Auto merge of #42727 - alexcrichton:allocators-new, r=eddyb

rustc: Implement the #[global_allocator] attribute

This PR is an implementation of [RFC 1974] which specifies a new method of
defining a global allocator for a program. This obsoletes the old
`#![allocator]` attribute and also removes support for it.

[RFC 1974]: rust-lang/rfcs#1974

The new `#[global_allocator]` attribute solves many issues encountered with the
`#![allocator]` attribute such as composition and restrictions on the crate
graph itself. The compiler now has much more control over the ABI of the
allocator and how it's implemented, allowing much more freedom in terms of how
this feature is implemented.

cc #27389

Author: bors

src/Cargo.lock

@@ -0,0 +1,7 @@
+# `allocator_internals`
+
+This feature does not have a tracking issue, it is an unstable implementation
+detail of the `global_allocator` feature not intended for use outside the
+compiler.
+
+------------------------

src/doc/unstable-book/src/language-features/allocator-internals.md

@@ -0,0 +1,71 @@
+# `global_allocator`
+
+The tracking issue for this feature is: [#27389]
+
+[#27389]: https://github.com/rust-lang/rust/issues/27389
+
+------------------------
+
+Rust programs may need to change the allocator that they're running with from
+time to time. This use case is distinct from an allocator-per-collection (e.g. a
+`Vec` with a custom allocator) and instead is more related to changing the
+global default allocator, e.g. what `Vec<T>` uses by default.
+
+Currently Rust programs don't have a specified global allocator. The compiler
+may link to a version of [jemalloc] on some platforms, but this is not
+guaranteed. Libraries, however, like cdylibs and staticlibs are guaranteed
+to use the "system allocator" which means something like `malloc` on Unixes and
+`HeapAlloc` on Windows.
+
+[jemalloc]: https://github.com/jemalloc/jemalloc
+
+The `#[global_allocator]` attribute, however, allows configuring this choice.
+You can use this to implement a completely custom global allocator to route all
+default allocation requests to a custom object. Defined in [RFC 1974] usage
+looks like:
+
+[RFC 1974]: https://github.com/rust-lang/rfcs/pull/1974
+
+```rust
+#![feature(global_allocator, heap_api)]
+
+use std::heap::{Alloc, System, Layout, AllocErr};
+
+struct MyAllocator;
+
+unsafe impl<'a> Alloc for &'a MyAllocator {
+    unsafe fn alloc(&mut self, layout: Layout) -> Result<*mut u8, AllocErr> {
+        System.alloc(layout)
+    }
+
+    unsafe fn dealloc(&mut self, ptr: *mut u8, layout: Layout) {
+        System.dealloc(ptr, layout)
+    }
+}
+
+#[global_allocator]
+static GLOBAL: MyAllocator = MyAllocator;
+
+fn main() {
+    // This `Vec` will allocate memory through `GLOBAL` above
+    let mut v = Vec::new();
+    v.push(1);
+}
+```
+
+And that's it! The `#[global_allocator]` attribute is applied to a `static`
+which implements the `Alloc` trait in the `std::heap` module. Note, though,
+that the implementation is defined for `&MyAllocator`, not just `MyAllocator`.
+You may wish, however, to also provide `Alloc for MyAllocator` for other use
+cases.
+
+A crate can only have one instance of `#[global_allocator]` and this instance
+may be loaded through a dependency. For example `#[global_allocator]` above
+could have been placed in one of the dependencies loaded through `extern crate`.
+
+Note that `Alloc` itself is an `unsafe` trait, with much documentation on the
+trait itself about usage and for implementors. Extra care should be taken when
+implementing a global allocator as well as the allocator may be called from many
+portions of the standard library, such as the panicking routine. As a result it
+is highly recommended to not panic during allocation and work in as many
+situations with as few dependencies as possible as well.

src/doc/unstable-book/src/language-features/allocator.md

@@ -13,7 +13,7 @@
                       slightly, especially to possibly take into account the \
                       types being stored to make room for a future \
                       tracing garbage collector",
-            issue = "27700")]
+            issue = "32838")]
 
 use core::cmp;
 use core::fmt;
@@ -73,6 +73,7 @@ impl Layout {
     /// * `size`, when rounded up to the nearest multiple of `align`,
     ///    must not overflow (i.e. the rounded value must be less than
     ///    `usize::MAX`).
+    #[inline]
     pub fn from_size_align(size: usize, align: usize) -> Option<Layout> {
         if !align.is_power_of_two() {
             return None;
@@ -96,13 +97,28 @@ impl Layout {
             return None;
         }
 
-        Some(Layout { size: size, align: align })
+        unsafe {
+            Some(Layout::from_size_align_unchecked(size, align))
+        }
+    }
+
+    /// Creates a layout, bypassing all checks.
+    ///
+    /// # Unsafety
+    ///
+    /// This function is unsafe as it does not verify that `align` is a power of
+    /// two nor that `size` aligned to `align` fits within the address space.
+    #[inline]
+    pub unsafe fn from_size_align_unchecked(size: usize, align: usize) -> Layout {
+        Layout { size: size, align: align }
     }
 
     /// The minimum size in bytes for a memory block of this layout.
+    #[inline]
     pub fn size(&self) -> usize { self.size }
 
     /// The minimum byte alignment for a memory block of this layout.
+    #[inline]
     pub fn align(&self) -> usize { self.align }
 
     /// Constructs a `Layout` suitable for holding a value of type `T`.
@@ -135,6 +151,7 @@ impl Layout {
     ///
     /// Panics if the combination of `self.size` and the given `align`
     /// violates the conditions listed in `from_size_align`.
+    #[inline]
     pub fn align_to(&self, align: usize) -> Self {
         Layout::from_size_align(self.size, cmp::max(self.align, align)).unwrap()
     }
@@ -155,6 +172,7 @@ impl Layout {
     /// to be less than or equal to the alignment of the starting
     /// address for the whole allocated block of memory. One way to
     /// satisfy this constraint is to ensure `align <= self.align`.
+    #[inline]
     pub fn padding_needed_for(&self, align: usize) -> usize {
         let len = self.size();
 
@@ -556,6 +574,7 @@ pub unsafe trait Alloc {
     /// However, for clients that do not wish to track the capacity
     /// returned by `alloc_excess` locally, this method is likely to
     /// produce useful results.
+    #[inline]
     fn usable_size(&self, layout: &Layout) -> (usize, usize) {
         (layout.size(), layout.size())
     }

src/doc/unstable-book/src/language-features/global-allocator.md

@@ -23,7 +23,6 @@ use core::sync::atomic::Ordering::{Acquire, Relaxed, Release, SeqCst};
 use core::borrow;
 use core::fmt;
 use core::cmp::Ordering;
-use core::mem::{align_of_val, size_of_val};
 use core::intrinsics::abort;
 use core::mem;
 use core::mem::uninitialized;
@@ -34,7 +33,8 @@ use core::marker::Unsize;
 use core::hash::{Hash, Hasher};
 use core::{isize, usize};
 use core::convert::From;
-use heap::deallocate;
+
+use heap::{Heap, Alloc, Layout};
 
 /// A soft limit on the amount of references that may be made to an `Arc`.
 ///
@@ -503,7 +503,7 @@ impl<T: ?Sized> Arc<T> {
 
         if self.inner().weak.fetch_sub(1, Release) == 1 {
             atomic::fence(Acquire);
-            deallocate(ptr as *mut u8, size_of_val(&*ptr), align_of_val(&*ptr))
+            Heap.dealloc(ptr as *mut u8, Layout::for_value(&*ptr))
         }
     }
 
@@ -1007,7 +1007,9 @@ impl<T: ?Sized> Drop for Weak<T> {
         // ref, which can only happen after the lock is released.
         if self.inner().weak.fetch_sub(1, Release) == 1 {
             atomic::fence(Acquire);
-            unsafe { deallocate(ptr as *mut u8, size_of_val(&*ptr), align_of_val(&*ptr)) }
+            unsafe {
+                Heap.dealloc(ptr as *mut u8, Layout::for_value(&*ptr))
+            }
         }
     }
 }