chore: prepare bytes v1.7.2 (#736)

Closes #730

.github/workflows/ci.yml

@@ -7,10 +7,13 @@ on:
   push:
     branches:
       - master
+  schedule:
+    - cron: '0 2 * * 0'
 
 env:
   RUSTFLAGS: -Dwarnings
   RUST_BACKTRACE: 1
+  nightly: nightly-2024-09-15
 
 defaults:
   run:
@@ -22,11 +25,11 @@ jobs:
     name: rustfmt
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v4
     - name: Install Rust
-      run: rustup update stable && rustup default stable
+      run: rustup update stable
     - name: Check formatting
-      run: cargo fmt --all -- --check
+      run: cargo fmt --all --check
 
   # TODO
   # # Apply clippy lints
@@ -34,7 +37,7 @@ jobs:
   #   name: clippy
   #   runs-on: ubuntu-latest
   #   steps:
-  #   - uses: actions/checkout@v2
+  #   - uses: actions/checkout@v4
   #   - name: Apply clippy lints
   #     run: cargo clippy --all-features
 
@@ -47,11 +50,11 @@ jobs:
     name: minrust
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
-      - name: Install Rust
-        run: rustup update 1.39.0 && rustup default 1.39.0
+      - uses: actions/checkout@v4
+      - name: Install cargo-hack
+        uses: taiki-e/install-action@cargo-hack
       - name: Check
-        run: . ci/test-stable.sh check
+        run: cargo hack check --feature-powerset --optional-deps --rust-version
 
   # Stable
   stable:
@@ -64,26 +67,27 @@ jobs:
           - windows-latest
     runs-on: ${{ matrix.os }}
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - name: Install Rust
         # --no-self-update is necessary because the windows environment cannot self-update rustup.exe.
-        run: rustup update stable --no-self-update && rustup default stable
+        run: rustup update stable --no-self-update
+      - name: Install cargo-hack
+        uses: taiki-e/install-action@cargo-hack
       - name: Test
-        run: . ci/test-stable.sh test
+        run: ci/test-stable.sh test
 
   # Nightly
   nightly:
     name: nightly
-    env:
-      # Pin nightly to avoid being impacted by breakage
-      RUST_VERSION: nightly-2019-09-25
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - name: Install Rust
-        run: rustup update $RUST_VERSION && rustup default $RUST_VERSION
+        run: rustup update $nightly && rustup default $nightly
+      - name: Install cargo-hack
+        uses: taiki-e/install-action@cargo-hack
       - name: Test
-        run: . ci/test-stable.sh test
+        run: ci/test-stable.sh test
 
   # Run tests on some extra platforms
   cross:
@@ -96,15 +100,16 @@ jobs:
           - powerpc-unknown-linux-gnu
           - powerpc64-unknown-linux-gnu
           - wasm32-unknown-unknown
-    runs-on: ubuntu-16.04
+    runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - name: Install Rust
-        run: rustup update stable && rustup default stable
+        run: rustup update stable
+      - name: Install cross
+        uses: taiki-e/install-action@cross
+        if: matrix.target != 'wasm32-unknown-unknown'
       - name: cross build --target ${{ matrix.target }}
-        run: |
-          cargo install cross
-          cross build --target ${{ matrix.target }}
+        run: cross build --target ${{ matrix.target }}
         if: matrix.target != 'wasm32-unknown-unknown'
       # WASM support
       - name: cargo build --target ${{ matrix.target }}
@@ -118,25 +123,32 @@ jobs:
     name: tsan
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - name: Install Rust
-        run: rustup update nightly && rustup default nightly
+        run: rustup update $nightly && rustup default $nightly
       - name: Install rust-src
         run: rustup component add rust-src
       - name: ASAN / TSAN
-        run: . ci/tsan.sh
+        run: ci/tsan.sh
+
+  miri:
+    name: miri
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Install Rust
+        run: rustup update $nightly && rustup default $nightly
+      - name: Miri
+        run: ci/miri.sh
 
   # Loom
   loom:
     name: loom
-    env:
-      # Pin nightly to avoid being impacted by breakage
-      RUST_VERSION: nightly-2020-05-19
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - name: Install Rust
-        run: rustup update $RUST_VERSION && rustup default $RUST_VERSION
+        run: rustup update $nightly && rustup default $nightly
       - name: Loom tests
         run: RUSTFLAGS="--cfg loom -Dwarnings" cargo test --lib
 
@@ -149,13 +161,18 @@ jobs:
       - nightly
       - minrust
       - cross
+      - tsan
+      - miri
+      - loom
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - name: Install Rust
-        run: rustup update stable && rustup default stable
+        run: rustup update $nightly && rustup default $nightly
       - name: Build documentation
         run: cargo doc --no-deps --all-features
+        env:
+          RUSTDOCFLAGS: --cfg docsrs
       - name: Publish documentation
         run: |
           cd target/doc

CHANGELOG.md

@@ -1,3 +1,208 @@
+# 1.7.2 (September 17, 2024)
+
+### Fixed
+
+- Fix default impl of `Buf::{get_int, get_int_le}` (#732)
+
+### Documented
+
+- Fix double spaces in comments and doc comments (#731)
+
+### Internal changes
+
+- Ensure BytesMut::advance reduces capacity (#728) 
+
+# 1.7.1 (August 1, 2024)
+
+This release reverts the following change due to a regression:
+
+- Reuse capacity when possible in `<BytesMut as Buf>::advance` impl (#698)
+
+The revert can be found at #726.
+
+# 1.7.0 (July 31, 2024)
+
+### Added
+
+- Add conversion from `Bytes` to `BytesMut` (#695, #710)
+- Add reclaim method without additional allocation (#686)
+
+### Documented
+
+- Clarify how `BytesMut::zeroed` works (#714)
+- Clarify the behavior of `Buf::chunk` (#717)
+
+### Changed
+
+- Change length condition of `BytesMut::truncate`
+- Reuse capacity when possible in `<BytesMut as Buf>::advance` impl (#698)
+- Improve `must_use` suggestion of `BytesMut::split` (#699)
+
+### Internal changes
+
+- Use `ManuallyDrop` instead of `mem::forget` (#678)
+- Don't set `len` in `BytesMut::reserve` (#682)
+- Optimize `Bytes::copy_to_bytes` (#688)
+- Refactor `BytesMut::truncate` (#694)
+- Refactor `BytesMut::resize` (#696)
+- Reorder assertion in `Bytes::split_to`, `Bytes::split_off` (#689, #693)
+- Use `offset_from` in more places (#705)
+- Correct the wrong usage of `IntoIter` (#707)
+
+# 1.6.1 (July 13, 2024)
+
+This release fixes a bug where `Bytes::is_unique` returns incorrect values when
+the `Bytes` originates from a shared `BytesMut`. (#718)
+
+# 1.6.0 (March 22, 2024)
+
+### Added
+
+- Add `Bytes::is_unique` (#643)
+
+### Documented
+
+- Fix changelog typo (#628)
+- Fix some spelling mistakes (#633)
+- Typo fix (#637)
+- Fix broken links (#639)
+- Add security policy (#649)
+
+### Internal changes
+
+- Move comment to correct constant (#629)
+- Various cleanup (#635)
+- Simplify `UninitSlice::as_uninit_slice_mut()` logic (#644)
+- Use `self.` instead of `Self::` (#642)
+- `BytesMut`: Assert alignment of `Shared` (#652)
+- Remove unnecessary namespace qualifier (#660)
+- Remove an unnecessary else branch (#662)
+- Remove unreachable else branch (#661)
+- make parameter mut in `From<Vec>` (#667)
+- Restore commented tests (#665)
+- Use `sub` instead of `offset` (#668)
+- Calculate original capacity only if necessary (#666)
+- `set_vec_pos` does not need a second parameter (#672)
+- `get_vec_pos`: use `&self` instead of `&mut self` (#670)
+- Refactor `split_at`/`split_to` (#663)
+- Use `Iterator` from the prelude (#673)
+- `copy_to_bytes`: Add panic section to docs (#676)
+- Remove redundant reserve call (#674)
+- Use `ManuallyDrop` instead of `mem::forget` (#675)
+
+
+# 1.5.0 (September 7, 2023)
+
+### Added
+
+- Add `UninitSlice::{new,uninit}` (#598, #599)
+- Implement `BufMut` for `&mut [MaybeUninit<u8>]` (#597)
+
+### Changed
+
+- Mark `BytesMut::extend_from_slice` as inline (#595)
+
+# 1.4.0 (January 31, 2023)
+
+### Added
+
+- Make `IntoIter` constructor public (#581)
+
+### Fixed
+
+- Avoid large reallocations when freezing `BytesMut` (#592)
+
+### Documented
+
+- Document which functions require `std` (#591)
+- Fix duplicate "the the" typos (#585)
+
+# 1.3.0 (November 20, 2022)
+
+### Added
+
+- Rename and expose `BytesMut::spare_capacity_mut` (#572)
+- Implement native-endian get and put functions for `Buf` and `BufMut` (#576)
+
+### Fixed
+
+- Don't have important data in unused capacity when calling reserve (#563)
+
+### Documented
+
+- `Bytes::new` etc should return `Self` not `Bytes` (#568)
+
+# 1.2.1 (July 30, 2022)
+
+### Fixed
+
+- Fix unbounded memory growth when using `reserve` (#560)
+
+# 1.2.0 (July 19, 2022)
+
+### Added
+
+- Add `BytesMut::zeroed` (#517)
+- Implement `Extend<Bytes>` for `BytesMut` (#527)
+- Add conversion from `BytesMut` to `Vec<u8>` (#543, #554)
+- Add conversion from `Bytes` to `Vec<u8>` (#547)
+- Add `UninitSlice::as_uninit_slice_mut()` (#548)
+- Add const to `Bytes::{len,is_empty}` (#514)
+
+### Changed
+
+- Reuse vector in `BytesMut::reserve` (#539, #544)
+
+### Fixed
+
+- Make miri happy (#515, #523, #542, #545, #553)
+- Make tsan happy (#541)
+- Fix `remaining_mut()` on chain (#488)
+- Fix amortized asymptotics of `BytesMut` (#555)
+
+### Documented
+
+- Redraw layout diagram with box drawing characters (#539)
+- Clarify `BytesMut::unsplit` docs (#535)
+
+# 1.1.0 (August 25, 2021)
+
+### Added
+
+- `BufMut::put_bytes(self, val, cnt)` (#487)
+- Implement `From<Box<[u8]>>` for `Bytes` (#504)
+
+### Changed
+
+- Override `put_slice` for `&mut [u8]` (#483)
+- Panic on integer overflow in `Chain::remaining` (#482)
+- Add inline tags to `UninitSlice` methods (#443)
+- Override `copy_to_bytes` for Chain and Take (#481)
+- Keep capacity when unsplit on empty other buf (#502)
+
+### Documented
+
+- Clarify `BufMut` allocation guarantees (#501)
+- Clarify `BufMut::put_int` behavior (#486)
+- Clarify actions of `clear` and `truncate`. (#508)
+
+# 1.0.1 (January 11, 2021)
+
+### Changed
+- mark `Vec::put_slice` with `#[inline]` (#459)
+
+### Fixed
+- Fix deprecation warning (#457)
+- use `Box::into_raw` instead of `mem::forget`-in-disguise (#458)
+
+# 1.0.0 (December 22, 2020)
+
+### Changed
+- Rename `Buf`/`BufMut` methods `bytes()` and `bytes_mut()` to `chunk()` and `chunk_mut()` (#450)
+
+### Removed
+- remove unused Buf implementation. (#449)
+
 # 0.6.0 (October 21, 2020)
 
 API polish in preparation for a 1.0 release.

Cargo.toml

@@ -2,23 +2,21 @@
 
 name = "bytes"
 # When releasing to crates.io:
-# - Update html_root_url.
 # - Update CHANGELOG.md.
-# - Update doc URL.
-# - Create "v0.6.x" git tag.
-version = "0.6.0"
+# - Create "v1.x.y" git tag.
+version = "1.7.2"
+edition = "2018"
+rust-version = "1.39"
 license = "MIT"
 authors = [
     "Carl Lerche <[email protected]>",
     "Sean McArthur <[email protected]>",
 ]
 description = "Types and traits for working with bytes"
-documentation = "https://docs.rs/bytes/0.6.0/bytes/"
 repository = "https://github.com/tokio-rs/bytes"
 readme = "README.md"
 keywords = ["buffers", "zero-copy", "io"]
 categories = ["network-programming", "data-structures"]
-edition = "2018"
 
 [features]
 default = ["std"]
@@ -31,4 +29,7 @@ serde = { version = "1.0.60", optional = true, default-features = false, feature
 serde_test = "1.0"
 
 [target.'cfg(loom)'.dev-dependencies]
-loom = "0.3"
+loom = "0.7"
+
+[package.metadata.docs.rs]
+rustdoc-args = ["--cfg", "docsrs"]

README.md

@@ -18,7 +18,7 @@ To use `bytes`, first add this to your `Cargo.toml`:
 
 ```toml
 [dependencies]
-bytes = "0.6"
+bytes = "1"
 ```
 
 Next, add this to your crate:
@@ -33,7 +33,16 @@ Serde support is optional and disabled by default. To enable use the feature `se
 
 ```toml
 [dependencies]
-bytes = { version = "0.6", features = ["serde"] }
+bytes = { version = "1", features = ["serde"] }
+```
+
+## Building documentation
+
+When building the `bytes` documentation the `docsrs` option should be used, otherwise
+feature gates will not be shown. This requires a nightly toolchain:
+
+```
+RUSTDOCFLAGS="--cfg docsrs" cargo +nightly doc
 ```
 
 ## License

SECURITY.md

@@ -0,0 +1,9 @@
+# Security Policy
+
+Bytes is part of the Tokio project and uses the same security policy as [Tokio][tokio-security].
+
+## Report a security issue
+
+The process for reporting an issue is the same as for [Tokio][tokio-security]. This includes private reporting via security@tokio.rs.
+
+[tokio-security]: https://github.com/tokio-rs/tokio/security/policy

benches/buf.rs

@@ -46,14 +46,14 @@ impl TestBuf {
 }
 impl Buf for TestBuf {
     fn remaining(&self) -> usize {
-        return self.buf.len() - self.pos;
+        self.buf.len() - self.pos
     }
     fn advance(&mut self, cnt: usize) {
         self.pos += cnt;
         assert!(self.pos <= self.buf.len());
         self.next_readlen();
     }
-    fn bytes(&self) -> &[u8] {
+    fn chunk(&self) -> &[u8] {
         if self.readlen == 0 {
             Default::default()
         } else {
@@ -87,8 +87,8 @@ impl Buf for TestBufC {
         self.inner.advance(cnt)
     }
     #[inline(never)]
-    fn bytes(&self) -> &[u8] {
-        self.inner.bytes()
+    fn chunk(&self) -> &[u8] {
+        self.inner.chunk()
     }
 }
 
@@ -159,7 +159,6 @@ macro_rules! bench_group {
 mod get_u8 {
     use super::*;
     bench_group!(get_u8);
-    bench!(option, option);
 }
 mod get_u16 {
     use super::*;

benches/bytes.rs

@@ -47,7 +47,7 @@ fn clone_static(b: &mut Bencher) {
 
     b.iter(|| {
         for _ in 0..1024 {
-            test::black_box(&bytes.clone());
+            test::black_box(test::black_box(&bytes).clone());
         }
     })
 }
@@ -58,7 +58,7 @@ fn clone_shared(b: &mut Bencher) {
 
     b.iter(|| {
         for _ in 0..1024 {
-            test::black_box(&bytes.clone());
+            test::black_box(test::black_box(&bytes).clone());
         }
     })
 }
@@ -70,7 +70,7 @@ fn clone_arc_vec(b: &mut Bencher) {
 
     b.iter(|| {
         for _ in 0..1024 {
-            test::black_box(&bytes.clone());
+            test::black_box(test::black_box(&bytes).clone());
         }
     })
 }
@@ -88,6 +88,7 @@ fn from_long_slice(b: &mut Bencher) {
 #[bench]
 fn slice_empty(b: &mut Bencher) {
     b.iter(|| {
+        // `clone` is to convert to ARC
         let b = Bytes::from(vec![17; 1024]).clone();
         for i in 0..1000 {
             test::black_box(b.slice(i % 100..i % 100));

ci/miri.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+set -e
+
+rustup component add miri
+cargo miri setup
+
+export MIRIFLAGS="-Zmiri-strict-provenance"
+
+cargo miri test
+cargo miri test --target mips64-unknown-linux-gnuabi64

ci/test-stable.sh

@@ -4,9 +4,6 @@ set -ex
 
 cmd="${1:-test}"
 
-# Install cargo-hack for feature flag test
-cargo install cargo-hack
-
 # Run with each feature
 # * --each-feature includes both default/no-default features
 # * --optional-deps is needed for serde feature
@@ -14,8 +11,6 @@ cargo hack "${cmd}" --each-feature --optional-deps
 # Run with all features
 cargo "${cmd}" --all-features
 
-cargo doc --no-deps --all-features
-
 if [[ "${RUST_VERSION}" == "nightly"* ]]; then
     # Check benchmarks
     cargo check --benches

clippy.toml

@@ -0,0 +1 @@
+msrv = "1.39"

src/buf/buf_impl.rs

@@ -1,8 +1,9 @@
 #[cfg(feature = "std")]
 use crate::buf::{reader, Reader};
 use crate::buf::{take, Chain, Take};
-
-use core::{cmp, mem, ptr};
+#[cfg(feature = "std")]
+use crate::{min_u64_usize, saturating_sub_usize_u64};
+use crate::{panic_advance, panic_does_not_fit};
 
 #[cfg(feature = "std")]
 use std::io::IoSlice;
@@ -11,12 +12,17 @@ use alloc::boxed::Box;
 
 macro_rules! buf_get_impl {
     ($this:ident, $typ:tt::$conv:tt) => {{
-        const SIZE: usize = mem::size_of::<$typ>();
+        const SIZE: usize = core::mem::size_of::<$typ>();
+
+        if $this.remaining() < SIZE {
+            panic_advance(SIZE, $this.remaining());
+        }
+
         // try to convert directly from the bytes
         // this Option<ret> trick is to avoid keeping a borrow on self
         // when advance() is called (mut borrow) and to call bytes() only once
         let ret = $this
-            .bytes()
+            .chunk()
             .get(..SIZE)
             .map(|src| unsafe { $typ::$conv(*(src as *const _ as *const [_; SIZE])) });
 
@@ -32,23 +38,40 @@ macro_rules! buf_get_impl {
         }
     }};
     (le => $this:ident, $typ:tt, $len_to_read:expr) => {{
-        debug_assert!(mem::size_of::<$typ>() >= $len_to_read);
+        const SIZE: usize = core::mem::size_of::<$typ>();
 
         // The same trick as above does not improve the best case speed.
         // It seems to be linked to the way the method is optimised by the compiler
-        let mut buf = [0; (mem::size_of::<$typ>())];
-        $this.copy_to_slice(&mut buf[..($len_to_read)]);
+        let mut buf = [0; SIZE];
+
+        let subslice = match buf.get_mut(..$len_to_read) {
+            Some(subslice) => subslice,
+            None => panic_does_not_fit(SIZE, $len_to_read),
+        };
+
+        $this.copy_to_slice(subslice);
         return $typ::from_le_bytes(buf);
     }};
     (be => $this:ident, $typ:tt, $len_to_read:expr) => {{
-        debug_assert!(mem::size_of::<$typ>() >= $len_to_read);
+        const SIZE: usize = core::mem::size_of::<$typ>();
 
-        let mut buf = [0; (mem::size_of::<$typ>())];
-        $this.copy_to_slice(&mut buf[mem::size_of::<$typ>() - ($len_to_read)..]);
+        let slice_at = match SIZE.checked_sub($len_to_read) {
+            Some(slice_at) => slice_at,
+            None => panic_does_not_fit(SIZE, $len_to_read),
+        };
+
+        let mut buf = [0; SIZE];
+        $this.copy_to_slice(&mut buf[slice_at..]);
         return $typ::from_be_bytes(buf);
     }};
 }
 
+// https://en.wikipedia.org/wiki/Sign_extension
+fn sign_extend(val: u64, nbytes: usize) -> i64 {
+    let shift = (8 - nbytes) * 8;
+    (val << shift) as i64 >> shift
+}
+
 /// Read bytes from a buffer.
 ///
 /// A buffer stores bytes in memory such that read operations are infallible.
@@ -78,7 +101,7 @@ pub trait Buf {
     /// the buffer.
     ///
     /// This value is greater than or equal to the length of the slice returned
-    /// by `bytes`.
+    /// by `chunk()`.
     ///
     /// # Examples
     ///
@@ -115,31 +138,36 @@ pub trait Buf {
     ///
     /// let mut buf = &b"hello world"[..];
     ///
-    /// assert_eq!(buf.bytes(), &b"hello world"[..]);
+    /// assert_eq!(buf.chunk(), &b"hello world"[..]);
     ///
     /// buf.advance(6);
     ///
-    /// assert_eq!(buf.bytes(), &b"world"[..]);
+    /// assert_eq!(buf.chunk(), &b"world"[..]);
     /// ```
     ///
     /// # Implementer notes
     ///
-    /// This function should never panic. Once the end of the buffer is reached,
-    /// i.e., `Buf::remaining` returns 0, calls to `bytes` should return an
-    /// empty slice.
-    fn bytes(&self) -> &[u8];
+    /// This function should never panic. `chunk()` should return an empty
+    /// slice **if and only if** `remaining()` returns 0. In other words,
+    /// `chunk()` returning an empty slice implies that `remaining()` will
+    /// return 0 and `remaining()` returning 0 implies that `chunk()` will
+    /// return an empty slice.
+    // The `chunk` method was previously called `bytes`. This alias makes the rename
+    // more easily discoverable.
+    #[cfg_attr(docsrs, doc(alias = "bytes"))]
+    fn chunk(&self) -> &[u8];
 
     /// Fills `dst` with potentially multiple slices starting at `self`'s
     /// current position.
     ///
-    /// If the `Buf` is backed by disjoint slices of bytes, `bytes_vectored` enables
+    /// If the `Buf` is backed by disjoint slices of bytes, `chunk_vectored` enables
     /// fetching more than one slice at once. `dst` is a slice of `IoSlice`
     /// references, enabling the slice to be directly used with [`writev`]
     /// without any further conversion. The sum of the lengths of all the
     /// buffers in `dst` will be less than or equal to `Buf::remaining()`.
     ///
     /// The entries in `dst` will be overwritten, but the data **contained** by
-    /// the slices **will not** be modified. If `bytes_vectored` does not fill every
+    /// the slices **will not** be modified. If `chunk_vectored` does not fill every
     /// entry in `dst`, then `dst` is guaranteed to contain all remaining slices
     /// in `self.
     ///
@@ -149,7 +177,7 @@ pub trait Buf {
     /// # Implementer notes
     ///
     /// This function should never panic. Once the end of the buffer is reached,
-    /// i.e., `Buf::remaining` returns 0, calls to `bytes_vectored` must return 0
+    /// i.e., `Buf::remaining` returns 0, calls to `chunk_vectored` must return 0
     /// without mutating `dst`.
     ///
     /// Implementations should also take care to properly handle being called
@@ -157,13 +185,14 @@ pub trait Buf {
     ///
     /// [`writev`]: http://man7.org/linux/man-pages/man2/readv.2.html
     #[cfg(feature = "std")]
-    fn bytes_vectored<'a>(&'a self, dst: &mut [IoSlice<'a>]) -> usize {
+    #[cfg_attr(docsrs, doc(cfg(feature = "std")))]
+    fn chunks_vectored<'a>(&'a self, dst: &mut [IoSlice<'a>]) -> usize {
         if dst.is_empty() {
             return 0;
         }
 
         if self.has_remaining() {
-            dst[0] = IoSlice::new(self.bytes());
+            dst[0] = IoSlice::new(self.chunk());
             1
         } else {
             0
@@ -172,7 +201,7 @@ pub trait Buf {
 
     /// Advance the internal cursor of the Buf
     ///
-    /// The next call to `bytes` will return a slice starting `cnt` bytes
+    /// The next call to `chunk()` will return a slice starting `cnt` bytes
     /// further into the underlying buffer.
     ///
     /// # Examples
@@ -182,11 +211,11 @@ pub trait Buf {
     ///
     /// let mut buf = &b"hello world"[..];
     ///
-    /// assert_eq!(buf.bytes(), &b"hello world"[..]);
+    /// assert_eq!(buf.chunk(), &b"hello world"[..]);
     ///
     /// buf.advance(6);
     ///
-    /// assert_eq!(buf.bytes(), &b"world"[..]);
+    /// assert_eq!(buf.chunk(), &b"world"[..]);
     /// ```
     ///
     /// # Panics
@@ -243,23 +272,18 @@ pub trait Buf {
     ///
     /// # Panics
     ///
-    /// This function panics if `self.remaining() < dst.len()`
-    fn copy_to_slice(&mut self, dst: &mut [u8]) {
-        let mut off = 0;
+    /// This function panics if `self.remaining() < dst.len()`.
+    fn copy_to_slice(&mut self, mut dst: &mut [u8]) {
+        if self.remaining() < dst.len() {
+            panic_advance(dst.len(), self.remaining());
+        }
 
-        assert!(self.remaining() >= dst.len());
+        while !dst.is_empty() {
+            let src = self.chunk();
+            let cnt = usize::min(src.len(), dst.len());
 
-        while off < dst.len() {
-            let cnt;
-
-            unsafe {
-                let src = self.bytes();
-                cnt = cmp::min(src.len(), dst.len() - off);
-
-                ptr::copy_nonoverlapping(src.as_ptr(), dst[off..].as_mut_ptr(), cnt);
-
-                off += cnt;
-            }
+            dst[..cnt].copy_from_slice(&src[..cnt]);
+            dst = &mut dst[cnt..];
 
             self.advance(cnt);
         }
@@ -282,8 +306,10 @@ pub trait Buf {
     ///
     /// This function panics if there is no more remaining data in `self`.
     fn get_u8(&mut self) -> u8 {
-        assert!(self.remaining() >= 1);
-        let ret = self.bytes()[0];
+        if self.remaining() < 1 {
+            panic_advance(1, 0);
+        }
+        let ret = self.chunk()[0];
         self.advance(1);
         ret
     }
@@ -305,8 +331,10 @@ pub trait Buf {
     ///
     /// This function panics if there is no more remaining data in `self`.
     fn get_i8(&mut self) -> i8 {
-        assert!(self.remaining() >= 1);
-        let ret = self.bytes()[0] as i8;
+        if self.remaining() < 1 {
+            panic_advance(1, 0);
+        }
+        let ret = self.chunk()[0] as i8;
         self.advance(1);
         ret
     }
@@ -351,6 +379,29 @@ pub trait Buf {
         buf_get_impl!(self, u16::from_le_bytes);
     }
 
+    /// Gets an unsigned 16 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 2.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09 hello",
+    ///     false => b"\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809, buf.get_u16_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u16_ne(&mut self) -> u16 {
+        buf_get_impl!(self, u16::from_ne_bytes);
+    }
+
     /// Gets a signed 16 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 2.
@@ -391,6 +442,29 @@ pub trait Buf {
         buf_get_impl!(self, i16::from_le_bytes);
     }
 
+    /// Gets a signed 16 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 2.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09 hello",
+    ///     false => b"\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809, buf.get_i16_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i16_ne(&mut self) -> i16 {
+        buf_get_impl!(self, i16::from_ne_bytes);
+    }
+
     /// Gets an unsigned 32 bit integer from `self` in the big-endian byte order.
     ///
     /// The current position is advanced by 4.
@@ -431,6 +505,29 @@ pub trait Buf {
         buf_get_impl!(self, u32::from_le_bytes);
     }
 
+    /// Gets an unsigned 32 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09\xA0\xA1 hello",
+    ///     false => b"\xA1\xA0\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809A0A1, buf.get_u32_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u32_ne(&mut self) -> u32 {
+        buf_get_impl!(self, u32::from_ne_bytes);
+    }
+
     /// Gets a signed 32 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 4.
@@ -471,6 +568,29 @@ pub trait Buf {
         buf_get_impl!(self, i32::from_le_bytes);
     }
 
+    /// Gets a signed 32 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09\xA0\xA1 hello",
+    ///     false => b"\xA1\xA0\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809A0A1, buf.get_i32_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i32_ne(&mut self) -> i32 {
+        buf_get_impl!(self, i32::from_ne_bytes);
+    }
+
     /// Gets an unsigned 64 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 8.
@@ -511,6 +631,29 @@ pub trait Buf {
         buf_get_impl!(self, u64::from_le_bytes);
     }
 
+    /// Gets an unsigned 64 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08 hello",
+    ///     false => b"\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x0102030405060708, buf.get_u64_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u64_ne(&mut self) -> u64 {
+        buf_get_impl!(self, u64::from_ne_bytes);
+    }
+
     /// Gets a signed 64 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 8.
@@ -551,6 +694,29 @@ pub trait Buf {
         buf_get_impl!(self, i64::from_le_bytes);
     }
 
+    /// Gets a signed 64 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08 hello",
+    ///     false => b"\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x0102030405060708, buf.get_i64_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i64_ne(&mut self) -> i64 {
+        buf_get_impl!(self, i64::from_ne_bytes);
+    }
+
     /// Gets an unsigned 128 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 16.
@@ -591,6 +757,29 @@ pub trait Buf {
         buf_get_impl!(self, u128::from_le_bytes);
     }
 
+    /// Gets an unsigned 128 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 16.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08\x09\x10\x11\x12\x13\x14\x15\x16 hello",
+    ///     false => b"\x16\x15\x14\x13\x12\x11\x10\x09\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x01020304050607080910111213141516, buf.get_u128_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u128_ne(&mut self) -> u128 {
+        buf_get_impl!(self, u128::from_ne_bytes);
+    }
+
     /// Gets a signed 128 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 16.
@@ -631,6 +820,29 @@ pub trait Buf {
         buf_get_impl!(self, i128::from_le_bytes);
     }
 
+    /// Gets a signed 128 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 16.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08\x09\x10\x11\x12\x13\x14\x15\x16 hello",
+    ///     false => b"\x16\x15\x14\x13\x12\x11\x10\x09\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x01020304050607080910111213141516, buf.get_i128_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i128_ne(&mut self) -> i128 {
+        buf_get_impl!(self, i128::from_ne_bytes);
+    }
+
     /// Gets an unsigned n-byte integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by `nbytes`.
@@ -671,6 +883,34 @@ pub trait Buf {
         buf_get_impl!(le => self, u64, nbytes);
     }
 
+    /// Gets an unsigned n-byte integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by `nbytes`.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03 hello",
+    ///     false => b"\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x010203, buf.get_uint_ne(3));
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`, or
+    /// if `nbytes` is greater than 8.
+    fn get_uint_ne(&mut self, nbytes: usize) -> u64 {
+        if cfg!(target_endian = "big") {
+            self.get_uint(nbytes)
+        } else {
+            self.get_uint_le(nbytes)
+        }
+    }
+
     /// Gets a signed n-byte integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by `nbytes`.
@@ -686,9 +926,10 @@ pub trait Buf {
     ///
     /// # Panics
     ///
-    /// This function panics if there is not enough remaining data in `self`.
+    /// This function panics if there is not enough remaining data in `self`, or
+    /// if `nbytes` is greater than 8.
     fn get_int(&mut self, nbytes: usize) -> i64 {
-        buf_get_impl!(be => self, i64, nbytes);
+        sign_extend(self.get_uint(nbytes), nbytes)
     }
 
     /// Gets a signed n-byte integer from `self` in little-endian byte order.
@@ -706,9 +947,38 @@ pub trait Buf {
     ///
     /// # Panics
     ///
-    /// This function panics if there is not enough remaining data in `self`.
+    /// This function panics if there is not enough remaining data in `self`, or
+    /// if `nbytes` is greater than 8.
     fn get_int_le(&mut self, nbytes: usize) -> i64 {
-        buf_get_impl!(le => self, i64, nbytes);
+        sign_extend(self.get_uint_le(nbytes), nbytes)
+    }
+
+    /// Gets a signed n-byte integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by `nbytes`.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03 hello",
+    ///     false => b"\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x010203, buf.get_int_ne(3));
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`, or
+    /// if `nbytes` is greater than 8.
+    fn get_int_ne(&mut self, nbytes: usize) -> i64 {
+        if cfg!(target_endian = "big") {
+            self.get_int(nbytes)
+        } else {
+            self.get_int_le(nbytes)
+        }
     }
 
     /// Gets an IEEE754 single-precision (4 bytes) floating point number from
@@ -729,7 +999,7 @@ pub trait Buf {
     ///
     /// This function panics if there is not enough remaining data in `self`.
     fn get_f32(&mut self) -> f32 {
-        f32::from_bits(Self::get_u32(self))
+        f32::from_bits(self.get_u32())
     }
 
     /// Gets an IEEE754 single-precision (4 bytes) floating point number from
@@ -750,7 +1020,31 @@ pub trait Buf {
     ///
     /// This function panics if there is not enough remaining data in `self`.
     fn get_f32_le(&mut self) -> f32 {
-        f32::from_bits(Self::get_u32_le(self))
+        f32::from_bits(self.get_u32_le())
+    }
+
+    /// Gets an IEEE754 single-precision (4 bytes) floating point number from
+    /// `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x3F\x99\x99\x9A hello",
+    ///     false => b"\x9A\x99\x99\x3F hello",
+    /// };
+    /// assert_eq!(1.2f32, buf.get_f32_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_f32_ne(&mut self) -> f32 {
+        f32::from_bits(self.get_u32_ne())
     }
 
     /// Gets an IEEE754 double-precision (8 bytes) floating point number from
@@ -771,7 +1065,7 @@ pub trait Buf {
     ///
     /// This function panics if there is not enough remaining data in `self`.
     fn get_f64(&mut self) -> f64 {
-        f64::from_bits(Self::get_u64(self))
+        f64::from_bits(self.get_u64())
     }
 
     /// Gets an IEEE754 double-precision (8 bytes) floating point number from
@@ -792,7 +1086,31 @@ pub trait Buf {
     ///
     /// This function panics if there is not enough remaining data in `self`.
     fn get_f64_le(&mut self) -> f64 {
-        f64::from_bits(Self::get_u64_le(self))
+        f64::from_bits(self.get_u64_le())
+    }
+
+    /// Gets an IEEE754 double-precision (8 bytes) floating point number from
+    /// `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x3F\xF3\x33\x33\x33\x33\x33\x33 hello",
+    ///     false => b"\x33\x33\x33\x33\x33\x33\xF3\x3F hello",
+    /// };
+    /// assert_eq!(1.2f64, buf.get_f64_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_f64_ne(&mut self) -> f64 {
+        f64::from_bits(self.get_u64_ne())
     }
 
     /// Consumes `len` bytes inside self and returns new instance of `Bytes`
@@ -810,10 +1128,16 @@ pub trait Buf {
     /// let bytes = (&b"hello world"[..]).copy_to_bytes(5);
     /// assert_eq!(&bytes[..], &b"hello"[..]);
     /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if `len > self.remaining()`.
     fn copy_to_bytes(&mut self, len: usize) -> crate::Bytes {
         use super::BufMut;
 
-        assert!(len <= self.remaining(), "`len` greater than remaining");
+        if self.remaining() < len {
+            panic_advance(len, self.remaining());
+        }
 
         let mut ret = crate::BytesMut::with_capacity(len);
         ret.put(self.take(len));
@@ -861,7 +1185,7 @@ pub trait Buf {
     /// let mut chain = b"hello "[..].chain(&b"world"[..]);
     ///
     /// let full = chain.copy_to_bytes(11);
-    /// assert_eq!(full.bytes(), b"hello world");
+    /// assert_eq!(full.chunk(), b"hello world");
     /// ```
     fn chain<U: Buf>(self, next: U) -> Chain<Self, U>
     where
@@ -894,6 +1218,7 @@ pub trait Buf {
     /// assert_eq!(&dst[..11], &b"hello world"[..]);
     /// ```
     #[cfg(feature = "std")]
+    #[cfg_attr(docsrs, doc(cfg(feature = "std")))]
     fn reader(self) -> Reader<Self>
     where
         Self: Sized,
@@ -904,103 +1229,168 @@ pub trait Buf {
 
 macro_rules! deref_forward_buf {
     () => {
+        #[inline]
         fn remaining(&self) -> usize {
             (**self).remaining()
         }
 
-        fn bytes(&self) -> &[u8] {
-            (**self).bytes()
+        #[inline]
+        fn chunk(&self) -> &[u8] {
+            (**self).chunk()
         }
 
         #[cfg(feature = "std")]
-        fn bytes_vectored<'b>(&'b self, dst: &mut [IoSlice<'b>]) -> usize {
-            (**self).bytes_vectored(dst)
+        #[inline]
+        fn chunks_vectored<'b>(&'b self, dst: &mut [IoSlice<'b>]) -> usize {
+            (**self).chunks_vectored(dst)
         }
 
+        #[inline]
         fn advance(&mut self, cnt: usize) {
             (**self).advance(cnt)
         }
 
+        #[inline]
         fn has_remaining(&self) -> bool {
             (**self).has_remaining()
         }
 
+        #[inline]
         fn copy_to_slice(&mut self, dst: &mut [u8]) {
             (**self).copy_to_slice(dst)
         }
 
+        #[inline]
         fn get_u8(&mut self) -> u8 {
             (**self).get_u8()
         }
 
+        #[inline]
         fn get_i8(&mut self) -> i8 {
             (**self).get_i8()
         }
 
+        #[inline]
         fn get_u16(&mut self) -> u16 {
             (**self).get_u16()
         }
 
+        #[inline]
         fn get_u16_le(&mut self) -> u16 {
             (**self).get_u16_le()
         }
 
+        #[inline]
+        fn get_u16_ne(&mut self) -> u16 {
+            (**self).get_u16_ne()
+        }
+
+        #[inline]
         fn get_i16(&mut self) -> i16 {
             (**self).get_i16()
         }
 
+        #[inline]
         fn get_i16_le(&mut self) -> i16 {
             (**self).get_i16_le()
         }
 
+        #[inline]
+        fn get_i16_ne(&mut self) -> i16 {
+            (**self).get_i16_ne()
+        }
+
+        #[inline]
         fn get_u32(&mut self) -> u32 {
             (**self).get_u32()
         }
 
+        #[inline]
         fn get_u32_le(&mut self) -> u32 {
             (**self).get_u32_le()
         }
 
+        #[inline]
+        fn get_u32_ne(&mut self) -> u32 {
+            (**self).get_u32_ne()
+        }
+
+        #[inline]
         fn get_i32(&mut self) -> i32 {
             (**self).get_i32()
         }
 
+        #[inline]
         fn get_i32_le(&mut self) -> i32 {
             (**self).get_i32_le()
         }
 
+        #[inline]
+        fn get_i32_ne(&mut self) -> i32 {
+            (**self).get_i32_ne()
+        }
+
+        #[inline]
         fn get_u64(&mut self) -> u64 {
             (**self).get_u64()
         }
 
+        #[inline]
         fn get_u64_le(&mut self) -> u64 {
             (**self).get_u64_le()
         }
 
+        #[inline]
+        fn get_u64_ne(&mut self) -> u64 {
+            (**self).get_u64_ne()
+        }
+
+        #[inline]
         fn get_i64(&mut self) -> i64 {
             (**self).get_i64()
         }
 
+        #[inline]
         fn get_i64_le(&mut self) -> i64 {
             (**self).get_i64_le()
         }
 
+        #[inline]
+        fn get_i64_ne(&mut self) -> i64 {
+            (**self).get_i64_ne()
+        }
+
+        #[inline]
         fn get_uint(&mut self, nbytes: usize) -> u64 {
             (**self).get_uint(nbytes)
         }
 
+        #[inline]
         fn get_uint_le(&mut self, nbytes: usize) -> u64 {
             (**self).get_uint_le(nbytes)
         }
 
+        #[inline]
+        fn get_uint_ne(&mut self, nbytes: usize) -> u64 {
+            (**self).get_uint_ne(nbytes)
+        }
+
+        #[inline]
         fn get_int(&mut self, nbytes: usize) -> i64 {
             (**self).get_int(nbytes)
         }
 
+        #[inline]
         fn get_int_le(&mut self, nbytes: usize) -> i64 {
             (**self).get_int_le(nbytes)
         }
 
+        #[inline]
+        fn get_int_ne(&mut self, nbytes: usize) -> i64 {
+            (**self).get_int_ne(nbytes)
+        }
+
+        #[inline]
         fn copy_to_bytes(&mut self, len: usize) -> crate::Bytes {
             (**self).copy_to_bytes(len)
         }
@@ -1022,76 +1412,58 @@ impl Buf for &[u8] {
     }
 
     #[inline]
-    fn bytes(&self) -> &[u8] {
+    fn chunk(&self) -> &[u8] {
         self
     }
 
     #[inline]
     fn advance(&mut self, cnt: usize) {
+        if self.len() < cnt {
+            panic_advance(cnt, self.len());
+        }
+
         *self = &self[cnt..];
     }
-}
 
-impl Buf for Option<[u8; 1]> {
-    fn remaining(&self) -> usize {
-        if self.is_some() {
-            1
-        } else {
-            0
-        }
-    }
-
-    fn bytes(&self) -> &[u8] {
-        self.as_ref()
-            .map(AsRef::as_ref)
-            .unwrap_or(Default::default())
-    }
-
-    fn advance(&mut self, cnt: usize) {
-        if cnt == 0 {
-            return;
+    #[inline]
+    fn copy_to_slice(&mut self, dst: &mut [u8]) {
+        if self.len() < dst.len() {
+            panic_advance(dst.len(), self.len());
         }
 
-        if self.is_none() {
-            panic!("overflow");
-        } else {
-            assert_eq!(1, cnt);
-            *self = None;
-        }
+        dst.copy_from_slice(&self[..dst.len()]);
+        self.advance(dst.len());
     }
 }
 
 #[cfg(feature = "std")]
 impl<T: AsRef<[u8]>> Buf for std::io::Cursor<T> {
+    #[inline]
     fn remaining(&self) -> usize {
-        let len = self.get_ref().as_ref().len();
-        let pos = self.position();
-
-        if pos >= len as u64 {
-            return 0;
-        }
-
-        len - pos as usize
+        saturating_sub_usize_u64(self.get_ref().as_ref().len(), self.position())
     }
 
-    fn bytes(&self) -> &[u8] {
-        let len = self.get_ref().as_ref().len();
-        let pos = self.position();
-
-        if pos >= len as u64 {
-            return &[];
-        }
-
-        &self.get_ref().as_ref()[pos as usize..]
+    #[inline]
+    fn chunk(&self) -> &[u8] {
+        let slice = self.get_ref().as_ref();
+        let pos = min_u64_usize(self.position(), slice.len());
+        &slice[pos..]
     }
 
+    #[inline]
     fn advance(&mut self, cnt: usize) {
-        let pos = (self.position() as usize)
-            .checked_add(cnt)
-            .expect("overflow");
+        let len = self.get_ref().as_ref().len();
+        let pos = self.position();
 
-        assert!(pos <= self.get_ref().as_ref().len());
-        self.set_position(pos as u64);
+        // We intentionally allow `cnt == 0` here even if `pos > len`.
+        let max_cnt = saturating_sub_usize_u64(len, pos);
+        if cnt > max_cnt {
+            panic_advance(cnt, max_cnt);
+        }
+
+        // This will not overflow because either `cnt == 0` or the sum is not
+        // greater than `len`.
+        self.set_position(pos + cnt as u64);
     }
 }
 

src/buf/chain.rs

@@ -1,5 +1,5 @@
 use crate::buf::{IntoIter, UninitSlice};
-use crate::{Buf, BufMut};
+use crate::{Buf, BufMut, Bytes};
 
 #[cfg(feature = "std")]
 use std::io::IoSlice;
@@ -25,9 +25,7 @@ use std::io::IoSlice;
 /// assert_eq!(full[..], b"hello world"[..]);
 /// ```
 ///
-/// [`Buf::chain`]: trait.Buf.html#method.chain
-/// [`Buf`]: trait.Buf.html
-/// [`BufMut`]: trait.BufMut.html
+/// [`Buf::chain`]: Buf::chain
 #[derive(Debug)]
 pub struct Chain<T, U> {
     a: T,
@@ -135,14 +133,14 @@ where
     U: Buf,
 {
     fn remaining(&self) -> usize {
-        self.a.remaining() + self.b.remaining()
+        self.a.remaining().saturating_add(self.b.remaining())
     }
 
-    fn bytes(&self) -> &[u8] {
+    fn chunk(&self) -> &[u8] {
         if self.a.has_remaining() {
-            self.a.bytes()
+            self.a.chunk()
         } else {
-            self.b.bytes()
+            self.b.chunk()
         }
     }
 
@@ -165,11 +163,29 @@ where
     }
 
     #[cfg(feature = "std")]
-    fn bytes_vectored<'a>(&'a self, dst: &mut [IoSlice<'a>]) -> usize {
-        let mut n = self.a.bytes_vectored(dst);
-        n += self.b.bytes_vectored(&mut dst[n..]);
+    fn chunks_vectored<'a>(&'a self, dst: &mut [IoSlice<'a>]) -> usize {
+        let mut n = self.a.chunks_vectored(dst);
+        n += self.b.chunks_vectored(&mut dst[n..]);
         n
     }
+
+    fn copy_to_bytes(&mut self, len: usize) -> Bytes {
+        let a_rem = self.a.remaining();
+        if a_rem >= len {
+            self.a.copy_to_bytes(len)
+        } else if a_rem == 0 {
+            self.b.copy_to_bytes(len)
+        } else {
+            assert!(
+                len - a_rem <= self.b.remaining(),
+                "`len` greater than remaining"
+            );
+            let mut ret = crate::BytesMut::with_capacity(len);
+            ret.put(&mut self.a);
+            ret.put((&mut self.b).take(len - a_rem));
+            ret.freeze()
+        }
+    }
 }
 
 unsafe impl<T, U> BufMut for Chain<T, U>
@@ -178,14 +194,16 @@ where
     U: BufMut,
 {
     fn remaining_mut(&self) -> usize {
-        self.a.remaining_mut() + self.b.remaining_mut()
+        self.a
+            .remaining_mut()
+            .saturating_add(self.b.remaining_mut())
     }
 
-    fn bytes_mut(&mut self) -> &mut UninitSlice {
+    fn chunk_mut(&mut self) -> &mut UninitSlice {
         if self.a.has_remaining_mut() {
-            self.a.bytes_mut()
+            self.a.chunk_mut()
         } else {
-            self.b.bytes_mut()
+            self.b.chunk_mut()
         }
     }
 

src/buf/iter.rs

@@ -2,8 +2,6 @@ use crate::Buf;
 
 /// Iterator over the bytes contained by the buffer.
 ///
-/// This struct is created by the [`iter`] method on [`Buf`].
-///
 /// # Examples
 ///
 /// Basic usage:
@@ -19,9 +17,6 @@ use crate::Buf;
 /// assert_eq!(iter.next(), Some(b'c'));
 /// assert_eq!(iter.next(), None);
 /// ```
-///
-/// [`iter`]: trait.Buf.html#method.iter
-/// [`Buf`]: trait.Buf.html
 #[derive(Debug)]
 pub struct IntoIter<T> {
     inner: T,
@@ -43,7 +38,7 @@ impl<T> IntoIter<T> {
     /// assert_eq!(iter.next(), Some(b'c'));
     /// assert_eq!(iter.next(), None);
     /// ```
-    pub(crate) fn new(inner: T) -> IntoIter<T> {
+    pub fn new(inner: T) -> IntoIter<T> {
         IntoIter { inner }
     }
 
@@ -117,7 +112,7 @@ impl<T: Buf> Iterator for IntoIter<T> {
             return None;
         }
 
-        let b = self.inner.bytes()[0];
+        let b = self.inner.chunk()[0];
         self.inner.advance(1);
 
         Some(b)

src/buf/limit.rs

@@ -61,8 +61,8 @@ unsafe impl<T: BufMut> BufMut for Limit<T> {
         cmp::min(self.inner.remaining_mut(), self.limit)
     }
 
-    fn bytes_mut(&mut self) -> &mut UninitSlice {
-        let bytes = self.inner.bytes_mut();
+    fn chunk_mut(&mut self) -> &mut UninitSlice {
+        let bytes = self.inner.chunk_mut();
         let end = cmp::min(bytes.len(), self.limit);
         &mut bytes[..end]
     }

src/buf/mod.rs

@@ -13,8 +13,6 @@
 //! See [`Buf`] and [`BufMut`] for more details.
 //!
 //! [rope]: https://en.wikipedia.org/wiki/Rope_(data_structure)
-//! [`Buf`]: trait.Buf.html
-//! [`BufMut`]: trait.BufMut.html
 
 mod buf_impl;
 mod buf_mut;

src/buf/reader.rs

@@ -5,7 +5,7 @@ use std::{cmp, io};
 /// A `Buf` adapter which implements `io::Read` for the inner value.
 ///
 /// This struct is generally created by calling `reader()` on `Buf`. See
-/// documentation of [`reader()`](trait.Buf.html#method.reader) for more
+/// documentation of [`reader()`](Buf::reader) for more
 /// details.
 #[derive(Debug)]
 pub struct Reader<B> {
@@ -73,7 +73,7 @@ impl<B: Buf + Sized> io::Read for Reader<B> {
 
 impl<B: Buf + Sized> io::BufRead for Reader<B> {
     fn fill_buf(&mut self) -> io::Result<&[u8]> {
-        Ok(self.buf.bytes())
+        Ok(self.buf.chunk())
     }
     fn consume(&mut self, amt: usize) {
         self.buf.advance(amt)

src/buf/take.rs

@@ -1,11 +1,11 @@
-use crate::Buf;
+use crate::{Buf, Bytes};
 
 use core::cmp;
 
 /// A `Buf` adapter which limits the bytes read from an underlying buffer.
 ///
 /// This struct is generally created by calling `take()` on `Buf`. See
-/// documentation of [`take()`](trait.BufExt.html#method.take) for more details.
+/// documentation of [`take()`](Buf::take) for more details.
 #[derive(Debug)]
 pub struct Take<T> {
     inner: T,
@@ -134,8 +134,8 @@ impl<T: Buf> Buf for Take<T> {
         cmp::min(self.inner.remaining(), self.limit)
     }
 
-    fn bytes(&self) -> &[u8] {
-        let bytes = self.inner.bytes();
+    fn chunk(&self) -> &[u8] {
+        let bytes = self.inner.chunk();
         &bytes[..cmp::min(bytes.len(), self.limit)]
     }
 
@@ -144,4 +144,12 @@ impl<T: Buf> Buf for Take<T> {
         self.inner.advance(cnt);
         self.limit -= cnt;
     }
+
+    fn copy_to_bytes(&mut self, len: usize) -> Bytes {
+        assert!(len <= self.remaining(), "`len` greater than remaining");
+
+        let r = self.inner.copy_to_bytes(len);
+        self.limit -= len;
+        r
+    }
 }

src/buf/uninit_slice.rs

@@ -6,7 +6,7 @@ use core::ops::{
 
 /// Uninitialized byte slice.
 ///
-/// Returned by `BufMut::bytes_mut()`, the referenced byte slice may be
+/// Returned by `BufMut::chunk_mut()`, the referenced byte slice may be
 /// uninitialized. The wrapper provides safe access without introducing
 /// undefined behavior.
 ///
@@ -22,6 +22,44 @@ use core::ops::{
 pub struct UninitSlice([MaybeUninit<u8>]);
 
 impl UninitSlice {
+    /// Creates a `&mut UninitSlice` wrapping a slice of initialised memory.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::buf::UninitSlice;
+    ///
+    /// let mut buffer = [0u8; 64];
+    /// let slice = UninitSlice::new(&mut buffer[..]);
+    /// ```
+    #[inline]
+    pub fn new(slice: &mut [u8]) -> &mut UninitSlice {
+        unsafe { &mut *(slice as *mut [u8] as *mut [MaybeUninit<u8>] as *mut UninitSlice) }
+    }
+
+    /// Creates a `&mut UninitSlice` wrapping a slice of uninitialised memory.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::buf::UninitSlice;
+    /// use core::mem::MaybeUninit;
+    ///
+    /// let mut buffer = [MaybeUninit::uninit(); 64];
+    /// let slice = UninitSlice::uninit(&mut buffer[..]);
+    ///
+    /// let mut vec = Vec::with_capacity(1024);
+    /// let spare: &mut UninitSlice = vec.spare_capacity_mut().into();
+    /// ```
+    #[inline]
+    pub fn uninit(slice: &mut [MaybeUninit<u8>]) -> &mut UninitSlice {
+        unsafe { &mut *(slice as *mut [MaybeUninit<u8>] as *mut UninitSlice) }
+    }
+
+    fn uninit_ref(slice: &[MaybeUninit<u8>]) -> &UninitSlice {
+        unsafe { &*(slice as *const [MaybeUninit<u8>] as *const UninitSlice) }
+    }
+
     /// Create a `&mut UninitSlice` from a pointer and a length.
     ///
     /// # Safety
@@ -40,10 +78,11 @@ impl UninitSlice {
     ///
     /// let slice = unsafe { UninitSlice::from_raw_parts_mut(ptr, len) };
     /// ```
+    #[inline]
     pub unsafe fn from_raw_parts_mut<'a>(ptr: *mut u8, len: usize) -> &'a mut UninitSlice {
         let maybe_init: &mut [MaybeUninit<u8>] =
             core::slice::from_raw_parts_mut(ptr as *mut _, len);
-        &mut *(maybe_init as *mut [MaybeUninit<u8>] as *mut UninitSlice)
+        Self::uninit(maybe_init)
     }
 
     /// Write a single byte at the specified offset.
@@ -64,13 +103,14 @@ impl UninitSlice {
     ///
     /// assert_eq!(b"boo", &data[..]);
     /// ```
+    #[inline]
     pub fn write_byte(&mut self, index: usize, byte: u8) {
         assert!(index < self.len());
 
         unsafe { self[index..].as_mut_ptr().write(byte) }
     }
 
-    /// Copies bytes  from `src` into `self`.
+    /// Copies bytes from `src` into `self`.
     ///
     /// The length of `src` must be the same as `self`.
     ///
@@ -90,6 +130,7 @@ impl UninitSlice {
     ///
     /// assert_eq!(b"bar", &data[..]);
     /// ```
+    #[inline]
     pub fn copy_from_slice(&mut self, src: &[u8]) {
         use core::ptr;
 
@@ -114,12 +155,39 @@ impl UninitSlice {
     ///
     /// let mut data = [0, 1, 2];
     /// let mut slice = &mut data[..];
-    /// let ptr = BufMut::bytes_mut(&mut slice).as_mut_ptr();
+    /// let ptr = BufMut::chunk_mut(&mut slice).as_mut_ptr();
     /// ```
+    #[inline]
     pub fn as_mut_ptr(&mut self) -> *mut u8 {
         self.0.as_mut_ptr() as *mut _
     }
 
+    /// Return a `&mut [MaybeUninit<u8>]` to this slice's buffer.
+    ///
+    /// # Safety
+    ///
+    /// The caller **must not** read from the referenced memory and **must not** write
+    /// **uninitialized** bytes to the slice either. This is because `BufMut` implementation
+    /// that created the `UninitSlice` knows which parts are initialized. Writing uninitialized
+    /// bytes to the slice may cause the `BufMut` to read those bytes and trigger undefined
+    /// behavior.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut data = [0, 1, 2];
+    /// let mut slice = &mut data[..];
+    /// unsafe {
+    ///     let uninit_slice = BufMut::chunk_mut(&mut slice).as_uninit_slice_mut();
+    /// };
+    /// ```
+    #[inline]
+    pub unsafe fn as_uninit_slice_mut(&mut self) -> &mut [MaybeUninit<u8>] {
+        &mut self.0
+    }
+
     /// Returns the number of bytes in the slice.
     ///
     /// # Examples
@@ -129,10 +197,11 @@ impl UninitSlice {
     ///
     /// let mut data = [0, 1, 2];
     /// let mut slice = &mut data[..];
-    /// let len = BufMut::bytes_mut(&mut slice).len();
+    /// let len = BufMut::chunk_mut(&mut slice).len();
     ///
     /// assert_eq!(len, 3);
     /// ```
+    #[inline]
     pub fn len(&self) -> usize {
         self.0.len()
     }
@@ -144,22 +213,34 @@ impl fmt::Debug for UninitSlice {
     }
 }
 
+impl<'a> From<&'a mut [u8]> for &'a mut UninitSlice {
+    fn from(slice: &'a mut [u8]) -> Self {
+        UninitSlice::new(slice)
+    }
+}
+
+impl<'a> From<&'a mut [MaybeUninit<u8>]> for &'a mut UninitSlice {
+    fn from(slice: &'a mut [MaybeUninit<u8>]) -> Self {
+        UninitSlice::uninit(slice)
+    }
+}
+
 macro_rules! impl_index {
     ($($t:ty),*) => {
         $(
             impl Index<$t> for UninitSlice {
                 type Output = UninitSlice;
 
+                #[inline]
                 fn index(&self, index: $t) -> &UninitSlice {
-                    let maybe_uninit: &[MaybeUninit<u8>] = &self.0[index];
-                    unsafe { &*(maybe_uninit as *const [MaybeUninit<u8>] as *const UninitSlice) }
+                    UninitSlice::uninit_ref(&self.0[index])
                 }
             }
 
             impl IndexMut<$t> for UninitSlice {
+                #[inline]
                 fn index_mut(&mut self, index: $t) -> &mut UninitSlice {
-                    let maybe_uninit: &mut [MaybeUninit<u8>] = &mut self.0[index];
-                    unsafe { &mut *(maybe_uninit as *mut [MaybeUninit<u8>] as *mut UninitSlice) }
+                    UninitSlice::uninit(&mut self.0[index])
                 }
             }
         )*

src/buf/vec_deque.rs

@@ -7,7 +7,7 @@ impl Buf for VecDeque<u8> {
         self.len()
     }
 
-    fn bytes(&self) -> &[u8] {
+    fn chunk(&self) -> &[u8] {
         let (s1, s2) = self.as_slices();
         if s1.is_empty() {
             s2

src/buf/writer.rs

@@ -5,7 +5,7 @@ use std::{cmp, io};
 /// A `BufMut` adapter which implements `io::Write` for the inner value.
 ///
 /// This struct is generally created by calling `writer()` on `BufMut`. See
-/// documentation of [`writer()`](trait.BufMut.html#method.writer) for more
+/// documentation of [`writer()`](BufMut::writer) for more
 /// details.
 #[derive(Debug)]
 pub struct Writer<B> {

src/bytes.rs

@@ -1,14 +1,21 @@
 use core::iter::FromIterator;
+use core::mem::{self, ManuallyDrop};
 use core::ops::{Deref, RangeBounds};
-use core::{cmp, fmt, hash, mem, ptr, slice, usize};
+use core::{cmp, fmt, hash, ptr, slice, usize};
 
-use alloc::{borrow::Borrow, boxed::Box, string::String, vec::Vec};
+use alloc::{
+    alloc::{dealloc, Layout},
+    borrow::Borrow,
+    boxed::Box,
+    string::String,
+    vec::Vec,
+};
 
 use crate::buf::IntoIter;
 #[allow(unused)]
 use crate::loom::sync::atomic::AtomicMut;
-use crate::loom::sync::atomic::{self, AtomicPtr, AtomicUsize, Ordering};
-use crate::Buf;
+use crate::loom::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
+use crate::{offset_from, Buf, BytesMut};
 
 /// A cheaply cloneable and sliceable chunk of contiguous memory.
 ///
@@ -26,7 +33,7 @@ use crate::Buf;
 /// All `Bytes` implementations must fulfill the following requirements:
 /// - They are cheaply cloneable and thereby shareable between an unlimited amount
 ///   of components, for example by modifying a reference count.
-/// - Instances can be sliced to refer to a subset of the the original buffer.
+/// - Instances can be sliced to refer to a subset of the original buffer.
 ///
 /// ```
 /// use bytes::Bytes;
@@ -55,17 +62,17 @@ use crate::Buf;
 /// # Sharing
 ///
 /// `Bytes` contains a vtable, which allows implementations of `Bytes` to define
-/// how sharing/cloneing is implemented in detail.
+/// how sharing/cloning is implemented in detail.
 /// When `Bytes::clone()` is called, `Bytes` will call the vtable function for
-/// cloning the backing storage in order to share it behind between multiple
-/// `Bytes` instances.
+/// cloning the backing storage in order to share it behind multiple `Bytes`
+/// instances.
 ///
 /// For `Bytes` implementations which refer to constant memory (e.g. created
 /// via `Bytes::from_static()`) the cloning implementation will be a no-op.
 ///
 /// For `Bytes` implementations which point to a reference counted shared storage
 /// (e.g. an `Arc<[u8]>`), sharing will be implemented by increasing the
-/// the reference count.
+/// reference count.
 ///
 /// Due to this mechanism, multiple `Bytes` instances may point to the same
 /// shared memory region.
@@ -78,18 +85,18 @@ use crate::Buf;
 ///
 /// ```text
 ///
-///    Arc ptrs                   +---------+
-///    ________________________ / | Bytes 2 |
-///   /                           +---------+
-///  /          +-----------+     |         |
-/// |_________/ |  Bytes 1  |     |         |
-/// |           +-----------+     |         |
+///    Arc ptrs                   ┌─────────┐
+///    ________________________ / │ Bytes 2 │
+///   /                           └─────────┘
+///  /          ┌───────────┐     |         |
+/// |_________/ │  Bytes 1  │     |         |
+/// |           └───────────┘     |         |
 /// |           |           | ___/ data     | tail
 /// |      data |      tail |/              |
 /// v           v           v               v
-/// +-----+---------------------------------+-----+
-/// | Arc |     |           |               |     |
-/// +-----+---------------------------------+-----+
+/// ┌─────┬─────┬───────────┬───────────────┬─────┐
+/// │ Arc │     │           │               │     │
+/// └─────┴─────┴───────────┴───────────────┴─────┘
 /// ```
 pub struct Bytes {
     ptr: *const u8,
@@ -103,6 +110,13 @@ pub(crate) struct Vtable {
     /// fn(data, ptr, len)
     pub clone: unsafe fn(&AtomicPtr<()>, *const u8, usize) -> Bytes,
     /// fn(data, ptr, len)
+    ///
+    /// takes `Bytes` to value
+    pub to_vec: unsafe fn(&AtomicPtr<()>, *const u8, usize) -> Vec<u8>,
+    pub to_mut: unsafe fn(&AtomicPtr<()>, *const u8, usize) -> BytesMut,
+    /// fn(data)
+    pub is_unique: unsafe fn(&AtomicPtr<()>) -> bool,
+    /// fn(data, ptr, len)
     pub drop: unsafe fn(&mut AtomicPtr<()>, *const u8, usize),
 }
 
@@ -121,15 +135,16 @@ impl Bytes {
     /// ```
     #[inline]
     #[cfg(not(all(loom, test)))]
-    pub const fn new() -> Bytes {
+    pub const fn new() -> Self {
         // Make it a named const to work around
         // "unsizing casts are not allowed in const fn"
         const EMPTY: &[u8] = &[];
         Bytes::from_static(EMPTY)
     }
 
+    /// Creates a new empty `Bytes`.
     #[cfg(all(loom, test))]
-    pub fn new() -> Bytes {
+    pub fn new() -> Self {
         const EMPTY: &[u8] = &[];
         Bytes::from_static(EMPTY)
     }
@@ -149,7 +164,7 @@ impl Bytes {
     /// ```
     #[inline]
     #[cfg(not(all(loom, test)))]
-    pub const fn from_static(bytes: &'static [u8]) -> Bytes {
+    pub const fn from_static(bytes: &'static [u8]) -> Self {
         Bytes {
             ptr: bytes.as_ptr(),
             len: bytes.len(),
@@ -158,8 +173,9 @@ impl Bytes {
         }
     }
 
+    /// Creates a new `Bytes` from a static slice.
     #[cfg(all(loom, test))]
-    pub fn from_static(bytes: &'static [u8]) -> Bytes {
+    pub fn from_static(bytes: &'static [u8]) -> Self {
         Bytes {
             ptr: bytes.as_ptr(),
             len: bytes.len(),
@@ -179,7 +195,7 @@ impl Bytes {
     /// assert_eq!(b.len(), 5);
     /// ```
     #[inline]
-    pub fn len(&self) -> usize {
+    pub const fn len(&self) -> usize {
         self.len
     }
 
@@ -194,10 +210,32 @@ impl Bytes {
     /// assert!(b.is_empty());
     /// ```
     #[inline]
-    pub fn is_empty(&self) -> bool {
+    pub const fn is_empty(&self) -> bool {
         self.len == 0
     }
 
+    /// Returns true if this is the only reference to the data.
+    ///
+    /// Always returns false if the data is backed by a static slice.
+    ///
+    /// The result of this method may be invalidated immediately if another
+    /// thread clones this value while this is being called. Ensure you have
+    /// unique access to this value (`&mut Bytes`) first if you need to be
+    /// certain the result is valid (i.e. for safety reasons)
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Bytes;
+    ///
+    /// let a = Bytes::from(vec![1, 2, 3]);
+    /// assert!(a.is_unique());
+    /// let b = a.clone();
+    /// assert!(!a.is_unique());
+    /// ```
+    pub fn is_unique(&self) -> bool {
+        unsafe { (self.vtable.is_unique)(&self.data) }
+    }
+
     /// Creates `Bytes` instance from slice, by copying it.
     pub fn copy_from_slice(data: &[u8]) -> Self {
         data.to_vec().into()
@@ -225,14 +263,14 @@ impl Bytes {
     ///
     /// Requires that `begin <= end` and `end <= self.len()`, otherwise slicing
     /// will panic.
-    pub fn slice(&self, range: impl RangeBounds<usize>) -> Bytes {
+    pub fn slice(&self, range: impl RangeBounds<usize>) -> Self {
         use core::ops::Bound;
 
         let len = self.len();
 
         let begin = match range.start_bound() {
             Bound::Included(&n) => n,
-            Bound::Excluded(&n) => n + 1,
+            Bound::Excluded(&n) => n.checked_add(1).expect("out of range"),
             Bound::Unbounded => 0,
         };
 
@@ -262,7 +300,7 @@ impl Bytes {
         let mut ret = self.clone();
 
         ret.len = end - begin;
-        ret.ptr = unsafe { ret.ptr.offset(begin as isize) };
+        ret.ptr = unsafe { ret.ptr.add(begin) };
 
         ret
     }
@@ -292,7 +330,7 @@ impl Bytes {
     ///
     /// Requires that the given `sub` slice is in fact contained within the
     /// `Bytes` buffer; otherwise this function will panic.
-    pub fn slice_ref(&self, subset: &[u8]) -> Bytes {
+    pub fn slice_ref(&self, subset: &[u8]) -> Self {
         // Empty slice and empty Bytes may have their pointers reset
         // so explicitly allow empty slice to be a subslice of any slice.
         if subset.is_empty() {
@@ -308,15 +346,15 @@ impl Bytes {
         assert!(
             sub_p >= bytes_p,
             "subset pointer ({:p}) is smaller than self pointer ({:p})",
-            sub_p as *const u8,
-            bytes_p as *const u8,
+            subset.as_ptr(),
+            self.as_ptr(),
         );
         assert!(
             sub_p + sub_len <= bytes_p + bytes_len,
             "subset is out of bounds: self = ({:p}, {}), subset = ({:p}, {})",
-            bytes_p as *const u8,
+            self.as_ptr(),
             bytes_len,
-            sub_p as *const u8,
+            subset.as_ptr(),
             sub_len,
         );
 
@@ -349,14 +387,7 @@ impl Bytes {
     ///
     /// Panics if `at > len`.
     #[must_use = "consider Bytes::truncate if you don't need the other half"]
-    pub fn split_off(&mut self, at: usize) -> Bytes {
-        assert!(
-            at <= self.len(),
-            "split_off out of bounds: {:?} <= {:?}",
-            at,
-            self.len(),
-        );
-
+    pub fn split_off(&mut self, at: usize) -> Self {
         if at == self.len() {
             return Bytes::new();
         }
@@ -365,6 +396,13 @@ impl Bytes {
             return mem::replace(self, Bytes::new());
         }
 
+        assert!(
+            at <= self.len(),
+            "split_off out of bounds: {:?} <= {:?}",
+            at,
+            self.len(),
+        );
+
         let mut ret = self.clone();
 
         self.len = at;
@@ -398,14 +436,7 @@ impl Bytes {
     ///
     /// Panics if `at > len`.
     #[must_use = "consider Bytes::advance if you don't need the other half"]
-    pub fn split_to(&mut self, at: usize) -> Bytes {
-        assert!(
-            at <= self.len(),
-            "split_to out of bounds: {:?} <= {:?}",
-            at,
-            self.len(),
-        );
-
+    pub fn split_to(&mut self, at: usize) -> Self {
         if at == self.len() {
             return mem::replace(self, Bytes::new());
         }
@@ -414,6 +445,13 @@ impl Bytes {
             return Bytes::new();
         }
 
+        assert!(
+            at <= self.len(),
+            "split_to out of bounds: {:?} <= {:?}",
+            at,
+            self.len(),
+        );
+
         let mut ret = self.clone();
 
         unsafe { self.inc_start(at) };
@@ -428,7 +466,7 @@ impl Bytes {
     /// If `len` is greater than the buffer's current length, this has no
     /// effect.
     ///
-    /// The [`split_off`] method can emulate `truncate`, but this causes the
+    /// The [split_off](`Self::split_off()`) method can emulate `truncate`, but this causes the
     /// excess bytes to be returned instead of dropped.
     ///
     /// # Examples
@@ -440,8 +478,6 @@ impl Bytes {
     /// buf.truncate(5);
     /// assert_eq!(buf, b"hello"[..]);
     /// ```
-    ///
-    /// [`split_off`]: #method.split_off
     #[inline]
     pub fn truncate(&mut self, len: usize) {
         if len < self.len {
@@ -474,6 +510,29 @@ impl Bytes {
         self.truncate(0);
     }
 
+    /// Try to convert self into `BytesMut`.
+    ///
+    /// If `self` is unique for the entire original buffer, this will succeed
+    /// and return a `BytesMut` with the contents of `self` without copying.
+    /// If `self` is not unique for the entire original buffer, this will fail
+    /// and return self.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::{Bytes, BytesMut};
+    ///
+    /// let bytes = Bytes::from(b"hello".to_vec());
+    /// assert_eq!(bytes.try_into_mut(), Ok(BytesMut::from(&b"hello"[..])));
+    /// ```
+    pub fn try_into_mut(self) -> Result<BytesMut, Bytes> {
+        if self.is_unique() {
+            Ok(self.into())
+        } else {
+            Err(self)
+        }
+    }
+
     #[inline]
     pub(crate) unsafe fn with_vtable(
         ptr: *const u8,
@@ -501,7 +560,7 @@ impl Bytes {
         // should already be asserted, but debug assert for tests
         debug_assert!(self.len >= by, "internal: inc_start out of bounds");
         self.len -= by;
-        self.ptr = self.ptr.offset(by as isize);
+        self.ptr = self.ptr.add(by);
     }
 }
 
@@ -530,7 +589,7 @@ impl Buf for Bytes {
     }
 
     #[inline]
-    fn bytes(&self) -> &[u8] {
+    fn chunk(&self) -> &[u8] {
         self.as_slice()
     }
 
@@ -548,14 +607,8 @@ impl Buf for Bytes {
         }
     }
 
-    fn copy_to_bytes(&mut self, len: usize) -> crate::Bytes {
-        if len == self.remaining() {
-            core::mem::replace(self, Bytes::new())
-        } else {
-            let ret = self.slice(..len);
-            self.advance(len);
-            ret
-        }
+    fn copy_to_bytes(&mut self, len: usize) -> Self {
+        self.split_to(len)
     }
 }
 
@@ -604,7 +657,7 @@ impl<'a> IntoIterator for &'a Bytes {
     type IntoIter = core::slice::Iter<'a, u8>;
 
     fn into_iter(self) -> Self::IntoIter {
-        self.as_slice().into_iter()
+        self.as_slice().iter()
     }
 }
 
@@ -686,7 +739,7 @@ impl PartialOrd<Bytes> for str {
 
 impl PartialEq<Vec<u8>> for Bytes {
     fn eq(&self, other: &Vec<u8>) -> bool {
-        *self == &other[..]
+        *self == other[..]
     }
 }
 
@@ -710,7 +763,7 @@ impl PartialOrd<Bytes> for Vec<u8> {
 
 impl PartialEq<String> for Bytes {
     fn eq(&self, other: &String) -> bool {
-        *self == &other[..]
+        *self == other[..]
     }
 }
 
@@ -797,43 +850,105 @@ impl From<&'static str> for Bytes {
 
 impl From<Vec<u8>> for Bytes {
     fn from(vec: Vec<u8>) -> Bytes {
-        // into_boxed_slice doesn't return a heap allocation for empty vectors,
+        let mut vec = ManuallyDrop::new(vec);
+        let ptr = vec.as_mut_ptr();
+        let len = vec.len();
+        let cap = vec.capacity();
+
+        // Avoid an extra allocation if possible.
+        if len == cap {
+            let vec = ManuallyDrop::into_inner(vec);
+            return Bytes::from(vec.into_boxed_slice());
+        }
+
+        let shared = Box::new(Shared {
+            buf: ptr,
+            cap,
+            ref_cnt: AtomicUsize::new(1),
+        });
+
+        let shared = Box::into_raw(shared);
+        // The pointer should be aligned, so this assert should
+        // always succeed.
+        debug_assert!(
+            0 == (shared as usize & KIND_MASK),
+            "internal: Box<Shared> should have an aligned pointer",
+        );
+        Bytes {
+            ptr,
+            len,
+            data: AtomicPtr::new(shared as _),
+            vtable: &SHARED_VTABLE,
+        }
+    }
+}
+
+impl From<Box<[u8]>> for Bytes {
+    fn from(slice: Box<[u8]>) -> Bytes {
+        // Box<[u8]> doesn't contain a heap allocation for empty slices,
         // so the pointer isn't aligned enough for the KIND_VEC stashing to
         // work.
-        if vec.is_empty() {
+        if slice.is_empty() {
             return Bytes::new();
         }
 
-        let slice = vec.into_boxed_slice();
         let len = slice.len();
-        let ptr = slice.as_ptr();
-        drop(Box::into_raw(slice));
+        let ptr = Box::into_raw(slice) as *mut u8;
 
         if ptr as usize & 0x1 == 0 {
-            let data = ptr as usize | KIND_VEC;
+            let data = ptr_map(ptr, |addr| addr | KIND_VEC);
             Bytes {
                 ptr,
                 len,
-                data: AtomicPtr::new(data as *mut _),
+                data: AtomicPtr::new(data.cast()),
                 vtable: &PROMOTABLE_EVEN_VTABLE,
             }
         } else {
             Bytes {
                 ptr,
                 len,
-                data: AtomicPtr::new(ptr as *mut _),
+                data: AtomicPtr::new(ptr.cast()),
                 vtable: &PROMOTABLE_ODD_VTABLE,
             }
         }
     }
 }
 
+impl From<Bytes> for BytesMut {
+    /// Convert self into `BytesMut`.
+    ///
+    /// If `bytes` is unique for the entire original buffer, this will return a
+    /// `BytesMut` with the contents of `bytes` without copying.
+    /// If `bytes` is not unique for the entire original buffer, this will make
+    /// a copy of `bytes` subset of the original buffer in a new `BytesMut`.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::{Bytes, BytesMut};
+    ///
+    /// let bytes = Bytes::from(b"hello".to_vec());
+    /// assert_eq!(BytesMut::from(bytes), BytesMut::from(&b"hello"[..]));
+    /// ```
+    fn from(bytes: Bytes) -> Self {
+        let bytes = ManuallyDrop::new(bytes);
+        unsafe { (bytes.vtable.to_mut)(&bytes.data, bytes.ptr, bytes.len) }
+    }
+}
+
 impl From<String> for Bytes {
     fn from(s: String) -> Bytes {
         Bytes::from(s.into_bytes())
     }
 }
 
+impl From<Bytes> for Vec<u8> {
+    fn from(bytes: Bytes) -> Vec<u8> {
+        let bytes = ManuallyDrop::new(bytes);
+        unsafe { (bytes.vtable.to_vec)(&bytes.data, bytes.ptr, bytes.len) }
+    }
+}
+
 // ===== impl Vtable =====
 
 impl fmt::Debug for Vtable {
@@ -849,6 +964,9 @@ impl fmt::Debug for Vtable {
 
 const STATIC_VTABLE: Vtable = Vtable {
     clone: static_clone,
+    to_vec: static_to_vec,
+    to_mut: static_to_mut,
+    is_unique: static_is_unique,
     drop: static_drop,
 };
 
@@ -857,6 +975,20 @@ unsafe fn static_clone(_: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Bytes {
     Bytes::from_static(slice)
 }
 
+unsafe fn static_to_vec(_: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    let slice = slice::from_raw_parts(ptr, len);
+    slice.to_vec()
+}
+
+unsafe fn static_to_mut(_: &AtomicPtr<()>, ptr: *const u8, len: usize) -> BytesMut {
+    let slice = slice::from_raw_parts(ptr, len);
+    BytesMut::from(slice)
+}
+
+fn static_is_unique(_: &AtomicPtr<()>) -> bool {
+    false
+}
+
 unsafe fn static_drop(_: &mut AtomicPtr<()>, _: *const u8, _: usize) {
     // nothing to drop for &'static [u8]
 }
@@ -865,11 +997,17 @@ unsafe fn static_drop(_: &mut AtomicPtr<()>, _: *const u8, _: usize) {
 
 static PROMOTABLE_EVEN_VTABLE: Vtable = Vtable {
     clone: promotable_even_clone,
+    to_vec: promotable_even_to_vec,
+    to_mut: promotable_even_to_mut,
+    is_unique: promotable_is_unique,
     drop: promotable_even_drop,
 };
 
 static PROMOTABLE_ODD_VTABLE: Vtable = Vtable {
     clone: promotable_odd_clone,
+    to_vec: promotable_odd_to_vec,
+    to_mut: promotable_odd_to_mut,
+    is_unique: promotable_is_unique,
     drop: promotable_odd_drop,
 };
 
@@ -878,25 +1016,92 @@ unsafe fn promotable_even_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize
     let kind = shared as usize & KIND_MASK;
 
     if kind == KIND_ARC {
-        shallow_clone_arc(shared as _, ptr, len)
+        shallow_clone_arc(shared.cast(), ptr, len)
     } else {
         debug_assert_eq!(kind, KIND_VEC);
-        let buf = (shared as usize & !KIND_MASK) as *mut u8;
+        let buf = ptr_map(shared.cast(), |addr| addr & !KIND_MASK);
         shallow_clone_vec(data, shared, buf, ptr, len)
     }
 }
 
+unsafe fn promotable_to_vec(
+    data: &AtomicPtr<()>,
+    ptr: *const u8,
+    len: usize,
+    f: fn(*mut ()) -> *mut u8,
+) -> Vec<u8> {
+    let shared = data.load(Ordering::Acquire);
+    let kind = shared as usize & KIND_MASK;
+
+    if kind == KIND_ARC {
+        shared_to_vec_impl(shared.cast(), ptr, len)
+    } else {
+        // If Bytes holds a Vec, then the offset must be 0.
+        debug_assert_eq!(kind, KIND_VEC);
+
+        let buf = f(shared);
+
+        let cap = offset_from(ptr, buf) + len;
+
+        // Copy back buffer
+        ptr::copy(ptr, buf, len);
+
+        Vec::from_raw_parts(buf, len, cap)
+    }
+}
+
+unsafe fn promotable_to_mut(
+    data: &AtomicPtr<()>,
+    ptr: *const u8,
+    len: usize,
+    f: fn(*mut ()) -> *mut u8,
+) -> BytesMut {
+    let shared = data.load(Ordering::Acquire);
+    let kind = shared as usize & KIND_MASK;
+
+    if kind == KIND_ARC {
+        shared_to_mut_impl(shared.cast(), ptr, len)
+    } else {
+        // KIND_VEC is a view of an underlying buffer at a certain offset.
+        // The ptr + len always represents the end of that buffer.
+        // Before truncating it, it is first promoted to KIND_ARC.
+        // Thus, we can safely reconstruct a Vec from it without leaking memory.
+        debug_assert_eq!(kind, KIND_VEC);
+
+        let buf = f(shared);
+        let off = offset_from(ptr, buf);
+        let cap = off + len;
+        let v = Vec::from_raw_parts(buf, cap, cap);
+
+        let mut b = BytesMut::from_vec(v);
+        b.advance_unchecked(off);
+        b
+    }
+}
+
+unsafe fn promotable_even_to_vec(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    promotable_to_vec(data, ptr, len, |shared| {
+        ptr_map(shared.cast(), |addr| addr & !KIND_MASK)
+    })
+}
+
+unsafe fn promotable_even_to_mut(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> BytesMut {
+    promotable_to_mut(data, ptr, len, |shared| {
+        ptr_map(shared.cast(), |addr| addr & !KIND_MASK)
+    })
+}
+
 unsafe fn promotable_even_drop(data: &mut AtomicPtr<()>, ptr: *const u8, len: usize) {
     data.with_mut(|shared| {
         let shared = *shared;
         let kind = shared as usize & KIND_MASK;
 
         if kind == KIND_ARC {
-            release_shared(shared as *mut Shared);
+            release_shared(shared.cast());
         } else {
             debug_assert_eq!(kind, KIND_VEC);
-            let buf = (shared as usize & !KIND_MASK) as *mut u8;
-            drop(rebuild_boxed_slice(buf, ptr, len));
+            let buf = ptr_map(shared.cast(), |addr| addr & !KIND_MASK);
+            free_boxed_slice(buf, ptr, len);
         }
     });
 }
@@ -909,38 +1114,65 @@ unsafe fn promotable_odd_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize)
         shallow_clone_arc(shared as _, ptr, len)
     } else {
         debug_assert_eq!(kind, KIND_VEC);
-        shallow_clone_vec(data, shared, shared as *mut u8, ptr, len)
+        shallow_clone_vec(data, shared, shared.cast(), ptr, len)
     }
 }
 
+unsafe fn promotable_odd_to_vec(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    promotable_to_vec(data, ptr, len, |shared| shared.cast())
+}
+
+unsafe fn promotable_odd_to_mut(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> BytesMut {
+    promotable_to_mut(data, ptr, len, |shared| shared.cast())
+}
+
 unsafe fn promotable_odd_drop(data: &mut AtomicPtr<()>, ptr: *const u8, len: usize) {
     data.with_mut(|shared| {
         let shared = *shared;
         let kind = shared as usize & KIND_MASK;
 
         if kind == KIND_ARC {
-            release_shared(shared as *mut Shared);
+            release_shared(shared.cast());
         } else {
             debug_assert_eq!(kind, KIND_VEC);
 
-            drop(rebuild_boxed_slice(shared as *mut u8, ptr, len));
+            free_boxed_slice(shared.cast(), ptr, len);
         }
     });
 }
 
-unsafe fn rebuild_boxed_slice(buf: *mut u8, offset: *const u8, len: usize) -> Box<[u8]> {
-    let cap = (offset as usize - buf as usize) + len;
-    Box::from_raw(slice::from_raw_parts_mut(buf, cap))
+unsafe fn promotable_is_unique(data: &AtomicPtr<()>) -> bool {
+    let shared = data.load(Ordering::Acquire);
+    let kind = shared as usize & KIND_MASK;
+
+    if kind == KIND_ARC {
+        let ref_cnt = (*shared.cast::<Shared>()).ref_cnt.load(Ordering::Relaxed);
+        ref_cnt == 1
+    } else {
+        true
+    }
+}
+
+unsafe fn free_boxed_slice(buf: *mut u8, offset: *const u8, len: usize) {
+    let cap = offset_from(offset, buf) + len;
+    dealloc(buf, Layout::from_size_align(cap, 1).unwrap())
 }
 
 // ===== impl SharedVtable =====
 
 struct Shared {
-    // holds vec for drop, but otherwise doesnt access it
-    _vec: Vec<u8>,
+    // Holds arguments to dealloc upon Drop, but otherwise doesn't use them
+    buf: *mut u8,
+    cap: usize,
     ref_cnt: AtomicUsize,
 }
 
+impl Drop for Shared {
+    fn drop(&mut self) {
+        unsafe { dealloc(self.buf, Layout::from_size_align(self.cap, 1).unwrap()) }
+    }
+}
+
 // Assert that the alignment of `Shared` is divisible by 2.
 // This is a necessary invariant since we depend on allocating `Shared` a
 // shared object to implicitly carry the `KIND_ARC` flag in its pointer.
@@ -949,6 +1181,9 @@ const _: [(); 0 - mem::align_of::<Shared>() % 2] = []; // Assert that the alignm
 
 static SHARED_VTABLE: Vtable = Vtable {
     clone: shared_clone,
+    to_vec: shared_to_vec,
+    to_mut: shared_to_mut,
+    is_unique: shared_is_unique,
     drop: shared_drop,
 };
 
@@ -961,9 +1196,87 @@ unsafe fn shared_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Byte
     shallow_clone_arc(shared as _, ptr, len)
 }
 
+unsafe fn shared_to_vec_impl(shared: *mut Shared, ptr: *const u8, len: usize) -> Vec<u8> {
+    // Check that the ref_cnt is 1 (unique).
+    //
+    // If it is unique, then it is set to 0 with AcqRel fence for the same
+    // reason in release_shared.
+    //
+    // Otherwise, we take the other branch and call release_shared.
+    if (*shared)
+        .ref_cnt
+        .compare_exchange(1, 0, Ordering::AcqRel, Ordering::Relaxed)
+        .is_ok()
+    {
+        // Deallocate the `Shared` instance without running its destructor.
+        let shared = *Box::from_raw(shared);
+        let shared = ManuallyDrop::new(shared);
+        let buf = shared.buf;
+        let cap = shared.cap;
+
+        // Copy back buffer
+        ptr::copy(ptr, buf, len);
+
+        Vec::from_raw_parts(buf, len, cap)
+    } else {
+        let v = slice::from_raw_parts(ptr, len).to_vec();
+        release_shared(shared);
+        v
+    }
+}
+
+unsafe fn shared_to_vec(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    shared_to_vec_impl(data.load(Ordering::Relaxed).cast(), ptr, len)
+}
+
+unsafe fn shared_to_mut_impl(shared: *mut Shared, ptr: *const u8, len: usize) -> BytesMut {
+    // The goal is to check if the current handle is the only handle
+    // that currently has access to the buffer. This is done by
+    // checking if the `ref_cnt` is currently 1.
+    //
+    // The `Acquire` ordering synchronizes with the `Release` as
+    // part of the `fetch_sub` in `release_shared`. The `fetch_sub`
+    // operation guarantees that any mutations done in other threads
+    // are ordered before the `ref_cnt` is decremented. As such,
+    // this `Acquire` will guarantee that those mutations are
+    // visible to the current thread.
+    //
+    // Otherwise, we take the other branch, copy the data and call `release_shared`.
+    if (*shared).ref_cnt.load(Ordering::Acquire) == 1 {
+        // Deallocate the `Shared` instance without running its destructor.
+        let shared = *Box::from_raw(shared);
+        let shared = ManuallyDrop::new(shared);
+        let buf = shared.buf;
+        let cap = shared.cap;
+
+        // Rebuild Vec
+        let off = offset_from(ptr, buf);
+        let v = Vec::from_raw_parts(buf, len + off, cap);
+
+        let mut b = BytesMut::from_vec(v);
+        b.advance_unchecked(off);
+        b
+    } else {
+        // Copy the data from Shared in a new Vec, then release it
+        let v = slice::from_raw_parts(ptr, len).to_vec();
+        release_shared(shared);
+        BytesMut::from_vec(v)
+    }
+}
+
+unsafe fn shared_to_mut(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> BytesMut {
+    shared_to_mut_impl(data.load(Ordering::Relaxed).cast(), ptr, len)
+}
+
+pub(crate) unsafe fn shared_is_unique(data: &AtomicPtr<()>) -> bool {
+    let shared = data.load(Ordering::Acquire);
+    let ref_cnt = (*shared.cast::<Shared>()).ref_cnt.load(Ordering::Relaxed);
+    ref_cnt == 1
+}
+
 unsafe fn shared_drop(data: &mut AtomicPtr<()>, _ptr: *const u8, _len: usize) {
     data.with_mut(|shared| {
-        release_shared(*shared as *mut Shared);
+        release_shared(shared.cast());
     });
 }
 
@@ -990,7 +1303,7 @@ unsafe fn shallow_clone_vec(
     offset: *const u8,
     len: usize,
 ) -> Bytes {
-    // If  the buffer is still tracked in a `Vec<u8>`. It is time to
+    // If the buffer is still tracked in a `Vec<u8>`. It is time to
     // promote the vec to an `Arc`. This could potentially be called
     // concurrently, so some care must be taken.
 
@@ -1001,9 +1314,9 @@ unsafe fn shallow_clone_vec(
     // updated and since the buffer hasn't been promoted to an
     // `Arc`, those three fields still are the components of the
     // vector.
-    let vec = rebuild_boxed_slice(buf, offset, len).into_vec();
     let shared = Box::new(Shared {
-        _vec: vec,
+        buf,
+        cap: offset_from(offset, buf) + len,
         // Initialize refcount to 2. One for this reference, and one
         // for the new clone that will be returned from
         // `shallow_clone`.
@@ -1023,33 +1336,35 @@ unsafe fn shallow_clone_vec(
     // `Release` is used synchronize with other threads that
     // will load the `arc` field.
     //
-    // If the `compare_and_swap` fails, then the thread lost the
+    // If the `compare_exchange` fails, then the thread lost the
     // race to promote the buffer to shared. The `Acquire`
-    // ordering will synchronize with the `compare_and_swap`
+    // ordering will synchronize with the `compare_exchange`
     // that happened in the other thread and the `Shared`
     // pointed to by `actual` will be visible.
-    let actual = atom.compare_and_swap(ptr as _, shared as _, Ordering::AcqRel);
+    match atom.compare_exchange(ptr as _, shared as _, Ordering::AcqRel, Ordering::Acquire) {
+        Ok(actual) => {
+            debug_assert!(actual as usize == ptr as usize);
+            // The upgrade was successful, the new handle can be
+            // returned.
+            Bytes {
+                ptr: offset,
+                len,
+                data: AtomicPtr::new(shared as _),
+                vtable: &SHARED_VTABLE,
+            }
+        }
+        Err(actual) => {
+            // The upgrade failed, a concurrent clone happened. Release
+            // the allocation that was made in this thread, it will not
+            // be needed.
+            let shared = Box::from_raw(shared);
+            mem::forget(*shared);
 
-    if actual as usize == ptr as usize {
-        // The upgrade was successful, the new handle can be
-        // returned.
-        return Bytes {
-            ptr: offset,
-            len,
-            data: AtomicPtr::new(shared as _),
-            vtable: &SHARED_VTABLE,
-        };
+            // Buffer already promoted to shared storage, so increment ref
+            // count.
+            shallow_clone_arc(actual as _, offset, len)
+        }
     }
-
-    // The upgrade failed, a concurrent clone happened. Release
-    // the allocation that was made in this thread, it will not
-    // be needed.
-    let shared = Box::from_raw(shared);
-    mem::forget(*shared);
-
-    // Buffer already promoted to shared storage, so increment ref
-    // count.
-    shallow_clone_arc(actual as _, offset, len)
 }
 
 unsafe fn release_shared(ptr: *mut Shared) {
@@ -1075,10 +1390,40 @@ unsafe fn release_shared(ptr: *mut Shared) {
     // > "acquire" operation before deleting the object.
     //
     // [1]: (www.boost.org/doc/libs/1_55_0/doc/html/atomic/usage_examples.html)
-    atomic::fence(Ordering::Acquire);
+    //
+    // Thread sanitizer does not support atomic fences. Use an atomic load
+    // instead.
+    (*ptr).ref_cnt.load(Ordering::Acquire);
 
     // Drop the data
-    Box::from_raw(ptr);
+    drop(Box::from_raw(ptr));
+}
+
+// Ideally we would always use this version of `ptr_map` since it is strict
+// provenance compatible, but it results in worse codegen. We will however still
+// use it on miri because it gives better diagnostics for people who test bytes
+// code with miri.
+//
+// See https://github.com/tokio-rs/bytes/pull/545 for more info.
+#[cfg(miri)]
+fn ptr_map<F>(ptr: *mut u8, f: F) -> *mut u8
+where
+    F: FnOnce(usize) -> usize,
+{
+    let old_addr = ptr as usize;
+    let new_addr = f(old_addr);
+    let diff = new_addr.wrapping_sub(old_addr);
+    ptr.wrapping_add(diff)
+}
+
+#[cfg(not(miri))]
+fn ptr_map<F>(ptr: *mut u8, f: F) -> *mut u8
+where
+    F: FnOnce(usize) -> usize,
+{
+    let old_addr = ptr as usize;
+    let new_addr = f(old_addr);
+    new_addr as *mut u8
 }
 
 // compile-fails

src/fmt/debug.rs

@@ -25,7 +25,7 @@ impl Debug for BytesRef<'_> {
             } else if b == b'\0' {
                 write!(f, "\\0")?;
             // ASCII printable
-            } else if b >= 0x20 && b < 0x7f {
+            } else if (0x20..0x7f).contains(&b) {
                 write!(f, "{}", b as char)?;
             } else {
                 write!(f, "\\x{:02x}", b)?;
@@ -38,12 +38,12 @@ impl Debug for BytesRef<'_> {
 
 impl Debug for Bytes {
     fn fmt(&self, f: &mut Formatter<'_>) -> Result {
-        Debug::fmt(&BytesRef(&self.as_ref()), f)
+        Debug::fmt(&BytesRef(self.as_ref()), f)
     }
 }
 
 impl Debug for BytesMut {
     fn fmt(&self, f: &mut Formatter<'_>) -> Result {
-        Debug::fmt(&BytesRef(&self.as_ref()), f)
+        Debug::fmt(&BytesRef(self.as_ref()), f)
     }
 }

src/lib.rs

@@ -1,20 +1,18 @@
+#![allow(unknown_lints, unexpected_cfgs)]
 #![warn(missing_docs, missing_debug_implementations, rust_2018_idioms)]
 #![doc(test(
     no_crate_inject,
     attr(deny(warnings, rust_2018_idioms), allow(dead_code, unused_variables))
 ))]
-#![doc(html_root_url = "https://docs.rs/bytes/0.6.0")]
 #![no_std]
+#![cfg_attr(docsrs, feature(doc_cfg))]
 
 //! Provides abstractions for working with bytes.
 //!
 //! The `bytes` crate provides an efficient byte buffer structure
-//! ([`Bytes`](struct.Bytes.html)) and traits for working with buffer
+//! ([`Bytes`]) and traits for working with buffer
 //! implementations ([`Buf`], [`BufMut`]).
 //!
-//! [`Buf`]: trait.Buf.html
-//! [`BufMut`]: trait.BufMut.html
-//!
 //! # `Bytes`
 //!
 //! `Bytes` is an efficient container for storing and operating on contiguous
@@ -52,9 +50,7 @@
 //! `a` and `b` will share the underlying buffer and maintain indices tracking
 //! the view into the buffer represented by the handle.
 //!
-//! See the [struct docs] for more details.
-//!
-//! [struct docs]: struct.Bytes.html
+//! See the [struct docs](`Bytes`) for more details.
 //!
 //! # `Buf`, `BufMut`
 //!
@@ -70,7 +66,7 @@
 //! ## Relation with `Read` and `Write`
 //!
 //! At first glance, it may seem that `Buf` and `BufMut` overlap in
-//! functionality with `std::io::Read` and `std::io::Write`. However, they
+//! functionality with [`std::io::Read`] and [`std::io::Write`]. However, they
 //! serve different purposes. A buffer is the value that is provided as an
 //! argument to `Read::read` and `Write::write`. `Read` and `Write` may then
 //! perform a syscall, which has the potential of failing. Operations on `Buf`
@@ -115,3 +111,55 @@ fn abort() -> ! {
         panic!("abort");
     }
 }
+
+#[inline(always)]
+#[cfg(feature = "std")]
+fn saturating_sub_usize_u64(a: usize, b: u64) -> usize {
+    use core::convert::TryFrom;
+    match usize::try_from(b) {
+        Ok(b) => a.saturating_sub(b),
+        Err(_) => 0,
+    }
+}
+
+#[inline(always)]
+#[cfg(feature = "std")]
+fn min_u64_usize(a: u64, b: usize) -> usize {
+    use core::convert::TryFrom;
+    match usize::try_from(a) {
+        Ok(a) => usize::min(a, b),
+        Err(_) => b,
+    }
+}
+
+/// Panic with a nice error message.
+#[cold]
+fn panic_advance(idx: usize, len: usize) -> ! {
+    panic!(
+        "advance out of bounds: the len is {} but advancing by {}",
+        len, idx
+    );
+}
+
+#[cold]
+fn panic_does_not_fit(size: usize, nbytes: usize) -> ! {
+    panic!(
+        "size too large: the integer type can fit {} bytes, but nbytes is {}",
+        size, nbytes
+    );
+}
+
+/// Precondition: dst >= original
+///
+/// The following line is equivalent to:
+///
+/// ```rust,ignore
+/// self.ptr.as_ptr().offset_from(ptr) as usize;
+/// ```
+///
+/// But due to min rust is 1.39 and it is only stabilized
+/// in 1.47, we cannot use it.
+#[inline]
+fn offset_from(dst: *const u8, original: *const u8) -> usize {
+    dst as usize - original as usize
+}

src/loom.rs

@@ -1,7 +1,7 @@
 #[cfg(not(all(test, loom)))]
 pub(crate) mod sync {
     pub(crate) mod atomic {
-        pub(crate) use core::sync::atomic::{fence, AtomicPtr, AtomicUsize, Ordering};
+        pub(crate) use core::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 
         pub(crate) trait AtomicMut<T> {
             fn with_mut<F, R>(&mut self, f: F) -> R
@@ -23,7 +23,7 @@ pub(crate) mod sync {
 #[cfg(all(test, loom))]
 pub(crate) mod sync {
     pub(crate) mod atomic {
-        pub(crate) use loom::sync::atomic::{fence, AtomicPtr, AtomicUsize, Ordering};
+        pub(crate) use loom::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 
         pub(crate) trait AtomicMut<T> {}
     }

tests/test_buf.rs

@@ -9,17 +9,17 @@ fn test_fresh_cursor_vec() {
     let mut buf = &b"hello"[..];
 
     assert_eq!(buf.remaining(), 5);
-    assert_eq!(buf.bytes(), b"hello");
+    assert_eq!(buf.chunk(), b"hello");
 
     buf.advance(2);
 
     assert_eq!(buf.remaining(), 3);
-    assert_eq!(buf.bytes(), b"llo");
+    assert_eq!(buf.chunk(), b"llo");
 
     buf.advance(3);
 
     assert_eq!(buf.remaining(), 0);
-    assert_eq!(buf.bytes(), b"");
+    assert_eq!(buf.chunk(), b"");
 }
 
 #[test]
@@ -36,6 +36,19 @@ fn test_get_u16() {
     assert_eq!(0x5421, buf.get_u16_le());
 }
 
+#[test]
+fn test_get_int() {
+    let mut buf = &b"\xd6zomg"[..];
+    assert_eq!(-42, buf.get_int(1));
+    let mut buf = &b"\xd6zomg"[..];
+    assert_eq!(-42, buf.get_int_le(1));
+
+    let mut buf = &b"\xfe\x1d\xc0zomg"[..];
+    assert_eq!(0xffffffffffc01dfeu64 as i64, buf.get_int_le(3));
+    let mut buf = &b"\xfe\x1d\xc0zomg"[..];
+    assert_eq!(0xfffffffffffe1dc0u64 as i64, buf.get_int(3));
+}
+
 #[test]
 #[should_panic]
 fn test_get_u16_buffer_underflow() {
@@ -53,7 +66,7 @@ fn test_bufs_vec() {
 
     let mut dst = [IoSlice::new(b1), IoSlice::new(b2)];
 
-    assert_eq!(1, buf.bytes_vectored(&mut dst[..]));
+    assert_eq!(1, buf.chunks_vectored(&mut dst[..]));
 }
 
 #[test]
@@ -63,15 +76,16 @@ fn test_vec_deque() {
     let mut buffer: VecDeque<u8> = VecDeque::new();
     buffer.extend(b"hello world");
     assert_eq!(11, buffer.remaining());
-    assert_eq!(b"hello world", buffer.bytes());
+    assert_eq!(b"hello world", buffer.chunk());
     buffer.advance(6);
-    assert_eq!(b"world", buffer.bytes());
+    assert_eq!(b"world", buffer.chunk());
     buffer.extend(b" piece");
     let mut out = [0; 11];
     buffer.copy_to_slice(&mut out);
     assert_eq!(b"world piece", &out[..]);
 }
 
+#[allow(unused_allocation)] // This is intentional.
 #[test]
 fn test_deref_buf_forwards() {
     struct Special;
@@ -81,8 +95,8 @@ fn test_deref_buf_forwards() {
             unreachable!("remaining");
         }
 
-        fn bytes(&self) -> &[u8] {
-            unreachable!("bytes");
+        fn chunk(&self) -> &[u8] {
+            unreachable!("chunk");
         }
 
         fn advance(&mut self, _: usize) {

tests/test_buf_mut.rs

@@ -3,21 +3,22 @@
 use bytes::buf::UninitSlice;
 use bytes::{BufMut, BytesMut};
 use core::fmt::Write;
+use core::mem::MaybeUninit;
 use core::usize;
 
 #[test]
 fn test_vec_as_mut_buf() {
     let mut buf = Vec::with_capacity(64);
 
-    assert_eq!(buf.remaining_mut(), usize::MAX);
+    assert_eq!(buf.remaining_mut(), isize::MAX as usize);
 
-    assert!(buf.bytes_mut().len() >= 64);
+    assert!(buf.chunk_mut().len() >= 64);
 
     buf.put(&b"zomg"[..]);
 
     assert_eq!(&buf, b"zomg");
 
-    assert_eq!(buf.remaining_mut(), usize::MAX - 4);
+    assert_eq!(buf.remaining_mut(), isize::MAX as usize - 4);
     assert_eq!(buf.capacity(), 64);
 
     for _ in 0..16 {
@@ -27,6 +28,14 @@ fn test_vec_as_mut_buf() {
     assert_eq!(buf.len(), 68);
 }
 
+#[test]
+fn test_vec_put_bytes() {
+    let mut buf = Vec::new();
+    buf.push(17);
+    buf.put_bytes(19, 2);
+    assert_eq!([17, 19, 19], &buf[..]);
+}
+
 #[test]
 fn test_put_u8() {
     let mut buf = Vec::with_capacity(8);
@@ -46,7 +55,35 @@ fn test_put_u16() {
 }
 
 #[test]
-#[should_panic(expected = "cannot advance")]
+fn test_put_int() {
+    let mut buf = Vec::with_capacity(8);
+    buf.put_int(0x1020304050607080, 3);
+    assert_eq!(b"\x60\x70\x80", &buf[..]);
+}
+
+#[test]
+#[should_panic]
+fn test_put_int_nbytes_overflow() {
+    let mut buf = Vec::with_capacity(8);
+    buf.put_int(0x1020304050607080, 9);
+}
+
+#[test]
+fn test_put_int_le() {
+    let mut buf = Vec::with_capacity(8);
+    buf.put_int_le(0x1020304050607080, 3);
+    assert_eq!(b"\x80\x70\x60", &buf[..]);
+}
+
+#[test]
+#[should_panic]
+fn test_put_int_le_nbytes_overflow() {
+    let mut buf = Vec::with_capacity(8);
+    buf.put_int_le(0x1020304050607080, 9);
+}
+
+#[test]
+#[should_panic(expected = "advance out of bounds: the len is 8 but advancing by 12")]
 fn test_vec_advance_mut() {
     // Verify fix for #354
     let mut buf = Vec::with_capacity(8);
@@ -65,13 +102,123 @@ fn test_clone() {
     assert!(buf != buf2);
 }
 
-#[test]
-fn test_mut_slice() {
-    let mut v = vec![0, 0, 0, 0];
-    let mut s = &mut v[..];
-    s.put_u32(42);
+fn do_test_slice_small<T: ?Sized>(make: impl Fn(&mut [u8]) -> &mut T)
+where
+    for<'r> &'r mut T: BufMut,
+{
+    let mut buf = [b'X'; 8];
+
+    let mut slice = make(&mut buf[..]);
+    slice.put_bytes(b'A', 2);
+    slice.put_u8(b'B');
+    slice.put_slice(b"BCC");
+    assert_eq!(2, slice.remaining_mut());
+    assert_eq!(b"AABBCCXX", &buf[..]);
+
+    let mut slice = make(&mut buf[..]);
+    slice.put_u32(0x61626364);
+    assert_eq!(4, slice.remaining_mut());
+    assert_eq!(b"abcdCCXX", &buf[..]);
+
+    let mut slice = make(&mut buf[..]);
+    slice.put_u32_le(0x30313233);
+    assert_eq!(4, slice.remaining_mut());
+    assert_eq!(b"3210CCXX", &buf[..]);
 }
 
+fn do_test_slice_large<T: ?Sized>(make: impl Fn(&mut [u8]) -> &mut T)
+where
+    for<'r> &'r mut T: BufMut,
+{
+    const LEN: usize = 100;
+    const FILL: [u8; LEN] = [b'Y'; LEN];
+
+    let test = |fill: &dyn Fn(&mut &mut T, usize)| {
+        for buf_len in 0..LEN {
+            let mut buf = [b'X'; LEN];
+            for fill_len in 0..=buf_len {
+                let mut slice = make(&mut buf[..buf_len]);
+                fill(&mut slice, fill_len);
+                assert_eq!(buf_len - fill_len, slice.remaining_mut());
+                let (head, tail) = buf.split_at(fill_len);
+                assert_eq!(&FILL[..fill_len], head);
+                assert!(tail.iter().all(|b| *b == b'X'));
+            }
+        }
+    };
+
+    test(&|slice, fill_len| slice.put_slice(&FILL[..fill_len]));
+    test(&|slice, fill_len| slice.put_bytes(FILL[0], fill_len));
+}
+
+fn do_test_slice_put_slice_panics<T: ?Sized>(make: impl Fn(&mut [u8]) -> &mut T)
+where
+    for<'r> &'r mut T: BufMut,
+{
+    let mut buf = [b'X'; 4];
+    let mut slice = make(&mut buf[..]);
+    slice.put_slice(b"12345");
+}
+
+fn do_test_slice_put_bytes_panics<T: ?Sized>(make: impl Fn(&mut [u8]) -> &mut T)
+where
+    for<'r> &'r mut T: BufMut,
+{
+    let mut buf = [b'X'; 4];
+    let mut slice = make(&mut buf[..]);
+    slice.put_bytes(b'1', 5);
+}
+
+#[test]
+fn test_slice_buf_mut_small() {
+    do_test_slice_small(|x| x);
+}
+
+#[test]
+fn test_slice_buf_mut_large() {
+    do_test_slice_large(|x| x);
+}
+
+#[test]
+#[should_panic]
+fn test_slice_buf_mut_put_slice_overflow() {
+    do_test_slice_put_slice_panics(|x| x);
+}
+
+#[test]
+#[should_panic]
+fn test_slice_buf_mut_put_bytes_overflow() {
+    do_test_slice_put_bytes_panics(|x| x);
+}
+
+fn make_maybe_uninit_slice(slice: &mut [u8]) -> &mut [MaybeUninit<u8>] {
+    // SAFETY: [u8] has the same layout as [MaybeUninit<u8>].
+    unsafe { core::mem::transmute(slice) }
+}
+
+#[test]
+fn test_maybe_uninit_buf_mut_small() {
+    do_test_slice_small(make_maybe_uninit_slice);
+}
+
+#[test]
+fn test_maybe_uninit_buf_mut_large() {
+    do_test_slice_large(make_maybe_uninit_slice);
+}
+
+#[test]
+#[should_panic]
+fn test_maybe_uninit_buf_mut_put_slice_overflow() {
+    do_test_slice_put_slice_panics(make_maybe_uninit_slice);
+}
+
+#[test]
+#[should_panic]
+fn test_maybe_uninit_buf_mut_put_bytes_overflow() {
+    do_test_slice_put_bytes_panics(make_maybe_uninit_slice);
+}
+
+#[allow(unused_allocation)] // This is intentional.
 #[test]
 fn test_deref_bufmut_forwards() {
     struct Special;
@@ -81,8 +228,8 @@ fn test_deref_bufmut_forwards() {
             unreachable!("remaining_mut");
         }
 
-        fn bytes_mut(&mut self) -> &mut UninitSlice {
-            unreachable!("bytes_mut");
+        fn chunk_mut(&mut self) -> &mut UninitSlice {
+            unreachable!("chunk_mut");
         }
 
         unsafe fn advance_mut(&mut self, _: usize) {

tests/test_bytes.rs

@@ -4,8 +4,8 @@ use bytes::{Buf, BufMut, Bytes, BytesMut};
 
 use std::usize;
 
-const LONG: &'static [u8] = b"mary had a little lamb, little lamb, little lamb";
-const SHORT: &'static [u8] = b"hello world";
+const LONG: &[u8] = b"mary had a little lamb, little lamb, little lamb";
+const SHORT: &[u8] = b"hello world";
 
 fn is_sync<T: Sync>() {}
 fn is_send<T: Send>() {}
@@ -411,8 +411,8 @@ fn freeze_after_split_off() {
 fn fns_defined_for_bytes_mut() {
     let mut bytes = BytesMut::from(&b"hello world"[..]);
 
-    bytes.as_ptr();
-    bytes.as_mut_ptr();
+    let _ = bytes.as_ptr();
+    let _ = bytes.as_mut_ptr();
 
     // Iterator
     let v: Vec<u8> = bytes.as_ref().iter().cloned().collect();
@@ -443,7 +443,7 @@ fn reserve_growth() {
     let _ = bytes.split();
 
     bytes.reserve(65);
-    assert_eq!(bytes.capacity(), 128);
+    assert_eq!(bytes.capacity(), 117);
 }
 
 #[test]
@@ -461,6 +461,7 @@ fn reserve_allocates_at_least_original_capacity() {
 }
 
 #[test]
+#[cfg_attr(miri, ignore)] // Miri is too slow
 fn reserve_max_original_capacity_value() {
     const SIZE: usize = 128 * 1024;
 
@@ -514,6 +515,34 @@ fn reserve_in_arc_unique_doubles() {
     assert_eq!(2000, bytes.capacity());
 }
 
+#[test]
+fn reserve_in_arc_unique_does_not_overallocate_after_split() {
+    let mut bytes = BytesMut::from(LONG);
+    let orig_capacity = bytes.capacity();
+    drop(bytes.split_off(LONG.len() / 2));
+
+    // now bytes is Arc and refcount == 1
+
+    let new_capacity = bytes.capacity();
+    bytes.reserve(orig_capacity - new_capacity);
+    assert_eq!(bytes.capacity(), orig_capacity);
+}
+
+#[test]
+fn reserve_in_arc_unique_does_not_overallocate_after_multiple_splits() {
+    let mut bytes = BytesMut::from(LONG);
+    let orig_capacity = bytes.capacity();
+    for _ in 0..10 {
+        drop(bytes.split_off(LONG.len() / 2));
+
+        // now bytes is Arc and refcount == 1
+
+        let new_capacity = bytes.capacity();
+        bytes.reserve(orig_capacity - new_capacity);
+    }
+    assert_eq!(bytes.capacity(), orig_capacity);
+}
+
 #[test]
 fn reserve_in_arc_nonunique_does_not_overallocate() {
     let mut bytes = BytesMut::with_capacity(1000);
@@ -526,6 +555,25 @@ fn reserve_in_arc_nonunique_does_not_overallocate() {
     assert_eq!(2001, bytes.capacity());
 }
 
+/// This function tests `BytesMut::reserve_inner`, where `BytesMut` holds
+/// a unique reference to the shared vector and decide to reuse it
+/// by reallocating the `Vec`.
+#[test]
+fn reserve_shared_reuse() {
+    let mut bytes = BytesMut::with_capacity(1000);
+    bytes.put_slice(b"Hello, World!");
+    drop(bytes.split());
+
+    bytes.put_slice(b"!123ex123,sadchELLO,_wORLD!");
+    // Use split_off so that v.capacity() - self.cap != off
+    drop(bytes.split_off(9));
+    assert_eq!(&*bytes, b"!123ex123");
+
+    bytes.reserve(2000);
+    assert_eq!(&*bytes, b"!123ex123");
+    assert_eq!(bytes.capacity(), 2009);
+}
+
 #[test]
 fn extend_mut() {
     let mut bytes = BytesMut::with_capacity(0);
@@ -543,6 +591,35 @@ fn extend_from_slice_mut() {
     }
 }
 
+#[test]
+fn extend_mut_from_bytes() {
+    let mut bytes = BytesMut::with_capacity(0);
+    bytes.extend([Bytes::from(LONG)]);
+    assert_eq!(*bytes, LONG[..]);
+}
+
+#[test]
+fn extend_past_lower_limit_of_size_hint() {
+    // See https://github.com/tokio-rs/bytes/pull/674#pullrequestreview-1913035700
+    struct Iter<I>(I);
+
+    impl<I: Iterator<Item = u8>> Iterator for Iter<I> {
+        type Item = u8;
+
+        fn next(&mut self) -> Option<Self::Item> {
+            self.0.next()
+        }
+
+        fn size_hint(&self) -> (usize, Option<usize>) {
+            (5, None)
+        }
+    }
+
+    let mut bytes = BytesMut::with_capacity(5);
+    bytes.extend(Iter(std::iter::repeat(0).take(10)));
+    assert_eq!(bytes.len(), 10);
+}
+
 #[test]
 fn extend_mut_without_size_hint() {
     let mut bytes = BytesMut::with_capacity(0);
@@ -599,6 +676,43 @@ fn advance_bytes_mut() {
     assert_eq!(a, b"d zomg wat wat"[..]);
 }
 
+// Ensures BytesMut::advance reduces always capacity
+//
+// See https://github.com/tokio-rs/bytes/issues/725
+#[test]
+fn advance_bytes_mut_remaining_capacity() {
+    // reduce the search space under miri
+    let max_capacity = if cfg!(miri) { 16 } else { 256 };
+    for capacity in 0..=max_capacity {
+        for len in 0..=capacity {
+            for advance in 0..=len {
+                eprintln!("testing capacity={capacity}, len={len}, advance={advance}");
+                let mut buf = BytesMut::with_capacity(capacity);
+
+                buf.resize(len, 42);
+                assert_eq!(buf.len(), len, "resize should write `len` bytes");
+                assert_eq!(
+                    buf.remaining(),
+                    len,
+                    "Buf::remaining() should equal BytesMut::len"
+                );
+
+                buf.advance(advance);
+                assert_eq!(
+                    buf.remaining(),
+                    len - advance,
+                    "Buf::advance should reduce the remaining len"
+                );
+                assert_eq!(
+                    buf.capacity(),
+                    capacity - advance,
+                    "Buf::advance should reduce the remaining capacity"
+                );
+            }
+        }
+    }
+}
+
 #[test]
 #[should_panic]
 fn advance_past_len() {
@@ -608,15 +722,15 @@ fn advance_past_len() {
 
 #[test]
 // Only run these tests on little endian systems. CI uses qemu for testing
-// little endian... and qemu doesn't really support threading all that well.
-#[cfg(target_endian = "little")]
+// big endian... and qemu doesn't really support threading all that well.
+#[cfg(any(miri, target_endian = "little"))]
 fn stress() {
     // Tests promoting a buffer from a vec -> shared in a concurrent situation
     use std::sync::{Arc, Barrier};
     use std::thread;
 
     const THREADS: usize = 8;
-    const ITERS: usize = 1_000;
+    const ITERS: usize = if cfg!(miri) { 100 } else { 1_000 };
 
     for i in 0..ITERS {
         let data = [i as u8; 256];
@@ -655,97 +769,6 @@ fn partial_eq_bytesmut() {
     assert!(bytesmut != bytes2);
 }
 
-/*
-#[test]
-fn bytes_unsplit_basic() {
-    let buf = Bytes::from(&b"aaabbbcccddd"[..]);
-
-    let splitted = buf.split_off(6);
-    assert_eq!(b"aaabbb", &buf[..]);
-    assert_eq!(b"cccddd", &splitted[..]);
-
-    buf.unsplit(splitted);
-    assert_eq!(b"aaabbbcccddd", &buf[..]);
-}
-
-#[test]
-fn bytes_unsplit_empty_other() {
-    let buf = Bytes::from(&b"aaabbbcccddd"[..]);
-
-    // empty other
-    let other = Bytes::new();
-
-    buf.unsplit(other);
-    assert_eq!(b"aaabbbcccddd", &buf[..]);
-}
-
-#[test]
-fn bytes_unsplit_empty_self() {
-    // empty self
-    let mut buf = Bytes::new();
-
-    let mut other = Bytes::with_capacity(64);
-    other.extend_from_slice(b"aaabbbcccddd");
-
-    buf.unsplit(other);
-    assert_eq!(b"aaabbbcccddd", &buf[..]);
-}
-
-#[test]
-fn bytes_unsplit_arc_different() {
-    let mut buf = Bytes::with_capacity(64);
-    buf.extend_from_slice(b"aaaabbbbeeee");
-
-    buf.split_off(8); //arc
-
-    let mut buf2 = Bytes::with_capacity(64);
-    buf2.extend_from_slice(b"ccccddddeeee");
-
-    buf2.split_off(8); //arc
-
-    buf.unsplit(buf2);
-    assert_eq!(b"aaaabbbbccccdddd", &buf[..]);
-}
-
-#[test]
-fn bytes_unsplit_arc_non_contiguous() {
-    let mut buf = Bytes::with_capacity(64);
-    buf.extend_from_slice(b"aaaabbbbeeeeccccdddd");
-
-    let mut buf2 = buf.split_off(8); //arc
-
-    let buf3 = buf2.split_off(4); //arc
-
-    buf.unsplit(buf3);
-    assert_eq!(b"aaaabbbbccccdddd", &buf[..]);
-}
-
-#[test]
-fn bytes_unsplit_two_split_offs() {
-    let mut buf = Bytes::with_capacity(64);
-    buf.extend_from_slice(b"aaaabbbbccccdddd");
-
-    let mut buf2 = buf.split_off(8); //arc
-    let buf3 = buf2.split_off(4); //arc
-
-    buf2.unsplit(buf3);
-    buf.unsplit(buf2);
-    assert_eq!(b"aaaabbbbccccdddd", &buf[..]);
-}
-
-#[test]
-fn bytes_unsplit_overlapping_references() {
-    let mut buf = Bytes::with_capacity(64);
-    buf.extend_from_slice(b"abcdefghijklmnopqrstuvwxyz");
-    let mut buf0010 = buf.slice(0..10);
-    let buf1020 = buf.slice(10..20);
-    let buf0515 = buf.slice(5..15);
-    buf0010.unsplit(buf1020);
-    assert_eq!(b"abcdefghijklmnopqrst", &buf0010[..]);
-    assert_eq!(b"fghijklmno", &buf0515[..]);
-}
-*/
-
 #[test]
 fn bytes_mut_unsplit_basic() {
     let mut buf = BytesMut::with_capacity(64);
@@ -783,6 +806,31 @@ fn bytes_mut_unsplit_empty_self() {
     assert_eq!(b"aaabbbcccddd", &buf[..]);
 }
 
+#[test]
+fn bytes_mut_unsplit_other_keeps_capacity() {
+    let mut buf = BytesMut::with_capacity(64);
+    buf.extend_from_slice(b"aabb");
+
+    // non empty other created "from" buf
+    let mut other = buf.split_off(buf.len());
+    other.extend_from_slice(b"ccddee");
+    buf.unsplit(other);
+
+    assert_eq!(buf.capacity(), 64);
+}
+
+#[test]
+fn bytes_mut_unsplit_empty_other_keeps_capacity() {
+    let mut buf = BytesMut::with_capacity(64);
+    buf.extend_from_slice(b"aabbccddee");
+
+    // empty other created "from" buf
+    let other = buf.split_off(buf.len());
+    buf.unsplit(other);
+
+    assert_eq!(buf.capacity(), 64);
+}
+
 #[test]
 fn bytes_mut_unsplit_arc_different() {
     let mut buf = BytesMut::with_capacity(64);
@@ -848,7 +896,7 @@ fn from_iter_no_size_hint() {
 
 fn test_slice_ref(bytes: &Bytes, start: usize, end: usize, expected: &[u8]) {
     let slice = &(bytes.as_ref()[start..end]);
-    let sub = bytes.slice_ref(&slice);
+    let sub = bytes.slice_ref(slice);
     assert_eq!(&sub[..], expected);
 }
 
@@ -868,7 +916,7 @@ fn slice_ref_empty() {
     let bytes = Bytes::from(&b""[..]);
     let slice = &(bytes.as_ref()[0..0]);
 
-    let sub = bytes.slice_ref(&slice);
+    let sub = bytes.slice_ref(slice);
     assert_eq!(&sub[..], b"");
 }
 
@@ -912,20 +960,20 @@ fn bytes_buf_mut_advance() {
     let mut bytes = BytesMut::with_capacity(1024);
 
     unsafe {
-        let ptr = bytes.bytes_mut().as_mut_ptr();
-        assert_eq!(1024, bytes.bytes_mut().len());
+        let ptr = bytes.chunk_mut().as_mut_ptr();
+        assert_eq!(1024, bytes.chunk_mut().len());
 
         bytes.advance_mut(10);
 
-        let next = bytes.bytes_mut().as_mut_ptr();
-        assert_eq!(1024 - 10, bytes.bytes_mut().len());
+        let next = bytes.chunk_mut().as_mut_ptr();
+        assert_eq!(1024 - 10, bytes.chunk_mut().len());
         assert_eq!(ptr.offset(10), next);
 
         // advance to the end
         bytes.advance_mut(1024 - 10);
 
         // The buffer size is doubled
-        assert_eq!(1024, bytes.bytes_mut().len());
+        assert_eq!(1024, bytes.chunk_mut().len());
     }
 }
 
@@ -960,3 +1008,394 @@ fn bytes_with_capacity_but_empty() {
     let vec = Vec::with_capacity(1);
     let _ = Bytes::from(vec);
 }
+
+#[test]
+fn bytes_put_bytes() {
+    let mut bytes = BytesMut::new();
+    bytes.put_u8(17);
+    bytes.put_bytes(19, 2);
+    assert_eq!([17, 19, 19], bytes.as_ref());
+}
+
+#[test]
+fn box_slice_empty() {
+    // See https://github.com/tokio-rs/bytes/issues/340
+    let empty: Box<[u8]> = Default::default();
+    let b = Bytes::from(empty);
+    assert!(b.is_empty());
+}
+
+#[test]
+fn bytes_into_vec() {
+    // Test kind == KIND_VEC
+    let content = b"helloworld";
+
+    let mut bytes = BytesMut::new();
+    bytes.put_slice(content);
+
+    let vec: Vec<u8> = bytes.into();
+    assert_eq!(&vec, content);
+
+    // Test kind == KIND_ARC, shared.is_unique() == True
+    let mut bytes = BytesMut::new();
+    bytes.put_slice(b"abcdewe23");
+    bytes.put_slice(content);
+
+    // Overwrite the bytes to make sure only one reference to the underlying
+    // Vec exists.
+    bytes = bytes.split_off(9);
+
+    let vec: Vec<u8> = bytes.into();
+    assert_eq!(&vec, content);
+
+    // Test kind == KIND_ARC, shared.is_unique() == False
+    let prefix = b"abcdewe23";
+
+    let mut bytes = BytesMut::new();
+    bytes.put_slice(prefix);
+    bytes.put_slice(content);
+
+    let vec: Vec<u8> = bytes.split_off(prefix.len()).into();
+    assert_eq!(&vec, content);
+
+    let vec: Vec<u8> = bytes.into();
+    assert_eq!(&vec, prefix);
+}
+
+#[test]
+fn test_bytes_into_vec() {
+    // Test STATIC_VTABLE.to_vec
+    let bs = b"1b23exfcz3r";
+    let vec: Vec<u8> = Bytes::from_static(bs).into();
+    assert_eq!(&*vec, bs);
+
+    // Test bytes_mut.SHARED_VTABLE.to_vec impl
+    eprintln!("1");
+    let mut bytes_mut: BytesMut = bs[..].into();
+
+    // Set kind to KIND_ARC so that after freeze, Bytes will use bytes_mut.SHARED_VTABLE
+    eprintln!("2");
+    drop(bytes_mut.split_off(bs.len()));
+
+    eprintln!("3");
+    let b1 = bytes_mut.freeze();
+    eprintln!("4");
+    let b2 = b1.clone();
+
+    eprintln!("{:#?}", (&*b1).as_ptr());
+
+    // shared.is_unique() = False
+    eprintln!("5");
+    assert_eq!(&*Vec::from(b2), bs);
+
+    // shared.is_unique() = True
+    eprintln!("6");
+    assert_eq!(&*Vec::from(b1), bs);
+
+    // Test bytes_mut.SHARED_VTABLE.to_vec impl where offset != 0
+    let mut bytes_mut1: BytesMut = bs[..].into();
+    let bytes_mut2 = bytes_mut1.split_off(9);
+
+    let b1 = bytes_mut1.freeze();
+    let b2 = bytes_mut2.freeze();
+
+    assert_eq!(Vec::from(b2), bs[9..]);
+    assert_eq!(Vec::from(b1), bs[..9]);
+}
+
+#[test]
+fn test_bytes_into_vec_promotable_even() {
+    let vec = vec![33u8; 1024];
+
+    // Test cases where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    assert_eq!(Vec::from(b2), vec);
+
+    // Test cases where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+
+    assert_eq!(Vec::from(b2), vec[20..]);
+    assert_eq!(Vec::from(b1), vec[..20]);
+}
+
+#[test]
+fn test_bytes_vec_conversion() {
+    let mut vec = Vec::with_capacity(10);
+    vec.extend(b"abcdefg");
+    let b = Bytes::from(vec);
+    let v = Vec::from(b);
+    assert_eq!(v.len(), 7);
+    assert_eq!(v.capacity(), 10);
+
+    let mut b = Bytes::from(v);
+    b.advance(1);
+    let v = Vec::from(b);
+    assert_eq!(v.len(), 6);
+    assert_eq!(v.capacity(), 10);
+    assert_eq!(v.as_slice(), b"bcdefg");
+}
+
+#[test]
+fn test_bytes_mut_conversion() {
+    let mut b1 = BytesMut::with_capacity(10);
+    b1.extend(b"abcdefg");
+    let b2 = Bytes::from(b1);
+    let v = Vec::from(b2);
+    assert_eq!(v.len(), 7);
+    assert_eq!(v.capacity(), 10);
+
+    let mut b = Bytes::from(v);
+    b.advance(1);
+    let v = Vec::from(b);
+    assert_eq!(v.len(), 6);
+    assert_eq!(v.capacity(), 10);
+    assert_eq!(v.as_slice(), b"bcdefg");
+}
+
+#[test]
+fn test_bytes_capacity_len() {
+    for cap in 0..100 {
+        for len in 0..=cap {
+            let mut v = Vec::with_capacity(cap);
+            v.resize(len, 0);
+            let _ = Bytes::from(v);
+        }
+    }
+}
+
+#[test]
+fn static_is_unique() {
+    let b = Bytes::from_static(LONG);
+    assert!(!b.is_unique());
+}
+
+#[test]
+fn vec_is_unique() {
+    let v: Vec<u8> = LONG.to_vec();
+    let b = Bytes::from(v);
+    assert!(b.is_unique());
+}
+
+#[test]
+fn arc_is_unique() {
+    let v: Vec<u8> = LONG.to_vec();
+    let b = Bytes::from(v);
+    let c = b.clone();
+    assert!(!b.is_unique());
+    drop(c);
+    assert!(b.is_unique());
+}
+
+#[test]
+fn shared_is_unique() {
+    let v: Vec<u8> = LONG.to_vec();
+    let b = Bytes::from(v);
+    let c = b.clone();
+    assert!(!c.is_unique());
+    drop(b);
+    assert!(c.is_unique());
+}
+
+#[test]
+fn mut_shared_is_unique() {
+    let mut b = BytesMut::from(LONG);
+    let c = b.split().freeze();
+    assert!(!c.is_unique());
+    drop(b);
+    assert!(c.is_unique());
+}
+
+#[test]
+fn test_bytesmut_from_bytes_static() {
+    let bs = b"1b23exfcz3r";
+
+    // Test STATIC_VTABLE.to_mut
+    let bytes_mut = BytesMut::from(Bytes::from_static(bs));
+    assert_eq!(bytes_mut, bs[..]);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_bytes_mut_vec() {
+    let bs = b"1b23exfcz3r";
+    let bs_long = b"1b23exfcz3r1b23exfcz3r";
+
+    // Test case where kind == KIND_VEC
+    let mut bytes_mut: BytesMut = bs[..].into();
+    bytes_mut = BytesMut::from(bytes_mut.freeze());
+    assert_eq!(bytes_mut, bs[..]);
+    bytes_mut.extend_from_slice(&bs[..]);
+    assert_eq!(bytes_mut, bs_long[..]);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_bytes_mut_shared() {
+    let bs = b"1b23exfcz3r";
+
+    // Set kind to KIND_ARC so that after freeze, Bytes will use bytes_mut.SHARED_VTABLE
+    let mut bytes_mut: BytesMut = bs[..].into();
+    drop(bytes_mut.split_off(bs.len()));
+
+    let b1 = bytes_mut.freeze();
+    let b2 = b1.clone();
+
+    // shared.is_unique() = False
+    let mut b1m = BytesMut::from(b1);
+    assert_eq!(b1m, bs[..]);
+    b1m[0] = b'9';
+
+    // shared.is_unique() = True
+    let b2m = BytesMut::from(b2);
+    assert_eq!(b2m, bs[..]);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_bytes_mut_offset() {
+    let bs = b"1b23exfcz3r";
+
+    // Test bytes_mut.SHARED_VTABLE.to_mut impl where offset != 0
+    let mut bytes_mut1: BytesMut = bs[..].into();
+    let bytes_mut2 = bytes_mut1.split_off(9);
+
+    let b1 = bytes_mut1.freeze();
+    let b2 = bytes_mut2.freeze();
+
+    let b1m = BytesMut::from(b1);
+    let b2m = BytesMut::from(b2);
+
+    assert_eq!(b2m, bs[9..]);
+    assert_eq!(b1m, bs[..9]);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_promotable_even_vec() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    let b1m = BytesMut::from(b1);
+    assert_eq!(b1m, vec);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_promotable_even_arc_1() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    let b1m = BytesMut::from(b1);
+    assert_eq!(b1m, vec);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_promotable_even_arc_2() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    let b1m = BytesMut::from(b1);
+    assert_eq!(b1m, vec);
+
+    // Test case where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    let b2m = BytesMut::from(b2);
+    assert_eq!(b2m, vec);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_promotable_even_arc_offset() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+    let b1m = BytesMut::from(b1);
+    let b2m = BytesMut::from(b2);
+
+    assert_eq!(b2m, vec[20..]);
+    assert_eq!(b1m, vec[..20]);
+}
+
+#[test]
+fn try_reclaim_empty() {
+    let mut buf = BytesMut::new();
+    assert_eq!(false, buf.try_reclaim(6));
+    buf.reserve(6);
+    assert_eq!(true, buf.try_reclaim(6));
+    let cap = buf.capacity();
+    assert!(cap >= 6);
+    assert_eq!(false, buf.try_reclaim(cap + 1));
+
+    let mut buf = BytesMut::new();
+    buf.reserve(6);
+    let cap = buf.capacity();
+    assert!(cap >= 6);
+    let mut split = buf.split();
+    drop(buf);
+    assert_eq!(0, split.capacity());
+    assert_eq!(true, split.try_reclaim(6));
+    assert_eq!(false, split.try_reclaim(cap + 1));
+}
+
+#[test]
+fn try_reclaim_vec() {
+    let mut buf = BytesMut::with_capacity(6);
+    buf.put_slice(b"abc");
+    // Reclaiming a ludicrous amount of space should calmly return false
+    assert_eq!(false, buf.try_reclaim(usize::MAX));
+
+    assert_eq!(false, buf.try_reclaim(6));
+    buf.advance(2);
+    assert_eq!(4, buf.capacity());
+    // We can reclaim 5 bytes, because the byte in the buffer can be moved to the front. 6 bytes
+    // cannot be reclaimed because there is already one byte stored
+    assert_eq!(false, buf.try_reclaim(6));
+    assert_eq!(true, buf.try_reclaim(5));
+    buf.advance(1);
+    assert_eq!(true, buf.try_reclaim(6));
+    assert_eq!(6, buf.capacity());
+}
+
+#[test]
+fn try_reclaim_arc() {
+    let mut buf = BytesMut::with_capacity(6);
+    buf.put_slice(b"abc");
+    let x = buf.split().freeze();
+    buf.put_slice(b"def");
+    // Reclaiming a ludicrous amount of space should calmly return false
+    assert_eq!(false, buf.try_reclaim(usize::MAX));
+
+    let y = buf.split().freeze();
+    let z = y.clone();
+    assert_eq!(false, buf.try_reclaim(6));
+    drop(x);
+    drop(z);
+    assert_eq!(false, buf.try_reclaim(6));
+    drop(y);
+    assert_eq!(true, buf.try_reclaim(6));
+    assert_eq!(6, buf.capacity());
+    assert_eq!(0, buf.len());
+    buf.put_slice(b"abc");
+    buf.put_slice(b"def");
+    assert_eq!(6, buf.capacity());
+    assert_eq!(6, buf.len());
+    assert_eq!(false, buf.try_reclaim(6));
+    buf.advance(4);
+    assert_eq!(true, buf.try_reclaim(4));
+    buf.advance(2);
+    assert_eq!(true, buf.try_reclaim(6));
+}

tests/test_bytes_odd_alloc.rs

@@ -1,10 +1,12 @@
 //! Test using `Bytes` with an allocator that hands out "odd" pointers for
 //! vectors (pointers where the LSB is set).
 
+#![cfg(not(miri))] // Miri does not support custom allocators (also, Miri is "odd" by default with 50% chance)
+
 use std::alloc::{GlobalAlloc, Layout, System};
 use std::ptr;
 
-use bytes::Bytes;
+use bytes::{Bytes, BytesMut};
 
 #[global_allocator]
 static ODD: Odd = Odd;
@@ -22,8 +24,7 @@ unsafe impl GlobalAlloc for Odd {
             };
             let ptr = System.alloc(new_layout);
             if !ptr.is_null() {
-                let ptr = ptr.offset(1);
-                ptr
+                ptr.offset(1)
             } else {
                 ptr
             }
@@ -65,3 +66,82 @@ fn test_bytes_clone_drop() {
     let b1 = Bytes::from(vec);
     let _b2 = b1.clone();
 }
+
+#[test]
+fn test_bytes_into_vec() {
+    let vec = vec![33u8; 1024];
+
+    // Test cases where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    assert_eq!(Vec::from(b2), vec);
+
+    // Test cases where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+
+    assert_eq!(Vec::from(b2), vec[20..]);
+    assert_eq!(Vec::from(b1), vec[..20]);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_vec() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    let b1m = BytesMut::from(b1);
+    assert_eq!(b1m, vec);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_arc_1() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    let b1m = BytesMut::from(b1);
+    assert_eq!(b1m, vec);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_arc_2() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    let b1m = BytesMut::from(b1);
+    assert_eq!(b1m, vec);
+
+    // Test case where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    let b2m = BytesMut::from(b2);
+    assert_eq!(b2m, vec);
+}
+
+#[test]
+fn test_bytesmut_from_bytes_arc_offset() {
+    let vec = vec![33u8; 1024];
+
+    // Test case where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+    let b1m = BytesMut::from(b1);
+    let b2m = BytesMut::from(b2);
+
+    assert_eq!(b2m, vec[20..]);
+    assert_eq!(b1m, vec[..20]);
+}

tests/test_bytes_vec_alloc.rs

@@ -1,61 +1,87 @@
 use std::alloc::{GlobalAlloc, Layout, System};
-use std::{mem, ptr};
+use std::ptr::null_mut;
+use std::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 
 use bytes::{Buf, Bytes};
 
 #[global_allocator]
-static LEDGER: Ledger = Ledger;
+static LEDGER: Ledger = Ledger::new();
 
-struct Ledger;
+const LEDGER_LENGTH: usize = 2048;
 
-const USIZE_SIZE: usize = mem::size_of::<usize>();
+struct Ledger {
+    alloc_table: [(AtomicPtr<u8>, AtomicUsize); LEDGER_LENGTH],
+}
 
-unsafe impl GlobalAlloc for Ledger {
-    unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
-        if layout.align() == 1 && layout.size() > 0 {
-            // Allocate extra space to stash a record of
-            // how much space there was.
-            let orig_size = layout.size();
-            let size = orig_size + USIZE_SIZE;
-            let new_layout = match Layout::from_size_align(size, 1) {
-                Ok(layout) => layout,
-                Err(_err) => return ptr::null_mut(),
-            };
-            let ptr = System.alloc(new_layout);
-            if !ptr.is_null() {
-                (ptr as *mut usize).write(orig_size);
-                let ptr = ptr.offset(USIZE_SIZE as isize);
-                ptr
-            } else {
-                ptr
+impl Ledger {
+    const fn new() -> Self {
+        const ELEM: (AtomicPtr<u8>, AtomicUsize) =
+            (AtomicPtr::new(null_mut()), AtomicUsize::new(0));
+        let alloc_table = [ELEM; LEDGER_LENGTH];
+
+        Self { alloc_table }
+    }
+
+    /// Iterate over our table until we find an open entry, then insert into said entry
+    fn insert(&self, ptr: *mut u8, size: usize) {
+        for (entry_ptr, entry_size) in self.alloc_table.iter() {
+            // SeqCst is good enough here, we don't care about perf, i just want to be correct!
+            if entry_ptr
+                .compare_exchange(null_mut(), ptr, Ordering::SeqCst, Ordering::SeqCst)
+                .is_ok()
+            {
+                entry_size.store(size, Ordering::SeqCst);
+                break;
             }
-        } else {
-            System.alloc(layout)
         }
     }
 
-    unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
-        if layout.align() == 1 && layout.size() > 0 {
-            let off_ptr = (ptr as *mut usize).offset(-1);
-            let orig_size = off_ptr.read();
-            if orig_size != layout.size() {
-                panic!(
-                    "bad dealloc: alloc size was {}, dealloc size is {}",
-                    orig_size,
-                    layout.size()
-                );
+    fn remove(&self, ptr: *mut u8) -> usize {
+        for (entry_ptr, entry_size) in self.alloc_table.iter() {
+            // set the value to be something that will never try and be deallocated, so that we
+            // don't have any chance of a race condition
+            //
+            // dont worry, LEDGER_LENGTH is really long to compensate for us not reclaiming space
+            if entry_ptr
+                .compare_exchange(
+                    ptr,
+                    invalid_ptr(usize::MAX),
+                    Ordering::SeqCst,
+                    Ordering::SeqCst,
+                )
+                .is_ok()
+            {
+                return entry_size.load(Ordering::SeqCst);
             }
+        }
 
-            let new_layout = match Layout::from_size_align(layout.size() + USIZE_SIZE, 1) {
-                Ok(layout) => layout,
-                Err(_err) => std::process::abort(),
-            };
-            System.dealloc(off_ptr as *mut u8, new_layout);
+        panic!("Couldn't find a matching entry for {:x?}", ptr);
+    }
+}
+
+unsafe impl GlobalAlloc for Ledger {
+    unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
+        let size = layout.size();
+        let ptr = System.alloc(layout);
+        self.insert(ptr, size);
+        ptr
+    }
+
+    unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
+        let orig_size = self.remove(ptr);
+
+        if orig_size != layout.size() {
+            panic!(
+                "bad dealloc: alloc size was {}, dealloc size is {}",
+                orig_size,
+                layout.size()
+            );
         } else {
             System.dealloc(ptr, layout);
         }
     }
 }
+
 #[test]
 fn test_bytes_advance() {
     let mut bytes = Bytes::from(vec![10, 20, 30]);
@@ -77,3 +103,41 @@ fn test_bytes_truncate_and_advance() {
     bytes.advance(1);
     drop(bytes);
 }
+
+/// Returns a dangling pointer with the given address. This is used to store
+/// integer data in pointer fields.
+#[inline]
+fn invalid_ptr<T>(addr: usize) -> *mut T {
+    let ptr = std::ptr::null_mut::<u8>().wrapping_add(addr);
+    debug_assert_eq!(ptr as usize, addr);
+    ptr.cast::<T>()
+}
+
+#[test]
+fn test_bytes_into_vec() {
+    let vec = vec![33u8; 1024];
+
+    // Test cases where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    assert_eq!(Vec::from(b2), vec);
+
+    // Test cases where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+
+    assert_eq!(Vec::from(b2), vec[20..]);
+    assert_eq!(Vec::from(b1), vec[..20]);
+}

tests/test_chain.rs

@@ -62,7 +62,7 @@ fn vectored_read() {
             IoSlice::new(b4),
         ];
 
-        assert_eq!(2, buf.bytes_vectored(&mut iovecs));
+        assert_eq!(2, buf.chunks_vectored(&mut iovecs));
         assert_eq!(iovecs[0][..], b"hello"[..]);
         assert_eq!(iovecs[1][..], b"world"[..]);
         assert_eq!(iovecs[2][..], b""[..]);
@@ -83,7 +83,7 @@ fn vectored_read() {
             IoSlice::new(b4),
         ];
 
-        assert_eq!(2, buf.bytes_vectored(&mut iovecs));
+        assert_eq!(2, buf.chunks_vectored(&mut iovecs));
         assert_eq!(iovecs[0][..], b"llo"[..]);
         assert_eq!(iovecs[1][..], b"world"[..]);
         assert_eq!(iovecs[2][..], b""[..]);
@@ -104,7 +104,7 @@ fn vectored_read() {
             IoSlice::new(b4),
         ];
 
-        assert_eq!(1, buf.bytes_vectored(&mut iovecs));
+        assert_eq!(1, buf.chunks_vectored(&mut iovecs));
         assert_eq!(iovecs[0][..], b"world"[..]);
         assert_eq!(iovecs[1][..], b""[..]);
         assert_eq!(iovecs[2][..], b""[..]);
@@ -125,10 +125,53 @@ fn vectored_read() {
             IoSlice::new(b4),
         ];
 
-        assert_eq!(1, buf.bytes_vectored(&mut iovecs));
+        assert_eq!(1, buf.chunks_vectored(&mut iovecs));
         assert_eq!(iovecs[0][..], b"ld"[..]);
         assert_eq!(iovecs[1][..], b""[..]);
         assert_eq!(iovecs[2][..], b""[..]);
         assert_eq!(iovecs[3][..], b""[..]);
     }
 }
+
+#[test]
+fn chain_growing_buffer() {
+    let mut buff = [' ' as u8; 10];
+    let mut vec = b"wassup".to_vec();
+
+    let mut chained = (&mut buff[..]).chain_mut(&mut vec).chain_mut(Vec::new()); // Required for potential overflow because remaining_mut for Vec is isize::MAX - vec.len(), but for chain_mut is usize::MAX
+
+    chained.put_slice(b"hey there123123");
+
+    assert_eq!(&buff, b"hey there1");
+    assert_eq!(&vec, b"wassup23123");
+}
+
+#[test]
+fn chain_overflow_remaining_mut() {
+    let mut chained = Vec::<u8>::new().chain_mut(Vec::new()).chain_mut(Vec::new());
+
+    assert_eq!(chained.remaining_mut(), usize::MAX);
+    chained.put_slice(&[0; 256]);
+    assert_eq!(chained.remaining_mut(), usize::MAX);
+}
+
+#[test]
+fn chain_get_bytes() {
+    let mut ab = Bytes::copy_from_slice(b"ab");
+    let mut cd = Bytes::copy_from_slice(b"cd");
+    let ab_ptr = ab.as_ptr();
+    let cd_ptr = cd.as_ptr();
+    let mut chain = (&mut ab).chain(&mut cd);
+    let a = chain.copy_to_bytes(1);
+    let bc = chain.copy_to_bytes(2);
+    let d = chain.copy_to_bytes(1);
+
+    assert_eq!(Bytes::copy_from_slice(b"a"), a);
+    assert_eq!(Bytes::copy_from_slice(b"bc"), bc);
+    assert_eq!(Bytes::copy_from_slice(b"d"), d);
+
+    // assert `get_bytes` did not allocate
+    assert_eq!(ab_ptr, a.as_ptr());
+    // assert `get_bytes` did not allocate
+    assert_eq!(cd_ptr.wrapping_offset(1), d.as_ptr());
+}

tests/test_iter.rs

@@ -1,11 +1,11 @@
 #![warn(rust_2018_idioms)]
 
-use bytes::Bytes;
+use bytes::{buf::IntoIter, Bytes};
 
 #[test]
 fn iter_len() {
     let buf = Bytes::from_static(b"hello world");
-    let iter = buf.iter();
+    let iter = IntoIter::new(buf);
 
     assert_eq!(iter.size_hint(), (11, Some(11)));
     assert_eq!(iter.len(), 11);
@@ -13,8 +13,8 @@ fn iter_len() {
 
 #[test]
 fn empty_iter_len() {
-    let buf = Bytes::from_static(b"");
-    let iter = buf.iter();
+    let buf = Bytes::new();
+    let iter = IntoIter::new(buf);
 
     assert_eq!(iter.size_hint(), (0, Some(0)));
     assert_eq!(iter.len(), 0);

tests/test_take.rs

@@ -1,6 +1,7 @@
 #![warn(rust_2018_idioms)]
 
 use bytes::buf::Buf;
+use bytes::Bytes;
 
 #[test]
 fn long_take() {
@@ -8,5 +9,24 @@ fn long_take() {
     // overrun the buffer. Regression test for #138.
     let buf = b"hello world".take(100);
     assert_eq!(11, buf.remaining());
-    assert_eq!(b"hello world", buf.bytes());
+    assert_eq!(b"hello world", buf.chunk());
+}
+
+#[test]
+fn take_copy_to_bytes() {
+    let mut abcd = Bytes::copy_from_slice(b"abcd");
+    let abcd_ptr = abcd.as_ptr();
+    let mut take = (&mut abcd).take(2);
+    let a = take.copy_to_bytes(1);
+    assert_eq!(Bytes::copy_from_slice(b"a"), a);
+    // assert `to_bytes` did not allocate
+    assert_eq!(abcd_ptr, a.as_ptr());
+    assert_eq!(Bytes::copy_from_slice(b"bcd"), abcd);
+}
+
+#[test]
+#[should_panic]
+fn take_copy_to_bytes_panics() {
+    let abcd = Bytes::copy_from_slice(b"abcd");
+    abcd.take(2).copy_to_bytes(3);
 }