Introduce ByteSymbol.

It's like `Symbol` but for byte strings. The interner is now used for
both `Symbol` and `ByteSymbol`. E.g. if you intern `"dog"` and `b"dog"`
you'll get a `Symbol` and a `ByteSymbol` with the same index and the
characters will only be stored once.

The motivation for this is to eliminate the `Arc`s in `ast::LitKind`, to
make `ast::LitKind` impl `Copy`, and to avoid the need to arena-allocate
`ast::LitKind` in HIR. The latter change reduces peak memory by a
non-trivial amount on literal-heavy benchmarks such as `deep-vector` and
`tuple-stress`.

`Encoder`, `Decoder`, `SpanEncoder`, and `SpanDecoder` all get some
changes so that they can handle normal strings and byte strings.

This change does slow down compilation of programs that use
`include_bytes!` on large files, because the contents of those files are
now interned (hashed). This makes `include_bytes!` more similar to
`include_str!`, though `include_bytes!` contents still aren't escaped,
and hashing is still much cheaper than escaping.

compiler/rustc_ast/src/ast.rs

@@ -19,7 +19,6 @@
 //! - [`UnOp`], [`BinOp`], and [`BinOpKind`]: Unary and binary operators.
 
 use std::borrow::Cow;
-use std::sync::Arc;
 use std::{cmp, fmt};
 
 pub use GenericArgs::*;
@@ -32,7 +31,7 @@ use rustc_data_structures::tagged_ptr::Tag;
 use rustc_macros::{Decodable, Encodable, HashStable_Generic};
 pub use rustc_span::AttrId;
 use rustc_span::source_map::{Spanned, respan};
-use rustc_span::{DUMMY_SP, ErrorGuaranteed, Ident, Span, Symbol, kw, sym};
+use rustc_span::{ByteSymbol, DUMMY_SP, ErrorGuaranteed, Ident, Span, Symbol, kw, sym};
 use thin_vec::{ThinVec, thin_vec};
 
 pub use crate::format::*;
@@ -1805,10 +1804,17 @@ pub enum ExprKind {
     Become(P<Expr>),
 
     /// Bytes included via `include_bytes!`
+    ///
     /// Added for optimization purposes to avoid the need to escape
     /// large binary blobs - should always behave like [`ExprKind::Lit`]
     /// with a `ByteStr` literal.
-    IncludedBytes(Arc<[u8]>),
+    ///
+    /// The value is stored as a `ByteSymbol`. It's unfortunate that we need to
+    /// intern (hash) the bytes because they're likely to be large and unique.
+    /// But it's necessary because this will eventually be lowered to
+    /// `LitKind::ByteStr`, which needs a `ByteSymbol` to impl `Copy` and avoid
+    /// arena allocation.
+    IncludedBytes(ByteSymbol),
 
     /// A `format_args!()` expression.
     FormatArgs(P<FormatArgs>),
@@ -2066,7 +2072,7 @@ impl YieldKind {
 }
 
 /// A literal in a meta item.
-#[derive(Clone, Encodable, Decodable, Debug, HashStable_Generic)]
+#[derive(Clone, Copy, Encodable, Decodable, Debug, HashStable_Generic)]
 pub struct MetaItemLit {
     /// The original literal as written in the source code.
     pub symbol: Symbol,
@@ -2129,16 +2135,18 @@ pub enum LitFloatType {
 /// deciding the `LitKind`. This means that float literals like `1f32` are
 /// classified by this type as `Float`. This is different to `token::LitKind`
 /// which does *not* consider the suffix.
-#[derive(Clone, Encodable, Decodable, Debug, Hash, Eq, PartialEq, HashStable_Generic)]
+#[derive(Clone, Copy, Encodable, Decodable, Debug, Hash, Eq, PartialEq, HashStable_Generic)]
 pub enum LitKind {
     /// A string literal (`"foo"`). The symbol is unescaped, and so may differ
     /// from the original token's symbol.
     Str(Symbol, StrStyle),
-    /// A byte string (`b"foo"`). Not stored as a symbol because it might be
-    /// non-utf8, and symbols only allow utf8 strings.
-    ByteStr(Arc<[u8]>, StrStyle),
-    /// A C String (`c"foo"`). Guaranteed to only have `\0` at the end.
-    CStr(Arc<[u8]>, StrStyle),
+    /// A byte string (`b"foo"`). The symbol is unescaped, and so may differ
+    /// from the original token's symbol.
+    ByteStr(ByteSymbol, StrStyle),
+    /// A C String (`c"foo"`). Guaranteed to only have `\0` at the end. The
+    /// symbol is unescaped, and so may differ from the original token's
+    /// symbol.
+    CStr(ByteSymbol, StrStyle),
     /// A byte char (`b'f'`).
     Byte(u8),
     /// A character literal (`'a'`).