113 /// let f2 = reader.get_ref();
117 #[stable(feature = "rust1", since = "1.0.0")]
118 pub fn get_ref(&self) -> &R {
122 /// Gets a mutable reference to the underlying reader.
124 /// It is inadvisable to directly read from the underlying reader.
129 /// use std::io::BufReader;
130 /// use std::fs::File;
132 /// fn main() -> std::io::Result<()> {
133 /// let f1 = File::open("log.txt")?;
134 /// let mut reader = BufReader::new(f1);
136 /// let f2 = reader.get_mut();
140 #[stable(feature = "rust1", since = "1.0.0")]
141 pub fn get_mut(&mut self) -> &mut R {
145 /// Returns a reference to the internally buffered data.
147 /// Unlike [`fill_buf`], this will not attempt to fill the buffer if it is empty.
149 /// [`fill_buf`]: BufRead::fill_buf
154 /// use std::io::{BufReader, BufRead};
155 /// use std::fs::File;
157 /// fn main() -> std::io::Result<()> {
158 /// let f = File::open("log.txt")?;
159 /// let mut reader = BufReader::new(f);
160 /// assert!(reader.buffer().is_empty());
162 /// if reader.fill_buf()?.len() > 0 {
163 /// assert!(!reader.buffer().is_empty());
168 #[stable(feature = "bufreader_buffer", since = "1.37.0")]
169 pub fn buffer(&self) -> &[u8] {
173 /// Returns the number of bytes the internal buffer can hold at once.
178 /// use std::io::{BufReader, BufRead};
179 /// use std::fs::File;
181 /// fn main() -> std::io::Result<()> {
182 /// let f = File::open("log.txt")?;
183 /// let mut reader = BufReader::new(f);
185 /// let capacity = reader.capacity();
186 /// let buffer = reader.fill_buf()?;
187 /// assert!(buffer.len() <= capacity);
191 #[stable(feature = "buffered_io_capacity", since = "1.46.0")]
192 pub fn capacity(&self) -> usize {
196 /// Unwraps this `BufReader<R>`, returning the underlying reader.
198 /// Note that any leftover data in the internal buffer is lost. Therefore,
199 /// a following read from the underlying reader may lead to data loss.
204 /// use std::io::BufReader;
205 /// use std::fs::File;
207 /// fn main() -> std::io::Result<()> {
208 /// let f1 = File::open("log.txt")?;
209 /// let reader = BufReader::new(f1);
211 /// let f2 = reader.into_inner();
215 #[stable(feature = "rust1", since = "1.0.0")]
216 pub fn into_inner(self) -> R {
220 /// Invalidates all data in the internal buffer.
222 fn discard_buffer(&mut self) {
223 self.buf.discard_buffer()
227 // This is only used by a test which asserts that the initialization-tracking is correct.
229 impl<R> BufReader<R> {
230 pub fn initialized(&self) -> usize {
231 self.buf.initialized()
235 impl<R: Seek> BufReader<R> {
236 /// Seeks relative to the current position. If the new position lies within the buffer,
237 /// the buffer will not be flushed, allowing for more efficient seeks.
238 /// This method does not return the location of the underlying reader, so the caller
239 /// must track this information themselves if it is required.
240 #[stable(feature = "bufreader_seek_relative", since = "1.53.0")]
241 pub fn seek_relative(&mut self, offset: i64) -> io::Result<()> {
242 let pos = self.buf.pos() as u64;
244 if let Some(_) = pos.checked_sub((-offset) as u64) {
245 self.buf.unconsume((-offset) as usize);
248 } else if let Some(new_pos) = pos.checked_add(offset as u64) {
249 if new_pos <= self.buf.filled() as u64 {
250 self.buf.consume(offset as usize);
255 self.seek(SeekFrom::Current(offset)).map(drop)
259 #[stable(feature = "rust1", since = "1.0.0")]
260 impl<R: Read> Read for BufReader<R> {
261 fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
262 // If we don't have any buffered data and we're doing a massive read
263 // (larger than our internal buffer), bypass our internal buffer
265 if self.buf.pos() == self.buf.filled() && buf.len() >= self.capacity() {
266 self.discard_buffer();
267 return self.inner.read(buf);
270 let mut rem = self.fill_buf()?;
277 fn read_buf(&mut self, mut cursor: BorrowedCursor<'_>) -> io::Result<()> {
278 // If we don't have any buffered data and we're doing a massive read
279 // (larger than our internal buffer), bypass our internal buffer
281 if self.buf.pos() == self.buf.filled() && cursor.capacity() >= self.capacity() {
282 self.discard_buffer();
283 return self.inner.read_buf(cursor);
286 let prev = cursor.written();
288 let mut rem = self.fill_buf()?;
289 rem.read_buf(cursor.reborrow())?;
291 self.consume(cursor.written() - prev); //slice impl of read_buf known to never unfill buf
296 // Small read_exacts from a BufReader are extremely common when used with a deserializer.
297 // The default implementation calls read in a loop, which results in surprisingly poor code
298 // generation for the common path where the buffer has enough bytes to fill the passed-in
300 fn read_exact(&mut self, buf: &mut [u8]) -> io::Result<()> {
301 if self.buf.consume_with(buf.len(), |claimed| buf.copy_from_slice(claimed)) {
305 crate::io::default_read_exact(self, buf)
308 fn read_vectored(&mut self, bufs: &mut [IoSliceMut<'_>]) -> io::Result<usize> {
309 let total_len = bufs.iter().map(|b| b.len()).sum::<usize>();
310 if self.buf.pos() == self.buf.filled() && total_len >= self.capacity() {
311 self.discard_buffer();
312 return self.inner.read_vectored(bufs);
315 let mut rem = self.fill_buf()?;
316 rem.read_vectored(bufs)?
322 fn is_read_vectored(&self) -> bool {
323 self.inner.is_read_vectored()
326 // The inner reader might have an optimized `read_to_end`. Drain our buffer and then
327 // delegate to the inner implementation.
328 fn read_to_end(&mut self, buf: &mut Vec<u8>) -> io::Result<usize> {
329 let inner_buf = self.buffer();
330 buf.extend_from_slice(inner_buf);
331 let nread = inner_buf.len();
332 self.discard_buffer();
333 Ok(nread + self.inner.read_to_end(buf)?)
336 // The inner reader might have an optimized `read_to_end`. Drain our buffer and then
337 // delegate to the inner implementation.
338 fn read_to_string(&mut self, buf: &mut String) -> io::Result<usize> {
339 // In the general `else` case below we must read bytes into a side buffer, check
340 // that they are valid UTF-8, and then append them to `buf`. This requires a
341 // potentially large memcpy.
343 // If `buf` is empty--the most common case--we can leverage `append_to_string`
344 // to read directly into `buf`'s internal byte buffer, saving an allocation and
347 // `append_to_string`'s safety relies on the buffer only being appended to since
348 // it only checks the UTF-8 validity of new data. If there were existing content in
349 // `buf` then an untrustworthy reader (i.e. `self.inner`) could not only append
350 // bytes but also modify existing bytes and render them invalid. On the other hand,
351 // if `buf` is empty then by definition any writes must be appends and
352 // `append_to_string` will validate all of the new bytes.
353 unsafe { crate::io::append_to_string(buf, |b| self.read_to_end(b)) }
355 // We cannot append our byte buffer directly onto the `buf` String as there could
356 // be an incomplete UTF-8 sequence that has only been partially read. We must read
357 // everything into a side buffer first and then call `from_utf8` on the complete
359 let mut bytes = Vec::new();
360 self.read_to_end(&mut bytes)?;
361 let string = crate::str::from_utf8(&bytes).map_err(|_| {
363 io::ErrorKind::InvalidData,
364 "stream did not contain valid UTF-8",
373 #[stable(feature = "rust1", since = "1.0.0")]
374 impl<R: Read> BufRead for BufReader<R> {
375 fn fill_buf(&mut self) -> io::Result<&[u8]> {
376 self.buf.fill_buf(&mut self.inner)
379 fn consume(&mut self, amt: usize) {
380 self.buf.consume(amt)
384 #[stable(feature = "rust1", since = "1.0.0")]
385 impl<R> fmt::Debug for BufReader<R>
389 fn fmt(&self, fmt: &mut fmt::Formatter<'_>) -> fmt::Result {
390 fmt.debug_struct("BufReader")
391 .field("reader", &self.inner)
394 &format_args!("{}/{}", self.buf.filled() - self.buf.pos(), self.capacity()),
400 #[stable(feature = "rust1", since = "1.0.0")]
401 impl<R: Seek> Seek for BufReader<R> {
402 /// Seek to an offset, in bytes, in the underlying reader.
404 /// The position used for seeking with <code>[SeekFrom::Current]\(_)</code> is the
405 /// position the underlying reader would be at if the `BufReader<R>` had no
408 /// Seeking always discards the internal buffer, even if the seek position
409 /// would otherwise fall within it. This guarantees that calling
410 /// [`BufReader::into_inner()`] immediately after a seek yields the underlying reader
411 /// at the same position.
413 /// To seek without discarding the internal buffer, use [`BufReader::seek_relative`].
415 /// See [`std::io::Seek`] for more details.
417 /// Note: In the edge case where you're seeking with <code>[SeekFrom::Current]\(n)</code>
418 /// where `n` minus the internal buffer length overflows an `i64`, two
419 /// seeks will be performed instead of one. If the second seek returns
420 /// [`Err`], the underlying reader will be left at the same position it would
421 /// have if you called `seek` with <code>[SeekFrom::Current]\(0)</code>.
423 /// [`std::io::Seek`]: Seek
424 fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
426 if let SeekFrom::Current(n) = pos {
427 let remainder = (self.buf.filled() - self.buf.pos()) as i64;
428 // it should be safe to assume that remainder fits within an i64 as the alternative
429 // means we managed to allocate 8 exbibytes and that's absurd.
430 // But it's not out of the realm of possibility for some weird underlying reader to
431 // support seeking by i64::MIN so we need to handle underflow when subtracting
433 if let Some(offset) = n.checked_sub(remainder) {
434 result = self.inner.seek(SeekFrom::Current(offset))?;
436 // seek backwards by our remainder, and then by the offset
437 self.inner.seek(SeekFrom::Current(-remainder))?;
438 self.discard_buffer();
439 result = self.inner.seek(SeekFrom::Current(n))?;
442 // Seeking with Start/End doesn't care about our buffer length.
443 result = self.inner.seek(pos)?;
445 self.discard_buffer();
449 /// Returns the current seek position from the start of the stream.
451 /// The value returned is equivalent to `self.seek(SeekFrom::Current(0))`
452 /// but does not flush the internal buffer. Due to this optimization the
453 /// function does not guarantee that calling `.into_inner()` immediately
454 /// afterwards will yield the underlying reader at the same position. Use
455 /// [`BufReader::seek`] instead if you require that guarantee.
459 /// This function will panic if the position of the inner reader is smaller
460 /// than the amount of buffered data. That can happen if the inner reader
461 /// has an incorrect implementation of [`Seek::stream_position`], or if the
462 /// position has gone out of sync due to calling [`Seek::seek`] directly on
463 /// the underlying reader.
469 /// io::{self, BufRead, BufReader, Seek},
473 /// fn main() -> io::Result<()> {
474 /// let mut f = BufReader::new(File::open("foo.txt")?);
476 /// let before = f.stream_position()?;
477 /// f.read_line(&mut String::new())?;
478 /// let after = f.stream_position()?;
480 /// println!("The first line was {} bytes long", after - before);
484 fn stream_position(&mut self) -> io::Result<u64> {
485 let remainder = (self.buf.filled() - self.buf.pos()) as u64;
486 self.inner.stream_position().map(|pos| {
487 pos.checked_sub(remainder).expect(
488 "overflow when subtracting remaining buffer size from inner stream position",
494 impl<T> SizeHint for BufReader<T> {
496 fn lower_bound(&self) -> usize {
497 SizeHint::lower_bound(self.get_ref()) + self.buffer().len()
501 fn upper_bound(&self) -> Option<usize> {
502 SizeHint::upper_bound(self.get_ref()).and_then(|up| self.buffer().len().checked_add(up))