perf(minifier): normalize undefined to void 0 before everything else (#8699)

Boshen · web-flow · commit 8587965e4557 · 2025-01-25T11:50:08.000+08:00
so subsequent code don't need to lookup `undefined`.
diff --git a/crates/oxc_minifier/examples/minifier.rs b/crates/oxc_minifier/examples/minifier.rs
@@ -27,7 +27,7 @@ fn main() -> std::io::Result<()> {
 
     let mut allocator = Allocator::default();
     let printed = minify(&allocator, &source_text, source_type, mangle, nospace);
-    println!("{printed}");
+    // println!("{printed}");
 
     if twice {
         allocator.reset();
diff --git a/crates/oxc_minifier/src/ctx.rs b/crates/oxc_minifier/src/ctx.rs
@@ -62,20 +62,23 @@ impl<'a> Ctx<'a, '_> {
         }
     }
 
+    #[inline]
     pub fn is_identifier_undefined(self, ident: &IdentifierReference) -> bool {
         if ident.name == "undefined" && ident.is_global_reference(self.symbols()) {
             return true;
         }
         false
     }
 
+    #[inline]
     pub fn is_identifier_infinity(self, ident: &IdentifierReference) -> bool {
         if ident.name == "Infinity" && ident.is_global_reference(self.symbols()) {
             return true;
         }
         false
     }
 
+    #[inline]
     pub fn is_identifier_nan(self, ident: &IdentifierReference) -> bool {
         if ident.name == "NaN" && ident.is_global_reference(self.symbols()) {
             return true;
diff --git a/crates/oxc_minifier/src/peephole/fold_constants.rs b/crates/oxc_minifier/src/peephole/fold_constants.rs
@@ -573,34 +573,34 @@ impl<'a, 'b> PeepholeOptimizations {
         if e.arguments.len() != 1 {
             return None;
         }
-        Some(ctx.value_to_expr(
-            e.span,
-            ConstantValue::Number(match &e.arguments[0] {
-                // `Number(undefined)` -> `NaN`
-                Argument::Identifier(ident) if ctx.is_identifier_undefined(ident) => f64::NAN,
-                // `Number(null)` -> `0`
-                Argument::NullLiteral(_) => 0.0,
-                // `Number(true)` -> `1` `Number(false)` -> `0`
-                Argument::BooleanLiteral(b) => f64::from(b.value),
-                // `Number(100)` -> `100`
-                Argument::NumericLiteral(n) => n.value,
-                // `Number("a")` -> `+"a"` -> `NaN`
-                // `Number("1")` -> `+"1"` -> `1`
-                Argument::StringLiteral(n) => {
-                    let argument = ctx.ast.expression_string_literal(n.span, n.value, n.raw);
-                    if let Some(n) = ctx.eval_to_number(&argument) {
-                        n
-                    } else {
-                        return Some(ctx.ast.expression_unary(
-                            e.span,
-                            UnaryOperator::UnaryPlus,
-                            argument,
-                        ));
-                    }
+        let arg = e.arguments[0].as_expression()?;
+        let value = ConstantValue::Number(match arg {
+            // `Number(undefined)` -> `NaN`
+            Expression::Identifier(ident) if ctx.is_identifier_undefined(ident) => f64::NAN,
+            // `Number(null)` -> `0`
+            Expression::NullLiteral(_) => 0.0,
+            // `Number(true)` -> `1` `Number(false)` -> `0`
+            Expression::BooleanLiteral(b) => f64::from(b.value),
+            // `Number(100)` -> `100`
+            Expression::NumericLiteral(n) => n.value,
+            // `Number("a")` -> `+"a"` -> `NaN`
+            // `Number("1")` -> `+"1"` -> `1`
+            Expression::StringLiteral(n) => {
+                let argument = ctx.ast.expression_string_literal(n.span, n.value, n.raw);
+                if let Some(n) = ctx.eval_to_number(&argument) {
+                    n
+                } else {
+                    return Some(ctx.ast.expression_unary(
+                        e.span,
+                        UnaryOperator::UnaryPlus,
+                        argument,
+                    ));
                 }
-                _ => return None,
-            }),
-        ))
+            }
+            e if e.is_void_0() => f64::NAN,
+            _ => return None,
+        });
+        Some(ctx.value_to_expr(e.span, value))
     }
 
     fn try_fold_binary_typeof_comparison(
@@ -1744,6 +1744,7 @@ mod test {
     #[test]
     fn test_number_constructor() {
         test("Number(undefined)", "NaN");
+        test("Number(void 0)", "NaN");
         test("Number(null)", "0");
         test("Number(true)", "1");
         test("Number(false)", "0");
diff --git a/crates/oxc_minifier/src/peephole/minimize_conditions.rs b/crates/oxc_minifier/src/peephole/minimize_conditions.rs
@@ -4,10 +4,9 @@ use oxc_ecmascript::{
     constant_evaluation::{ConstantEvaluation, ValueType},
     ToInt32,
 };
-use oxc_semantic::ReferenceFlags;
 use oxc_span::{cmp::ContentEq, GetSpan};
 use oxc_syntax::es_target::ESTarget;
-use oxc_traverse::{Ancestor, MaybeBoundIdentifier, TraverseCtx};
+use oxc_traverse::{Ancestor, TraverseCtx};
 
 use crate::ctx::Ctx;
 
@@ -366,13 +365,9 @@ impl<'a> PeepholeOptimizations {
             unreachable!()
         };
         let return_stmt = return_stmt.unbox();
-        if let Some(e) = return_stmt.argument {
-            e
-        } else {
-            let name = "undefined";
-            let symbol_id = ctx.scopes().find_binding(ctx.current_scope_id(), name);
-            let ident = MaybeBoundIdentifier::new(Atom::from(name), symbol_id);
-            ident.create_expression(ReferenceFlags::read(), ctx)
+        match return_stmt.argument {
+            Some(e) => e,
+            None => ctx.ast.void_0(return_stmt.span),
         }
     }
 
diff --git a/crates/oxc_minifier/src/peephole/normalize.rs b/crates/oxc_minifier/src/peephole/normalize.rs
@@ -3,7 +3,7 @@ use oxc_ast::ast::*;
 use oxc_ecmascript::constant_evaluation::ConstantEvaluation;
 use oxc_span::GetSpan;
 use oxc_syntax::scope::ScopeFlags;
-use oxc_traverse::{traverse_mut_with_ctx, ReusableTraverseCtx, Traverse, TraverseCtx};
+use oxc_traverse::{traverse_mut_with_ctx, Ancestor, ReusableTraverseCtx, Traverse, TraverseCtx};
 
 use crate::{ctx::Ctx, CompressOptions};
 
@@ -22,6 +22,7 @@ pub struct NormalizeOptions {
 /// * convert `Infinity` to `f64::INFINITY`
 /// * convert `NaN` to `f64::NaN`
 /// * convert `var x; void x` to `void 0`
+/// * convert `undefined` to `void 0`
 ///
 /// Also
 ///
@@ -62,7 +63,10 @@ impl<'a> Traverse<'a> for Normalize {
             *expr = ctx.ast.move_expression(&mut paren_expr.expression);
         }
         match expr {
-            Expression::Identifier(_) => {
+            Expression::Identifier(ident) => {
+                if let Some(e) = Self::try_compress_undefined(ident, ctx) {
+                    *expr = e;
+                }
                 Self::convert_infinity_or_nan_into_number(expr, ctx);
             }
             Expression::UnaryExpression(e) if e.operator.is_void() => {
@@ -154,6 +158,24 @@ impl<'a> Normalize {
         }
         e.argument = ctx.ast.expression_numeric_literal(ident.span, 0.0, None, NumberBase::Decimal);
     }
+
+    /// Transforms `undefined` => `void 0`
+    /// So subsequent passes don't need to look up whether `undefined` is shadowed or not.
+    fn try_compress_undefined(
+        ident: &IdentifierReference<'a>,
+        ctx: &mut TraverseCtx<'a>,
+    ) -> Option<Expression<'a>> {
+        if !Ctx(ctx).is_identifier_undefined(ident) {
+            return None;
+        }
+        // `delete undefined` returns `false`
+        // `delete void 0` returns `true`
+        if matches!(ctx.parent(), Ancestor::UnaryExpressionArgument(e) if e.operator().is_delete())
+        {
+            return None;
+        }
+        Some(ctx.ast.void_0(ident.span))
+    }
 }
 
 #[cfg(test)]
diff --git a/crates/oxc_minifier/src/peephole/substitute_alternate_syntax.rs b/crates/oxc_minifier/src/peephole/substitute_alternate_syntax.rs
@@ -871,7 +871,6 @@ impl<'a> LatePeepholeOptimizations {
         }
 
         if let Some(folded_expr) = match expr {
-            Expression::Identifier(ident) => Self::try_compress_undefined(ident, ctx),
             Expression::BooleanLiteral(_) => Self::try_compress_boolean(expr, ctx),
             _ => None,
         } {
@@ -893,23 +892,6 @@ impl<'a> LatePeepholeOptimizations {
         }
     }
 
-    /// Transforms `undefined` => `void 0`
-    fn try_compress_undefined(
-        ident: &IdentifierReference<'a>,
-        ctx: &mut TraverseCtx<'a>,
-    ) -> Option<Expression<'a>> {
-        if !Ctx(ctx).is_identifier_undefined(ident) {
-            return None;
-        }
-        // `delete undefined` returns `false`
-        // `delete void 0` returns `true`
-        if matches!(ctx.parent(), Ancestor::UnaryExpressionArgument(e) if e.operator().is_delete())
-        {
-            return None;
-        }
-        Some(ctx.ast.void_0(ident.span))
-    }
-
     /// Transforms boolean expression `true` => `!0` `false` => `!1`.
     fn try_compress_boolean(
         expr: &mut Expression<'a>,
diff --git a/tasks/minsize/minsize.snap b/tasks/minsize/minsize.snap
@@ -9,7 +9,7 @@ Original   | minified   | minified   | gzip       | gzip       | Fixture
 
 342.15 kB  | 118.19 kB  | 118.14 kB  | 44.45 kB   | 44.37 kB   | vue.js    
 
-544.10 kB  | 71.73 kB   | 72.48 kB   | 26.14 kB   | 26.20 kB   | lodash.js 
+544.10 kB  | 71.75 kB   | 72.48 kB   | 26.15 kB   | 26.20 kB   | lodash.js 
 
 555.77 kB  | 272.89 kB  | 270.13 kB  | 90.90 kB   | 90.80 kB   | d3.js     
 

Original file line number	Diff line number	Diff line change
`@@ -62,20 +62,23 @@ impl<'a> Ctx<'a, '_> {`
`62`	`62`	`}`
`63`	`63`	`}`
`64`	`64`
	`65`	`+ #[inline]`
`65`	`66`	`pub fn is_identifier_undefined(self, ident: &IdentifierReference) -> bool {`
`66`	`67`	`if ident.name == "undefined" && ident.is_global_reference(self.symbols()) {`
`67`	`68`	`return true;`
`68`	`69`	`}`
`69`	`70`	`false`
`70`	`71`	`}`
`71`	`72`
	`73`	`+ #[inline]`
`72`	`74`	`pub fn is_identifier_infinity(self, ident: &IdentifierReference) -> bool {`
`73`	`75`	`if ident.name == "Infinity" && ident.is_global_reference(self.symbols()) {`
`74`	`76`	`return true;`
`75`	`77`	`}`
`76`	`78`	`false`
`77`	`79`	`}`
`78`	`80`
	`81`	`+ #[inline]`
`79`	`82`	`pub fn is_identifier_nan(self, ident: &IdentifierReference) -> bool {`
`80`	`83`	`if ident.name == "NaN" && ident.is_global_reference(self.symbols()) {`
`81`	`84`	`return true;`