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.

2 files changed, 30 insertions, 23 deletions

diff --git a/compiler/rustc_ast/src/ast.rs b/compiler/rustc_ast/src/ast.rs
index b2d8881e3f6..f5418402377 100644
--- a/compiler/rustc_ast/src/ast.rs
+++ b/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'`).
diff --git a/compiler/rustc_ast/src/util/literal.rs b/compiler/rustc_ast/src/util/literal.rs
index ad9e5d1468b..fa7878873e5 100644
--- a/compiler/rustc_ast/src/util/literal.rs
+++ b/compiler/rustc_ast/src/util/literal.rs
@@ -5,7 +5,7 @@ use std::{ascii, fmt, str};
 use rustc_literal_escaper::{
     MixedUnit, unescape_byte, unescape_byte_str, unescape_c_str, unescape_char, unescape_str,
 };
-use rustc_span::{Span, Symbol, kw, sym};
+use rustc_span::{ByteSymbol, Span, Symbol, kw, sym};
 use tracing::debug;
 
 use crate::ast::{self, LitKind, MetaItemLit, StrStyle};
@@ -116,13 +116,12 @@ impl LitKind {
                         assert!(!err.is_fatal(), "failed to unescape string literal")
                     }
                 });
-                LitKind::ByteStr(buf.into(), StrStyle::Cooked)
+                LitKind::ByteStr(ByteSymbol::intern(&buf), StrStyle::Cooked)
             }
             token::ByteStrRaw(n) => {
-                // Raw strings have no escapes so we can convert the symbol
-                // directly to a `Arc<u8>`.
+                // Raw byte strings have no escapes so no work is needed here.
                 let buf = symbol.as_str().to_owned().into_bytes();
-                LitKind::ByteStr(buf.into(), StrStyle::Raw(n))
+                LitKind::ByteStr(ByteSymbol::intern(&buf), StrStyle::Raw(n))
             }
             token::CStr => {
                 let s = symbol.as_str();
@@ -137,7 +136,7 @@ impl LitKind {
                     }
                 });
                 buf.push(0);
-                LitKind::CStr(buf.into(), StrStyle::Cooked)
+                LitKind::CStr(ByteSymbol::intern(&buf), StrStyle::Cooked)
             }
             token::CStrRaw(n) => {
                 // Raw strings have no escapes so we can convert the symbol
@@ -145,7 +144,7 @@ impl LitKind {
                 // char.
                 let mut buf = symbol.as_str().to_owned().into_bytes();
                 buf.push(0);
-                LitKind::CStr(buf.into(), StrStyle::Raw(n))
+                LitKind::CStr(ByteSymbol::intern(&buf), StrStyle::Raw(n))
             }
             token::Err(guar) => LitKind::Err(guar),
         })
@@ -167,12 +166,12 @@ impl fmt::Display for LitKind {
                 delim = "#".repeat(n as usize),
                 string = sym
             )?,
-            LitKind::ByteStr(ref bytes, StrStyle::Cooked) => {
-                write!(f, "b\"{}\"", escape_byte_str_symbol(bytes))?
+            LitKind::ByteStr(ref byte_sym, StrStyle::Cooked) => {
+                write!(f, "b\"{}\"", escape_byte_str_symbol(byte_sym.as_byte_str()))?
             }
-            LitKind::ByteStr(ref bytes, StrStyle::Raw(n)) => {
+            LitKind::ByteStr(ref byte_sym, StrStyle::Raw(n)) => {
                 // Unwrap because raw byte string literals can only contain ASCII.
-                let symbol = str::from_utf8(bytes).unwrap();
+                let symbol = str::from_utf8(byte_sym.as_byte_str()).unwrap();
                 write!(
                     f,
                     "br{delim}\"{string}\"{delim}",
@@ -181,11 +180,11 @@ impl fmt::Display for LitKind {
                 )?;
             }
             LitKind::CStr(ref bytes, StrStyle::Cooked) => {
-                write!(f, "c\"{}\"", escape_byte_str_symbol(bytes))?
+                write!(f, "c\"{}\"", escape_byte_str_symbol(bytes.as_byte_str()))?
             }
             LitKind::CStr(ref bytes, StrStyle::Raw(n)) => {
                 // This can only be valid UTF-8.
-                let symbol = str::from_utf8(bytes).unwrap();
+                let symbol = str::from_utf8(bytes.as_byte_str()).unwrap();
                 write!(f, "cr{delim}\"{symbol}\"{delim}", delim = "#".repeat(n as usize),)?;
             }
             LitKind::Int(n, ty) => {