blob: 97c4b879793b71cd4f7ac958788586bf253f7930 [file] [log] [blame]
//! Buffering wrappers for I/O traits
#[cfg(test)]
mod tests;
use crate::io::prelude::*;
use crate::cmp;
use crate::error;
use crate::fmt;
use crate::io::{
self, Error, ErrorKind, Initializer, IoSlice, IoSliceMut, SeekFrom, DEFAULT_BUF_SIZE,
};
use crate::memchr;
/// The `BufReader<R>` struct adds buffering to any reader.
///
/// It can be excessively inefficient to work directly with a [`Read`] instance.
/// For example, every call to [`read`][`TcpStream::read`] on [`TcpStream`]
/// results in a system call. A `BufReader<R>` performs large, infrequent reads on
/// the underlying [`Read`] and maintains an in-memory buffer of the results.
///
/// `BufReader<R>` can improve the speed of programs that make *small* and
/// *repeated* read calls to the same file or network socket. It does not
/// help when reading very large amounts at once, or reading just one or a few
/// times. It also provides no advantage when reading from a source that is
/// already in memory, like a [`Vec`]`<u8>`.
///
/// When the `BufReader<R>` is dropped, the contents of its buffer will be
/// discarded. Creating multiple instances of a `BufReader<R>` on the same
/// stream can cause data loss. Reading from the underlying reader after
/// unwrapping the `BufReader<R>` with [`BufReader::into_inner`] can also cause
/// data loss.
///
/// [`TcpStream::read`]: Read::read
/// [`TcpStream`]: crate::net::TcpStream
///
/// # Examples
///
/// ```no_run
/// use std::io::prelude::*;
/// use std::io::BufReader;
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f = File::open("log.txt")?;
/// let mut reader = BufReader::new(f);
///
/// let mut line = String::new();
/// let len = reader.read_line(&mut line)?;
/// println!("First line is {} bytes long", len);
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub struct BufReader<R> {
inner: R,
buf: Box<[u8]>,
pos: usize,
cap: usize,
}
impl<R: Read> BufReader<R> {
/// Creates a new `BufReader<R>` with a default buffer capacity. The default is currently 8 KB,
/// but may change in the future.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufReader;
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f = File::open("log.txt")?;
/// let reader = BufReader::new(f);
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn new(inner: R) -> BufReader<R> {
BufReader::with_capacity(DEFAULT_BUF_SIZE, inner)
}
/// Creates a new `BufReader<R>` with the specified buffer capacity.
///
/// # Examples
///
/// Creating a buffer with ten bytes of capacity:
///
/// ```no_run
/// use std::io::BufReader;
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f = File::open("log.txt")?;
/// let reader = BufReader::with_capacity(10, f);
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn with_capacity(capacity: usize, inner: R) -> BufReader<R> {
unsafe {
let mut buffer = Vec::with_capacity(capacity);
buffer.set_len(capacity);
inner.initializer().initialize(&mut buffer);
BufReader { inner, buf: buffer.into_boxed_slice(), pos: 0, cap: 0 }
}
}
}
impl<R> BufReader<R> {
/// Gets a reference to the underlying reader.
///
/// It is inadvisable to directly read from the underlying reader.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufReader;
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f1 = File::open("log.txt")?;
/// let reader = BufReader::new(f1);
///
/// let f2 = reader.get_ref();
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn get_ref(&self) -> &R {
&self.inner
}
/// Gets a mutable reference to the underlying reader.
///
/// It is inadvisable to directly read from the underlying reader.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufReader;
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f1 = File::open("log.txt")?;
/// let mut reader = BufReader::new(f1);
///
/// let f2 = reader.get_mut();
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn get_mut(&mut self) -> &mut R {
&mut self.inner
}
/// Returns a reference to the internally buffered data.
///
/// Unlike [`fill_buf`], this will not attempt to fill the buffer if it is empty.
///
/// [`fill_buf`]: BufRead::fill_buf
///
/// # Examples
///
/// ```no_run
/// use std::io::{BufReader, BufRead};
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f = File::open("log.txt")?;
/// let mut reader = BufReader::new(f);
/// assert!(reader.buffer().is_empty());
///
/// if reader.fill_buf()?.len() > 0 {
/// assert!(!reader.buffer().is_empty());
/// }
/// Ok(())
/// }
/// ```
#[stable(feature = "bufreader_buffer", since = "1.37.0")]
pub fn buffer(&self) -> &[u8] {
&self.buf[self.pos..self.cap]
}
/// Returns the number of bytes the internal buffer can hold at once.
///
/// # Examples
///
/// ```no_run
/// use std::io::{BufReader, BufRead};
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f = File::open("log.txt")?;
/// let mut reader = BufReader::new(f);
///
/// let capacity = reader.capacity();
/// let buffer = reader.fill_buf()?;
/// assert!(buffer.len() <= capacity);
/// Ok(())
/// }
/// ```
#[stable(feature = "buffered_io_capacity", since = "1.46.0")]
pub fn capacity(&self) -> usize {
self.buf.len()
}
/// Unwraps this `BufReader<R>`, returning the underlying reader.
///
/// Note that any leftover data in the internal buffer is lost. Therefore,
/// a following read from the underlying reader may lead to data loss.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufReader;
/// use std::fs::File;
///
/// fn main() -> std::io::Result<()> {
/// let f1 = File::open("log.txt")?;
/// let reader = BufReader::new(f1);
///
/// let f2 = reader.into_inner();
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn into_inner(self) -> R {
self.inner
}
/// Invalidates all data in the internal buffer.
#[inline]
fn discard_buffer(&mut self) {
self.pos = 0;
self.cap = 0;
}
}
impl<R: Seek> BufReader<R> {
/// Seeks relative to the current position. If the new position lies within the buffer,
/// the buffer will not be flushed, allowing for more efficient seeks.
/// This method does not return the location of the underlying reader, so the caller
/// must track this information themselves if it is required.
#[unstable(feature = "bufreader_seek_relative", issue = "31100")]
pub fn seek_relative(&mut self, offset: i64) -> io::Result<()> {
let pos = self.pos as u64;
if offset < 0 {
if let Some(new_pos) = pos.checked_sub((-offset) as u64) {
self.pos = new_pos as usize;
return Ok(());
}
} else {
if let Some(new_pos) = pos.checked_add(offset as u64) {
if new_pos <= self.cap as u64 {
self.pos = new_pos as usize;
return Ok(());
}
}
}
self.seek(SeekFrom::Current(offset)).map(drop)
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<R: Read> Read for BufReader<R> {
fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
// If we don't have any buffered data and we're doing a massive read
// (larger than our internal buffer), bypass our internal buffer
// entirely.
if self.pos == self.cap && buf.len() >= self.buf.len() {
self.discard_buffer();
return self.inner.read(buf);
}
let nread = {
let mut rem = self.fill_buf()?;
rem.read(buf)?
};
self.consume(nread);
Ok(nread)
}
fn read_vectored(&mut self, bufs: &mut [IoSliceMut<'_>]) -> io::Result<usize> {
let total_len = bufs.iter().map(|b| b.len()).sum::<usize>();
if self.pos == self.cap && total_len >= self.buf.len() {
self.discard_buffer();
return self.inner.read_vectored(bufs);
}
let nread = {
let mut rem = self.fill_buf()?;
rem.read_vectored(bufs)?
};
self.consume(nread);
Ok(nread)
}
fn is_read_vectored(&self) -> bool {
self.inner.is_read_vectored()
}
// we can't skip unconditionally because of the large buffer case in read.
unsafe fn initializer(&self) -> Initializer {
self.inner.initializer()
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<R: Read> BufRead for BufReader<R> {
fn fill_buf(&mut self) -> io::Result<&[u8]> {
// If we've reached the end of our internal buffer then we need to fetch
// some more data from the underlying reader.
// Branch using `>=` instead of the more correct `==`
// to tell the compiler that the pos..cap slice is always valid.
if self.pos >= self.cap {
debug_assert!(self.pos == self.cap);
self.cap = self.inner.read(&mut self.buf)?;
self.pos = 0;
}
Ok(&self.buf[self.pos..self.cap])
}
fn consume(&mut self, amt: usize) {
self.pos = cmp::min(self.pos + amt, self.cap);
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<R> fmt::Debug for BufReader<R>
where
R: fmt::Debug,
{
fn fmt(&self, fmt: &mut fmt::Formatter<'_>) -> fmt::Result {
fmt.debug_struct("BufReader")
.field("reader", &self.inner)
.field("buffer", &format_args!("{}/{}", self.cap - self.pos, self.buf.len()))
.finish()
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<R: Seek> Seek for BufReader<R> {
/// Seek to an offset, in bytes, in the underlying reader.
///
/// The position used for seeking with [`SeekFrom::Current`]`(_)` is the
/// position the underlying reader would be at if the `BufReader<R>` had no
/// internal buffer.
///
/// Seeking always discards the internal buffer, even if the seek position
/// would otherwise fall within it. This guarantees that calling
/// [`BufReader::into_inner()`] immediately after a seek yields the underlying reader
/// at the same position.
///
/// To seek without discarding the internal buffer, use [`BufReader::seek_relative`].
///
/// See [`std::io::Seek`] for more details.
///
/// Note: In the edge case where you're seeking with [`SeekFrom::Current`]`(n)`
/// where `n` minus the internal buffer length overflows an `i64`, two
/// seeks will be performed instead of one. If the second seek returns
/// [`Err`], the underlying reader will be left at the same position it would
/// have if you called `seek` with [`SeekFrom::Current`]`(0)`.
///
/// [`std::io::Seek`]: Seek
fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
let result: u64;
if let SeekFrom::Current(n) = pos {
let remainder = (self.cap - self.pos) as i64;
// it should be safe to assume that remainder fits within an i64 as the alternative
// means we managed to allocate 8 exbibytes and that's absurd.
// But it's not out of the realm of possibility for some weird underlying reader to
// support seeking by i64::MIN so we need to handle underflow when subtracting
// remainder.
if let Some(offset) = n.checked_sub(remainder) {
result = self.inner.seek(SeekFrom::Current(offset))?;
} else {
// seek backwards by our remainder, and then by the offset
self.inner.seek(SeekFrom::Current(-remainder))?;
self.discard_buffer();
result = self.inner.seek(SeekFrom::Current(n))?;
}
} else {
// Seeking with Start/End doesn't care about our buffer length.
result = self.inner.seek(pos)?;
}
self.discard_buffer();
Ok(result)
}
/// Returns the current seek position from the start of the stream.
///
/// The value returned is equivalent to `self.seek(SeekFrom::Current(0))`
/// but does not flush the internal buffer. Due to this optimization the
/// function does not guarantee that calling `.into_inner()` immediately
/// afterwards will yield the underlying reader at the same position. Use
/// [`BufReader::seek`] instead if you require that guarantee.
///
/// # Panics
///
/// This function will panic if the position of the inner reader is smaller
/// than the amount of buffered data. That can happen if the inner reader
/// has an incorrect implementation of [`Seek::stream_position`], or if the
/// position has gone out of sync due to calling [`Seek::seek`] directly on
/// the underlying reader.
///
/// # Example
///
/// ```no_run
/// #![feature(seek_convenience)]
/// use std::{
/// io::{self, BufRead, BufReader, Seek},
/// fs::File,
/// };
///
/// fn main() -> io::Result<()> {
/// let mut f = BufReader::new(File::open("foo.txt")?);
///
/// let before = f.stream_position()?;
/// f.read_line(&mut String::new())?;
/// let after = f.stream_position()?;
///
/// println!("The first line was {} bytes long", after - before);
/// Ok(())
/// }
/// ```
fn stream_position(&mut self) -> io::Result<u64> {
let remainder = (self.cap - self.pos) as u64;
self.inner.stream_position().map(|pos| {
pos.checked_sub(remainder).expect(
"overflow when subtracting remaining buffer size from inner stream position",
)
})
}
}
/// Wraps a writer and buffers its output.
///
/// It can be excessively inefficient to work directly with something that
/// implements [`Write`]. For example, every call to
/// [`write`][`TcpStream::write`] on [`TcpStream`] results in a system call. A
/// `BufWriter<W>` keeps an in-memory buffer of data and writes it to an underlying
/// writer in large, infrequent batches.
///
/// `BufWriter<W>` can improve the speed of programs that make *small* and
/// *repeated* write calls to the same file or network socket. It does not
/// help when writing very large amounts at once, or writing just one or a few
/// times. It also provides no advantage when writing to a destination that is
/// in memory, like a [`Vec`]<u8>`.
///
/// It is critical to call [`flush`] before `BufWriter<W>` is dropped. Though
/// dropping will attempt to flush the contents of the buffer, any errors
/// that happen in the process of dropping will be ignored. Calling [`flush`]
/// ensures that the buffer is empty and thus dropping will not even attempt
/// file operations.
///
/// # Examples
///
/// Let's write the numbers one through ten to a [`TcpStream`]:
///
/// ```no_run
/// use std::io::prelude::*;
/// use std::net::TcpStream;
///
/// let mut stream = TcpStream::connect("127.0.0.1:34254").unwrap();
///
/// for i in 0..10 {
/// stream.write(&[i+1]).unwrap();
/// }
/// ```
///
/// Because we're not buffering, we write each one in turn, incurring the
/// overhead of a system call per byte written. We can fix this with a
/// `BufWriter<W>`:
///
/// ```no_run
/// use std::io::prelude::*;
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut stream = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// for i in 0..10 {
/// stream.write(&[i+1]).unwrap();
/// }
/// stream.flush().unwrap();
/// ```
///
/// By wrapping the stream with a `BufWriter<W>`, these ten writes are all grouped
/// together by the buffer and will all be written out in one system call when
/// the `stream` is flushed.
///
/// [`TcpStream::write`]: Write::write
/// [`TcpStream`]: crate::net::TcpStream
/// [`flush`]: Write::flush
#[stable(feature = "rust1", since = "1.0.0")]
pub struct BufWriter<W: Write> {
inner: Option<W>,
buf: Vec<u8>,
// #30888: If the inner writer panics in a call to write, we don't want to
// write the buffered data a second time in BufWriter's destructor. This
// flag tells the Drop impl if it should skip the flush.
panicked: bool,
}
/// An error returned by [`BufWriter::into_inner`] which combines an error that
/// happened while writing out the buffer, and the buffered writer object
/// which may be used to recover from the condition.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut stream = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // do stuff with the stream
///
/// // we want to get our `TcpStream` back, so let's try:
///
/// let stream = match stream.into_inner() {
/// Ok(s) => s,
/// Err(e) => {
/// // Here, e is an IntoInnerError
/// panic!("An error occurred");
/// }
/// };
/// ```
#[derive(Debug)]
#[stable(feature = "rust1", since = "1.0.0")]
pub struct IntoInnerError<W>(W, Error);
impl<W: Write> BufWriter<W> {
/// Creates a new `BufWriter<W>` with a default buffer capacity. The default is currently 8 KB,
/// but may change in the future.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut buffer = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn new(inner: W) -> BufWriter<W> {
BufWriter::with_capacity(DEFAULT_BUF_SIZE, inner)
}
/// Creates a new `BufWriter<W>` with the specified buffer capacity.
///
/// # Examples
///
/// Creating a buffer with a buffer of a hundred bytes.
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let stream = TcpStream::connect("127.0.0.1:34254").unwrap();
/// let mut buffer = BufWriter::with_capacity(100, stream);
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn with_capacity(capacity: usize, inner: W) -> BufWriter<W> {
BufWriter { inner: Some(inner), buf: Vec::with_capacity(capacity), panicked: false }
}
/// Send data in our local buffer into the inner writer, looping as
/// necessary until either it's all been sent or an error occurs.
///
/// Because all the data in the buffer has been reported to our owner as
/// "successfully written" (by returning nonzero success values from
/// `write`), any 0-length writes from `inner` must be reported as i/o
/// errors from this method.
fn flush_buf(&mut self) -> io::Result<()> {
/// Helper struct to ensure the buffer is updated after all the writes
/// are complete. It tracks the number of written bytes and drains them
/// all from the front of the buffer when dropped.
struct BufGuard<'a> {
buffer: &'a mut Vec<u8>,
written: usize,
}
impl<'a> BufGuard<'a> {
fn new(buffer: &'a mut Vec<u8>) -> Self {
Self { buffer, written: 0 }
}
/// The unwritten part of the buffer
fn remaining(&self) -> &[u8] {
&self.buffer[self.written..]
}
/// Flag some bytes as removed from the front of the buffer
fn consume(&mut self, amt: usize) {
self.written += amt;
}
/// true if all of the bytes have been written
fn done(&self) -> bool {
self.written >= self.buffer.len()
}
}
impl Drop for BufGuard<'_> {
fn drop(&mut self) {
if self.written > 0 {
self.buffer.drain(..self.written);
}
}
}
let mut guard = BufGuard::new(&mut self.buf);
let inner = self.inner.as_mut().unwrap();
while !guard.done() {
self.panicked = true;
let r = inner.write(guard.remaining());
self.panicked = false;
match r {
Ok(0) => {
return Err(Error::new(
ErrorKind::WriteZero,
"failed to write the buffered data",
));
}
Ok(n) => guard.consume(n),
Err(ref e) if e.kind() == io::ErrorKind::Interrupted => {}
Err(e) => return Err(e),
}
}
Ok(())
}
/// Buffer some data without flushing it, regardless of the size of the
/// data. Writes as much as possible without exceeding capacity. Returns
/// the number of bytes written.
fn write_to_buf(&mut self, buf: &[u8]) -> usize {
let available = self.buf.capacity() - self.buf.len();
let amt_to_buffer = available.min(buf.len());
self.buf.extend_from_slice(&buf[..amt_to_buffer]);
amt_to_buffer
}
/// Gets a reference to the underlying writer.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut buffer = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // we can use reference just like buffer
/// let reference = buffer.get_ref();
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn get_ref(&self) -> &W {
self.inner.as_ref().unwrap()
}
/// Gets a mutable reference to the underlying writer.
///
/// It is inadvisable to directly write to the underlying writer.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut buffer = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // we can use reference just like buffer
/// let reference = buffer.get_mut();
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn get_mut(&mut self) -> &mut W {
self.inner.as_mut().unwrap()
}
/// Returns a reference to the internally buffered data.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let buf_writer = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // See how many bytes are currently buffered
/// let bytes_buffered = buf_writer.buffer().len();
/// ```
#[stable(feature = "bufreader_buffer", since = "1.37.0")]
pub fn buffer(&self) -> &[u8] {
&self.buf
}
/// Returns the number of bytes the internal buffer can hold without flushing.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let buf_writer = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // Check the capacity of the inner buffer
/// let capacity = buf_writer.capacity();
/// // Calculate how many bytes can be written without flushing
/// let without_flush = capacity - buf_writer.buffer().len();
/// ```
#[stable(feature = "buffered_io_capacity", since = "1.46.0")]
pub fn capacity(&self) -> usize {
self.buf.capacity()
}
/// Unwraps this `BufWriter<W>`, returning the underlying writer.
///
/// The buffer is written out before returning the writer.
///
/// # Errors
///
/// An [`Err`] will be returned if an error occurs while flushing the buffer.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut buffer = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // unwrap the TcpStream and flush the buffer
/// let stream = buffer.into_inner().unwrap();
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn into_inner(mut self) -> Result<W, IntoInnerError<BufWriter<W>>> {
match self.flush_buf() {
Err(e) => Err(IntoInnerError(self, e)),
Ok(()) => Ok(self.inner.take().unwrap()),
}
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W: Write> Write for BufWriter<W> {
fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
if self.buf.len() + buf.len() > self.buf.capacity() {
self.flush_buf()?;
}
// FIXME: Why no len > capacity? Why not buffer len == capacity? #72919
if buf.len() >= self.buf.capacity() {
self.panicked = true;
let r = self.get_mut().write(buf);
self.panicked = false;
r
} else {
self.buf.extend_from_slice(buf);
Ok(buf.len())
}
}
fn write_all(&mut self, buf: &[u8]) -> io::Result<()> {
// Normally, `write_all` just calls `write` in a loop. We can do better
// by calling `self.get_mut().write_all()` directly, which avoids
// round trips through the buffer in the event of a series of partial
// writes in some circumstances.
if self.buf.len() + buf.len() > self.buf.capacity() {
self.flush_buf()?;
}
// FIXME: Why no len > capacity? Why not buffer len == capacity? #72919
if buf.len() >= self.buf.capacity() {
self.panicked = true;
let r = self.get_mut().write_all(buf);
self.panicked = false;
r
} else {
self.buf.extend_from_slice(buf);
Ok(())
}
}
fn write_vectored(&mut self, bufs: &[IoSlice<'_>]) -> io::Result<usize> {
let total_len = bufs.iter().map(|b| b.len()).sum::<usize>();
if self.buf.len() + total_len > self.buf.capacity() {
self.flush_buf()?;
}
// FIXME: Why no len > capacity? Why not buffer len == capacity? #72919
if total_len >= self.buf.capacity() {
self.panicked = true;
let r = self.get_mut().write_vectored(bufs);
self.panicked = false;
r
} else {
bufs.iter().for_each(|b| self.buf.extend_from_slice(b));
Ok(total_len)
}
}
fn is_write_vectored(&self) -> bool {
self.get_ref().is_write_vectored()
}
fn flush(&mut self) -> io::Result<()> {
self.flush_buf().and_then(|()| self.get_mut().flush())
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W: Write> fmt::Debug for BufWriter<W>
where
W: fmt::Debug,
{
fn fmt(&self, fmt: &mut fmt::Formatter<'_>) -> fmt::Result {
fmt.debug_struct("BufWriter")
.field("writer", &self.inner.as_ref().unwrap())
.field("buffer", &format_args!("{}/{}", self.buf.len(), self.buf.capacity()))
.finish()
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W: Write + Seek> Seek for BufWriter<W> {
/// Seek to the offset, in bytes, in the underlying writer.
///
/// Seeking always writes out the internal buffer before seeking.
fn seek(&mut self, pos: SeekFrom) -> io::Result<u64> {
self.flush_buf()?;
self.get_mut().seek(pos)
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W: Write> Drop for BufWriter<W> {
fn drop(&mut self) {
if self.inner.is_some() && !self.panicked {
// dtors should not panic, so we ignore a failed flush
let _r = self.flush_buf();
}
}
}
impl<W> IntoInnerError<W> {
/// Returns the error which caused the call to [`BufWriter::into_inner()`]
/// to fail.
///
/// This error was returned when attempting to write the internal buffer.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut stream = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // do stuff with the stream
///
/// // we want to get our `TcpStream` back, so let's try:
///
/// let stream = match stream.into_inner() {
/// Ok(s) => s,
/// Err(e) => {
/// // Here, e is an IntoInnerError, let's log the inner error.
/// //
/// // We'll just 'log' to stdout for this example.
/// println!("{}", e.error());
///
/// panic!("An unexpected error occurred.");
/// }
/// };
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn error(&self) -> &Error {
&self.1
}
/// Returns the buffered writer instance which generated the error.
///
/// The returned object can be used for error recovery, such as
/// re-inspecting the buffer.
///
/// # Examples
///
/// ```no_run
/// use std::io::BufWriter;
/// use std::net::TcpStream;
///
/// let mut stream = BufWriter::new(TcpStream::connect("127.0.0.1:34254").unwrap());
///
/// // do stuff with the stream
///
/// // we want to get our `TcpStream` back, so let's try:
///
/// let stream = match stream.into_inner() {
/// Ok(s) => s,
/// Err(e) => {
/// // Here, e is an IntoInnerError, let's re-examine the buffer:
/// let buffer = e.into_inner();
///
/// // do stuff to try to recover
///
/// // afterwards, let's just return the stream
/// buffer.into_inner().unwrap()
/// }
/// };
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn into_inner(self) -> W {
self.0
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W> From<IntoInnerError<W>> for Error {
fn from(iie: IntoInnerError<W>) -> Error {
iie.1
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W: Send + fmt::Debug> error::Error for IntoInnerError<W> {
#[allow(deprecated, deprecated_in_future)]
fn description(&self) -> &str {
error::Error::description(self.error())
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W> fmt::Display for IntoInnerError<W> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
self.error().fmt(f)
}
}
/// Private helper struct for implementing the line-buffered writing logic.
/// This shim temporarily wraps a BufWriter, and uses its internals to
/// implement a line-buffered writer (specifically by using the internal
/// methods like write_to_buf and flush_buf). In this way, a more
/// efficient abstraction can be created than one that only had access to
/// `write` and `flush`, without needlessly duplicating a lot of the
/// implementation details of BufWriter. This also allows existing
/// `BufWriters` to be temporarily given line-buffering logic; this is what
/// enables Stdout to be alternately in line-buffered or block-buffered mode.
#[derive(Debug)]
pub(super) struct LineWriterShim<'a, W: Write> {
buffer: &'a mut BufWriter<W>,
}
impl<'a, W: Write> LineWriterShim<'a, W> {
pub fn new(buffer: &'a mut BufWriter<W>) -> Self {
Self { buffer }
}
/// Get a mutable reference to the inner writer (that is, the writer
/// wrapped by the BufWriter). Be careful with this writer, as writes to
/// it will bypass the buffer.
fn inner_mut(&mut self) -> &mut W {
self.buffer.get_mut()
}
/// Get the content currently buffered in self.buffer
fn buffered(&self) -> &[u8] {
self.buffer.buffer()
}
/// Flush the buffer iff the last byte is a newline (indicating that an
/// earlier write only succeeded partially, and we want to retry flushing
/// the buffered line before continuing with a subsequent write)
fn flush_if_completed_line(&mut self) -> io::Result<()> {
match self.buffered().last().copied() {
Some(b'\n') => self.buffer.flush_buf(),
_ => Ok(()),
}
}
}
impl<'a, W: Write> Write for LineWriterShim<'a, W> {
/// Write some data into this BufReader with line buffering. This means
/// that, if any newlines are present in the data, the data up to the last
/// newline is sent directly to the underlying writer, and data after it
/// is buffered. Returns the number of bytes written.
///
/// This function operates on a "best effort basis"; in keeping with the
/// convention of `Write::write`, it makes at most one attempt to write
/// new data to the underlying writer. If that write only reports a partial
/// success, the remaining data will be buffered.
///
/// Because this function attempts to send completed lines to the underlying
/// writer, it will also flush the existing buffer if it ends with a
/// newline, even if the incoming data does not contain any newlines.
fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
let newline_idx = match memchr::memrchr(b'\n', buf) {
// If there are no new newlines (that is, if this write is less than
// one line), just do a regular buffered write (which may flush if
// we exceed the inner buffer's size)
None => {
self.flush_if_completed_line()?;
return self.buffer.write(buf);
}
// Otherwise, arrange for the lines to be written directly to the
// inner writer.
Some(newline_idx) => newline_idx + 1,
};
// Flush existing content to prepare for our write. We have to do this
// before attempting to write `buf` in order to maintain consistency;
// if we add `buf` to the buffer then try to flush it all at once,
// we're obligated to return Ok(), which would mean suppressing any
// errors that occur during flush.
self.buffer.flush_buf()?;
// This is what we're going to try to write directly to the inner
// writer. The rest will be buffered, if nothing goes wrong.
let lines = &buf[..newline_idx];
// Write `lines` directly to the inner writer. In keeping with the
// `write` convention, make at most one attempt to add new (unbuffered)
// data. Because this write doesn't touch the BufWriter state directly,
// and the buffer is known to be empty, we don't need to worry about
// self.buffer.panicked here.
let flushed = self.inner_mut().write(lines)?;
// If buffer returns Ok(0), propagate that to the caller without
// doing additional buffering; otherwise we're just guaranteeing
// an "ErrorKind::WriteZero" later.
if flushed == 0 {
return Ok(0);
}
// Now that the write has succeeded, buffer the rest (or as much of
// the rest as possible). If there were any unwritten newlines, we
// only buffer out to the last unwritten newline that fits in the
// buffer; this helps prevent flushing partial lines on subsequent
// calls to LineWriterShim::write.
// Handle the cases in order of most-common to least-common, under
// the presumption that most writes succeed in totality, and that most
// writes are smaller than the buffer.
// - Is this a partial line (ie, no newlines left in the unwritten tail)
// - If not, does the data out to the last unwritten newline fit in
// the buffer?
// - If not, scan for the last newline that *does* fit in the buffer
let tail = if flushed >= newline_idx {
&buf[flushed..]
} else if newline_idx - flushed <= self.buffer.capacity() {
&buf[flushed..newline_idx]
} else {
let scan_area = &buf[flushed..];
let scan_area = &scan_area[..self.buffer.capacity()];
match memchr::memrchr(b'\n', scan_area) {
Some(newline_idx) => &scan_area[..newline_idx + 1],
None => scan_area,
}
};
let buffered = self.buffer.write_to_buf(tail);
Ok(flushed + buffered)
}
fn flush(&mut self) -> io::Result<()> {
self.buffer.flush()
}
/// Write some vectored data into this BufReader with line buffering. This
/// means that, if any newlines are present in the data, the data up to
/// and including the buffer containing the last newline is sent directly
/// to the inner writer, and the data after it is buffered. Returns the
/// number of bytes written.
///
/// This function operates on a "best effort basis"; in keeping with the
/// convention of `Write::write`, it makes at most one attempt to write
/// new data to the underlying writer.
///
/// Because this function attempts to send completed lines to the underlying
/// writer, it will also flush the existing buffer if it contains any
/// newlines.
///
/// Because sorting through an array of `IoSlice` can be a bit convoluted,
/// This method differs from write in the following ways:
///
/// - It attempts to write the full content of all the buffers up to and
/// including the one containing the last newline. This means that it
/// may attempt to write a partial line, that buffer has data past the
/// newline.
/// - If the write only reports partial success, it does not attempt to
/// find the precise location of the written bytes and buffer the rest.
///
/// If the underlying vector doesn't support vectored writing, we instead
/// simply write the first non-empty buffer with `write`. This way, we
/// get the benefits of more granular partial-line handling without losing
/// anything in efficiency
fn write_vectored(&mut self, bufs: &[IoSlice<'_>]) -> io::Result<usize> {
// If there's no specialized behavior for write_vectored, just use
// write. This has the benefit of more granular partial-line handling.
if !self.is_write_vectored() {
return match bufs.iter().find(|buf| !buf.is_empty()) {
Some(buf) => self.write(buf),
None => Ok(0),
};
}
// Find the buffer containing the last newline
let last_newline_buf_idx = bufs
.iter()
.enumerate()
.rev()
.find_map(|(i, buf)| memchr::memchr(b'\n', buf).map(|_| i));
// If there are no new newlines (that is, if this write is less than
// one line), just do a regular buffered write
let last_newline_buf_idx = match last_newline_buf_idx {
// No newlines; just do a normal buffered write
None => {
self.flush_if_completed_line()?;
return self.buffer.write_vectored(bufs);
}
Some(i) => i,
};
// Flush existing content to prepare for our write
self.buffer.flush_buf()?;
// This is what we're going to try to write directly to the inner
// writer. The rest will be buffered, if nothing goes wrong.
let (lines, tail) = bufs.split_at(last_newline_buf_idx + 1);
// Write `lines` directly to the inner writer. In keeping with the
// `write` convention, make at most one attempt to add new (unbuffered)
// data. Because this write doesn't touch the BufWriter state directly,
// and the buffer is known to be empty, we don't need to worry about
// self.panicked here.
let flushed = self.inner_mut().write_vectored(lines)?;
// If inner returns Ok(0), propagate that to the caller without
// doing additional buffering; otherwise we're just guaranteeing
// an "ErrorKind::WriteZero" later.
if flushed == 0 {
return Ok(0);
}
// Don't try to reconstruct the exact amount written; just bail
// in the event of a partial write
let lines_len = lines.iter().map(|buf| buf.len()).sum();
if flushed < lines_len {
return Ok(flushed);
}
// Now that the write has succeeded, buffer the rest (or as much of the
// rest as possible)
let buffered: usize = tail
.iter()
.filter(|buf| !buf.is_empty())
.map(|buf| self.buffer.write_to_buf(buf))
.take_while(|&n| n > 0)
.sum();
Ok(flushed + buffered)
}
fn is_write_vectored(&self) -> bool {
self.buffer.is_write_vectored()
}
/// Write some data into this BufReader with line buffering. This means
/// that, if any newlines are present in the data, the data up to the last
/// newline is sent directly to the underlying writer, and data after it
/// is buffered.
///
/// Because this function attempts to send completed lines to the underlying
/// writer, it will also flush the existing buffer if it contains any
/// newlines, even if the incoming data does not contain any newlines.
fn write_all(&mut self, buf: &[u8]) -> io::Result<()> {
match memchr::memrchr(b'\n', buf) {
// If there are no new newlines (that is, if this write is less than
// one line), just do a regular buffered write (which may flush if
// we exceed the inner buffer's size)
None => {
self.flush_if_completed_line()?;
self.buffer.write_all(buf)
}
Some(newline_idx) => {
let (lines, tail) = buf.split_at(newline_idx + 1);
if self.buffered().is_empty() {
self.inner_mut().write_all(lines)?;
} else {
// If there is any buffered data, we add the incoming lines
// to that buffer before flushing, which saves us at least
// one write call. We can't really do this with `write`,
// since we can't do this *and* not suppress errors *and*
// report a consistent state to the caller in a return
// value, but here in write_all it's fine.
self.buffer.write_all(lines)?;
self.buffer.flush_buf()?;
}
self.buffer.write_all(tail)
}
}
}
}
/// Wraps a writer and buffers output to it, flushing whenever a newline
/// (`0x0a`, `'\n'`) is detected.
///
/// The [`BufWriter`] struct wraps a writer and buffers its output.
/// But it only does this batched write when it goes out of scope, or when the
/// internal buffer is full. Sometimes, you'd prefer to write each line as it's
/// completed, rather than the entire buffer at once. Enter `LineWriter`. It
/// does exactly that.
///
/// Like [`BufWriter`], a `LineWriter`’s buffer will also be flushed when the
/// `LineWriter` goes out of scope or when its internal buffer is full.
///
/// If there's still a partial line in the buffer when the `LineWriter` is
/// dropped, it will flush those contents.
///
/// # Examples
///
/// We can use `LineWriter` to write one line at a time, significantly
/// reducing the number of actual writes to the file.
///
/// ```no_run
/// use std::fs::{self, File};
/// use std::io::prelude::*;
/// use std::io::LineWriter;
///
/// fn main() -> std::io::Result<()> {
/// let road_not_taken = b"I shall be telling this with a sigh
/// Somewhere ages and ages hence:
/// Two roads diverged in a wood, and I -
/// I took the one less traveled by,
/// And that has made all the difference.";
///
/// let file = File::create("poem.txt")?;
/// let mut file = LineWriter::new(file);
///
/// file.write_all(b"I shall be telling this with a sigh")?;
///
/// // No bytes are written until a newline is encountered (or
/// // the internal buffer is filled).
/// assert_eq!(fs::read_to_string("poem.txt")?, "");
/// file.write_all(b"\n")?;
/// assert_eq!(
/// fs::read_to_string("poem.txt")?,
/// "I shall be telling this with a sigh\n",
/// );
///
/// // Write the rest of the poem.
/// file.write_all(b"Somewhere ages and ages hence:
/// Two roads diverged in a wood, and I -
/// I took the one less traveled by,
/// And that has made all the difference.")?;
///
/// // The last line of the poem doesn't end in a newline, so
/// // we have to flush or drop the `LineWriter` to finish
/// // writing.
/// file.flush()?;
///
/// // Confirm the whole poem was written.
/// assert_eq!(fs::read("poem.txt")?, &road_not_taken[..]);
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub struct LineWriter<W: Write> {
inner: BufWriter<W>,
}
impl<W: Write> LineWriter<W> {
/// Creates a new `LineWriter`.
///
/// # Examples
///
/// ```no_run
/// use std::fs::File;
/// use std::io::LineWriter;
///
/// fn main() -> std::io::Result<()> {
/// let file = File::create("poem.txt")?;
/// let file = LineWriter::new(file);
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn new(inner: W) -> LineWriter<W> {
// Lines typically aren't that long, don't use a giant buffer
LineWriter::with_capacity(1024, inner)
}
/// Creates a new `LineWriter` with a specified capacity for the internal
/// buffer.
///
/// # Examples
///
/// ```no_run
/// use std::fs::File;
/// use std::io::LineWriter;
///
/// fn main() -> std::io::Result<()> {
/// let file = File::create("poem.txt")?;
/// let file = LineWriter::with_capacity(100, file);
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn with_capacity(capacity: usize, inner: W) -> LineWriter<W> {
LineWriter { inner: BufWriter::with_capacity(capacity, inner) }
}
/// Gets a reference to the underlying writer.
///
/// # Examples
///
/// ```no_run
/// use std::fs::File;
/// use std::io::LineWriter;
///
/// fn main() -> std::io::Result<()> {
/// let file = File::create("poem.txt")?;
/// let file = LineWriter::new(file);
///
/// let reference = file.get_ref();
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn get_ref(&self) -> &W {
self.inner.get_ref()
}
/// Gets a mutable reference to the underlying writer.
///
/// Caution must be taken when calling methods on the mutable reference
/// returned as extra writes could corrupt the output stream.
///
/// # Examples
///
/// ```no_run
/// use std::fs::File;
/// use std::io::LineWriter;
///
/// fn main() -> std::io::Result<()> {
/// let file = File::create("poem.txt")?;
/// let mut file = LineWriter::new(file);
///
/// // we can use reference just like file
/// let reference = file.get_mut();
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn get_mut(&mut self) -> &mut W {
self.inner.get_mut()
}
/// Unwraps this `LineWriter`, returning the underlying writer.
///
/// The internal buffer is written out before returning the writer.
///
/// # Errors
///
/// An [`Err`] will be returned if an error occurs while flushing the buffer.
///
/// # Examples
///
/// ```no_run
/// use std::fs::File;
/// use std::io::LineWriter;
///
/// fn main() -> std::io::Result<()> {
/// let file = File::create("poem.txt")?;
///
/// let writer: LineWriter<File> = LineWriter::new(file);
///
/// let file: File = writer.into_inner()?;
/// Ok(())
/// }
/// ```
#[stable(feature = "rust1", since = "1.0.0")]
pub fn into_inner(self) -> Result<W, IntoInnerError<LineWriter<W>>> {
self.inner
.into_inner()
.map_err(|IntoInnerError(buf, e)| IntoInnerError(LineWriter { inner: buf }, e))
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W: Write> Write for LineWriter<W> {
fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
LineWriterShim::new(&mut self.inner).write(buf)
}
fn flush(&mut self) -> io::Result<()> {
self.inner.flush()
}
fn write_vectored(&mut self, bufs: &[IoSlice<'_>]) -> io::Result<usize> {
LineWriterShim::new(&mut self.inner).write_vectored(bufs)
}
fn is_write_vectored(&self) -> bool {
self.inner.is_write_vectored()
}
fn write_all(&mut self, buf: &[u8]) -> io::Result<()> {
LineWriterShim::new(&mut self.inner).write_all(buf)
}
fn write_all_vectored(&mut self, bufs: &mut [IoSlice<'_>]) -> io::Result<()> {
LineWriterShim::new(&mut self.inner).write_all_vectored(bufs)
}
fn write_fmt(&mut self, fmt: fmt::Arguments<'_>) -> io::Result<()> {
LineWriterShim::new(&mut self.inner).write_fmt(fmt)
}
}
#[stable(feature = "rust1", since = "1.0.0")]
impl<W: Write> fmt::Debug for LineWriter<W>
where
W: fmt::Debug,
{
fn fmt(&self, fmt: &mut fmt::Formatter<'_>) -> fmt::Result {
fmt.debug_struct("LineWriter")
.field("writer", &self.inner.inner)
.field(
"buffer",
&format_args!("{}/{}", self.inner.buf.len(), self.inner.buf.capacity()),
)
.finish()
}
}