Tell rustc about unused bits in Span.

This shrinks the size of `Option<Span>` and similar types, which shrinks
several AST and HIR nodes. The most important of these are `hir::Expr`
and `ast::Ty`.

This change was first attempted in rust-lang#93747 where it had little effect.
But the improved niche-filling implemented by rust-lang#94075 means this change
now has a much bigger effect.

Co-authored-by: Camille Gillot <[email protected]>

Author: nnethercote

compiler/rustc_ast/src/ast.rs

@@ -3040,19 +3040,19 @@ mod size_asserts {
     use super::*;
     use rustc_data_structures::static_assert_size;
     // tidy-alphabetical-start
-    static_assert_size!(AssocItem, 104);
-    static_assert_size!(AssocItemKind, 32);
+    static_assert_size!(AssocItem, 96);
+    static_assert_size!(AssocItemKind, 24);
     static_assert_size!(Attribute, 32);
     static_assert_size!(Block, 48);
     static_assert_size!(Expr, 104);
     static_assert_size!(ExprKind, 72);
-    static_assert_size!(Fn, 184);
+    static_assert_size!(Fn, 168);
     static_assert_size!(ForeignItem, 96);
     static_assert_size!(ForeignItemKind, 24);
     static_assert_size!(GenericArg, 24);
     static_assert_size!(GenericBound, 88);
     static_assert_size!(Generics, 72);
-    static_assert_size!(Impl, 200);
+    static_assert_size!(Impl, 184);
     static_assert_size!(Item, 184);
     static_assert_size!(ItemKind, 112);
     static_assert_size!(Lit, 48);
@@ -3065,7 +3065,7 @@ mod size_asserts {
     static_assert_size!(PatKind, 96);
     static_assert_size!(Stmt, 32);
     static_assert_size!(StmtKind, 16);
-    static_assert_size!(Ty, 96);
-    static_assert_size!(TyKind, 72);
+    static_assert_size!(Ty, 88);
+    static_assert_size!(TyKind, 64);
     // tidy-alphabetical-end
 }

compiler/rustc_hir/src/hir.rs

@@ -3517,20 +3517,20 @@ mod size_asserts {
     // tidy-alphabetical-start
     static_assert_size!(Block<'_>, 48);
     static_assert_size!(Body<'_>, 32);
-    static_assert_size!(Expr<'_>, 64);
-    static_assert_size!(ExprKind<'_>, 48);
+    static_assert_size!(Expr<'_>, 56);
+    static_assert_size!(ExprKind<'_>, 40);
     static_assert_size!(FnDecl<'_>, 40);
     static_assert_size!(ForeignItem<'_>, 72);
     static_assert_size!(ForeignItemKind<'_>, 40);
     static_assert_size!(GenericArg<'_>, 24);
     static_assert_size!(GenericBound<'_>, 48);
     static_assert_size!(Generics<'_>, 56);
-    static_assert_size!(Impl<'_>, 80);
+    static_assert_size!(Impl<'_>, 72);
     static_assert_size!(ImplItem<'_>, 80);
     static_assert_size!(ImplItemKind<'_>, 32);
     static_assert_size!(Item<'_>, 80);
     static_assert_size!(ItemKind<'_>, 48);
-    static_assert_size!(Local<'_>, 64);
+    static_assert_size!(Local<'_>, 56);
     static_assert_size!(Param<'_>, 32);
     static_assert_size!(Pat<'_>, 72);
     static_assert_size!(Path<'_>, 40);
@@ -3540,8 +3540,8 @@ mod size_asserts {
     static_assert_size!(Res, 12);
     static_assert_size!(Stmt<'_>, 32);
     static_assert_size!(StmtKind<'_>, 16);
-    static_assert_size!(TraitItem<'_>, 88);
-    static_assert_size!(TraitItemKind<'_>, 48);
+    static_assert_size!(TraitItem<'_>, 80);
+    static_assert_size!(TraitItemKind<'_>, 40);
     static_assert_size!(Ty<'_>, 48);
     static_assert_size!(TyKind<'_>, 32);
     // tidy-alphabetical-end

compiler/rustc_parse/src/parser/mod.rs

@@ -156,7 +156,7 @@ pub struct Parser<'a> {
 // This type is used a lot, e.g. it's cloned when matching many declarative macro rules. Make sure
 // it doesn't unintentionally get bigger.
 #[cfg(all(target_arch = "x86_64", target_pointer_width = "64"))]
-rustc_data_structures::static_assert_size!(Parser<'_>, 328);
+rustc_data_structures::static_assert_size!(Parser<'_>, 320);
 
 /// Stores span information about a closure.
 #[derive(Clone)]

compiler/rustc_span/src/span_encoding.rs

@@ -69,17 +69,23 @@ use rustc_data_structures::fx::FxIndexSet;
 #[rustc_pass_by_value]
 pub struct Span {
     base_or_index: u32,
-    len_or_tag: u16,
+    len_or_tag: LenOrTag,
     ctxt_or_tag: u16,
 }
 
-const LEN_TAG: u16 = 0b1000_0000_0000_0000;
+// LEN_TAG allows for some extra values at the top. Declare them to rustc to use as niches.
+#[rustc_layout_scalar_valid_range_end(0b1000_0000_0000_0000)]
+#[derive(Copy, Clone, Eq, PartialEq, Hash)]
+struct LenOrTag(u16);
+
+const LEN_TAG: LenOrTag = unsafe { LenOrTag(0b1000_0000_0000_0000) };
 const MAX_LEN: u32 = 0b0111_1111_1111_1111;
 const CTXT_TAG: u32 = 0b1111_1111_1111_1111;
 const MAX_CTXT: u32 = CTXT_TAG - 1;
 
 /// Dummy span, both position and length are zero, syntax context is zero as well.
-pub const DUMMY_SP: Span = Span { base_or_index: 0, len_or_tag: 0, ctxt_or_tag: 0 };
+pub const DUMMY_SP: Span =
+    Span { base_or_index: 0, len_or_tag: unsafe { LenOrTag(0) }, ctxt_or_tag: 0 };
 
 impl Span {
     #[inline]
@@ -97,7 +103,11 @@ impl Span {
 
         if len <= MAX_LEN && ctxt2 <= MAX_CTXT && parent.is_none() {
             // Inline format.
-            Span { base_or_index: base, len_or_tag: len as u16, ctxt_or_tag: ctxt2 as u16 }
+            Span {
+                base_or_index: base,
+                len_or_tag: unsafe { LenOrTag(len as u16) },
+                ctxt_or_tag: ctxt2 as u16,
+            }
         } else {
             // Interned format.
             let index =
@@ -122,10 +132,10 @@ impl Span {
     pub fn data_untracked(self) -> SpanData {
         if self.len_or_tag != LEN_TAG {
             // Inline format.
-            debug_assert!(self.len_or_tag as u32 <= MAX_LEN);
+            debug_assert!(self.len_or_tag.0 as u32 <= MAX_LEN);
             SpanData {
                 lo: BytePos(self.base_or_index),
-                hi: BytePos(self.base_or_index + self.len_or_tag as u32),
+                hi: BytePos(self.base_or_index + self.len_or_tag.0 as u32),
                 ctxt: SyntaxContext::from_u32(self.ctxt_or_tag as u32),
                 parent: None,
             }