Merge pull request #2712 from jfinkels/head-take-lines-reader

head: use std::io::copy() with TakeLines reader

Cargo.lock

@@ -2507,6 +2507,7 @@ name = "uu_head"
 version = "0.0.7"
 dependencies = [
  "clap",
+ "memchr 2.4.0",
  "uucore",
  "uucore_procs",
 ]

src/uu/head/BENCHMARKING.md

@@ -0,0 +1,41 @@
+# Benchmarking to measure performance
+
+To compare the performance of the `uutils` version of `head` with the
+GNU version of `head`, you can use a benchmarking tool like
+[hyperfine][0]. On Ubuntu 18.04 or later, you can install `hyperfine` by
+running
+
+    sudo apt-get install hyperfine
+
+Next, build the `head` binary under the release profile:
+
+    cargo build --release -p uu_head
+
+Now, get a text file to test `head` on. I used the *Complete Works of
+William Shakespeare*, which is in the public domain in the United States
+and most other parts of the world.
+
+    wget -O shakespeare.txt https://www.gutenberg.org/files/100/100-0.txt
+
+This particular file has about 170,000 lines, each of which is no longer
+than 96 characters:
+
+    $ wc -lL shakespeare.txt 
+    170592      96 shakespeare.txt
+
+You could use files of different shapes and sizes to test the
+performance of `head` in different situations. For a larger file, you
+could download a [database dump of Wikidata][1] or some related files
+that the Wikimedia project provides. For example, [this file][2]
+contains about 130 million lines.
+
+Finally, you can compare the performance of the two versions of `head`
+by running, for example,
+
+    hyperfine \
+        "head -n 100000 shakespeare.txt" \
+        "target/release/head -n 100000 shakespeare.txt"
+
+[0]: https://github.com/sharkdp/hyperfine
+[1]: https://www.wikidata.org/wiki/Wikidata:Database_download
+[2]: https://dumps.wikimedia.org/wikidatawiki/20211001/wikidatawiki-20211001-pages-logging.xml.gz

src/uu/head/Cargo.toml

@@ -16,6 +16,7 @@ path = "src/head.rs"
 
 [dependencies]
 clap = { version = "2.33", features = ["wrap_help"] }
+memchr = "2"
 uucore = { version=">=0.0.9", package="uucore", path="../../uucore", features=["ringbuffer"] }
 uucore_procs = { version=">=0.0.6", package="uucore_procs", path="../../uucore_procs" }
 

src/uu/head/src/head.rs

@@ -3,18 +3,21 @@
 //  * For the full copyright and license information, please view the LICENSE
 //  * file that was distributed with this source code.
 
-// spell-checker:ignore (vars) zlines
+// spell-checker:ignore (vars) zlines BUFWRITER
 
 use clap::{crate_version, App, Arg};
 use std::convert::TryFrom;
 use std::ffi::OsString;
-use std::io::{self, ErrorKind, Read, Seek, SeekFrom, Write};
+use std::io::{self, BufWriter, ErrorKind, Read, Seek, SeekFrom, Write};
 use uucore::display::Quotable;
 use uucore::error::{UResult, USimpleError};
 use uucore::show_error_custom_description;
 
 const BUF_SIZE: usize = 65536;
 
+/// The capacity in bytes for buffered writers.
+const BUFWRITER_CAPACITY: usize = 16_384; // 16 kilobytes
+
 const ABOUT: &str = "\
                      Print the first 10 lines of each FILE to standard output.\n\
                      With more than one FILE, precede each with a header giving the file name.\n\
@@ -34,10 +37,10 @@ mod options {
 }
 mod lines;
 mod parse;
-mod split;
 mod take;
 use lines::zlines;
 use take::take_all_but;
+use take::take_lines;
 
 pub fn uu_app() -> App<'static, 'static> {
     App::new(uucore::util_name())
@@ -208,26 +211,18 @@ where
 }
 
 fn read_n_lines(input: &mut impl std::io::BufRead, n: usize, zero: bool) -> std::io::Result<()> {
-    if n == 0 {
-        return Ok(());
-    }
+    // Read the first `n` lines from the `input` reader.
+    let separator = if zero { b'\0' } else { b'\n' };
+    let mut reader = take_lines(input, n, separator);
+
+    // Write those bytes to `stdout`.
     let stdout = std::io::stdout();
-    let mut stdout = stdout.lock();
-    let mut lines = 0usize;
-    split::walk_lines(input, zero, |e| match e {
-        split::Event::Data(dat) => {
-            stdout.write_all(dat)?;
-            Ok(true)
-        }
-        split::Event::Line => {
-            lines += 1;
-            if lines == n {
-                Ok(false)
-            } else {
-                Ok(true)
-            }
-        }
-    })
+    let stdout = stdout.lock();
+    let mut writer = BufWriter::with_capacity(BUFWRITER_CAPACITY, stdout);
+
+    io::copy(&mut reader, &mut writer)?;
+
+    Ok(())
 }
 
 fn read_but_last_n_bytes(input: &mut impl std::io::BufRead, n: usize) -> std::io::Result<()> {

src/uu/head/src/split.rs

@@ -1,60 +0,0 @@
-#[derive(Debug)]
-pub enum Event<'a> {
-    Data(&'a [u8]),
-    Line,
-}
-/// Loops over the lines read from a BufRead.
-/// # Arguments
-/// * `input` the ReadBuf to read from
-/// * `zero` whether to use 0u8 as a line delimiter
-/// * `on_event` a closure receiving some bytes read in a slice, or
-///     event signalling a line was just read.
-///     this is guaranteed to be signalled *directly* after the
-///     slice containing the (CR on win)LF / 0 is passed
-///
-///     Return whether to continue
-pub fn walk_lines<F>(
-    input: &mut impl std::io::BufRead,
-    zero: bool,
-    mut on_event: F,
-) -> std::io::Result<()>
-where
-    F: FnMut(Event) -> std::io::Result<bool>,
-{
-    let mut buffer = [0u8; super::BUF_SIZE];
-    loop {
-        let read = loop {
-            match input.read(&mut buffer) {
-                Ok(n) => break n,
-                Err(e) => match e.kind() {
-                    std::io::ErrorKind::Interrupted => {}
-                    _ => return Err(e),
-                },
-            }
-        };
-        if read == 0 {
-            return Ok(());
-        }
-        let mut base = 0usize;
-        for (i, byte) in buffer[..read].iter().enumerate() {
-            match byte {
-                b'\n' if !zero => {
-                    on_event(Event::Data(&buffer[base..=i]))?;
-                    base = i + 1;
-                    if !on_event(Event::Line)? {
-                        return Ok(());
-                    }
-                }
-                0u8 if zero => {
-                    on_event(Event::Data(&buffer[base..=i]))?;
-                    base = i + 1;
-                    if !on_event(Event::Line)? {
-                        return Ok(());
-                    }
-                }
-                _ => {}
-            }
-        }
-        on_event(Event::Data(&buffer[base..read]))?;
-    }
-}

src/uu/head/src/take.rs

@@ -1,4 +1,8 @@
 //! Take all but the last elements of an iterator.
+use std::io::Read;
+
+use memchr::memchr_iter;
+
 use uucore::ringbuffer::RingBuffer;
 
 /// Create an iterator over all but the last `n` elements of `iter`.
@@ -58,10 +62,63 @@ where
     }
 }
 
+/// Like `std::io::Take`, but for lines instead of bytes.
+///
+/// This struct is generally created by calling [`take_lines`] on a
+/// reader. Please see the documentation of [`take`] for more
+/// details.
+pub struct TakeLines<T> {
+    inner: T,
+    limit: usize,
+    separator: u8,
+}
+
+impl<T: Read> Read for TakeLines<T> {
+    /// Read bytes from a buffer up to the requested number of lines.
+    fn read(&mut self, buf: &mut [u8]) -> std::io::Result<usize> {
+        if self.limit == 0 {
+            return Ok(0);
+        }
+        match self.inner.read(buf) {
+            Ok(0) => Ok(0),
+            Ok(n) => {
+                for i in memchr_iter(self.separator, &buf[..n]) {
+                    self.limit -= 1;
+                    if self.limit == 0 {
+                        return Ok(i + 1);
+                    }
+                }
+                Ok(n)
+            }
+            Err(e) => Err(e),
+        }
+    }
+}
+
+/// Create an adaptor that will read at most `limit` lines from a given reader.
+///
+/// This function returns a new instance of `Read` that will read at
+/// most `limit` lines, after which it will always return EOF
+/// (`Ok(0)`).
+///
+/// The `separator` defines the character to interpret as the line
+/// ending. For the usual notion of "line", set this to `b'\n'`.
+pub fn take_lines<R>(reader: R, limit: usize, separator: u8) -> TakeLines<R> {
+    TakeLines {
+        inner: reader,
+        limit,
+        separator,
+    }
+}
+
 #[cfg(test)]
 mod tests {
 
+    use std::io::BufRead;
+    use std::io::BufReader;
+
     use crate::take::take_all_but;
+    use crate::take::take_lines;
 
     #[test]
     fn test_fewer_elements() {
@@ -90,4 +147,33 @@ mod tests {
         assert_eq!(Some(&2), iter.next());
         assert_eq!(None, iter.next());
     }
+
+    #[test]
+    fn test_zero_lines() {
+        let input_reader = std::io::Cursor::new("a\nb\nc\n");
+        let output_reader = BufReader::new(take_lines(input_reader, 0, b'\n'));
+        let mut iter = output_reader.lines().map(|l| l.unwrap());
+        assert_eq!(None, iter.next());
+    }
+
+    #[test]
+    fn test_fewer_lines() {
+        let input_reader = std::io::Cursor::new("a\nb\nc\n");
+        let output_reader = BufReader::new(take_lines(input_reader, 2, b'\n'));
+        let mut iter = output_reader.lines().map(|l| l.unwrap());
+        assert_eq!(Some(String::from("a")), iter.next());
+        assert_eq!(Some(String::from("b")), iter.next());
+        assert_eq!(None, iter.next());
+    }
+
+    #[test]
+    fn test_more_lines() {
+        let input_reader = std::io::Cursor::new("a\nb\nc\n");
+        let output_reader = BufReader::new(take_lines(input_reader, 4, b'\n'));
+        let mut iter = output_reader.lines().map(|l| l.unwrap());
+        assert_eq!(Some(String::from("a")), iter.next());
+        assert_eq!(Some(String::from("b")), iter.next());
+        assert_eq!(Some(String::from("c")), iter.next());
+        assert_eq!(None, iter.next());
+    }
 }