feat(ast_codegen): add alignment and size data to the schema.

Author: rzvxa

Cargo.lock

@@ -1,4 +1,5 @@
 #![allow(clippy::wildcard_imports)]
+#![allow(clippy::inconsistent_struct_constructor)]
 // TODO: I'm not sure if it is a but or intentional but clippy needs this allowed both on this
 // module and the generated one.
 #![allow(clippy::self_named_module_files)]
@@ -24,7 +25,6 @@
 //! [`AssignmentTarget`]: ast::AssignmentTarget
 //! [`oxc_parser`]: <https://docs.rs/oxc_parser>
 //! [`Parser`]: <https://docs.rs/oxc_parser/latest/oxc_parser/struct.Parser.html>
-
 #[cfg(feature = "serialize")]
 mod serialize;
 
@@ -39,6 +39,8 @@ mod trivia;
 mod generated {
     #[cfg(test)]
     pub mod assert_layouts;
+    // TODO: remove me; don't merge to upstream!!
+    pub mod assert_repr_rust_layout;
     pub mod ast_builder;
     pub mod ast_kind;
     pub mod span;

crates/oxc_ast/src/generated/assert_layouts.rs

@@ -24,3 +24,5 @@ doctest    = false
 quote = { workspace = true }
 syn = { workspace = true, features = ["full"] }
 proc-macro2 = { workspace = true }
+lazy_static = { workspace = true }
+rustc-hash  = { workspace = true }

crates/oxc_ast/src/generated/assert_repr_rust_layout.rs

@@ -1,6 +1,12 @@
+mod reorder_fields;
+mod generated {
+    pub mod ast_field_order_data;
+}
+
 use proc_macro::TokenStream;
 use proc_macro2::TokenStream as TokenStream2;
 use quote::quote;
+use reorder_fields::reorder_fields;
 
 /// returns `#[repr(C, u8)]` if `enum_` has any non-unit variant,
 /// Otherwise it would return `#[repr(u8)]`.
@@ -25,10 +31,16 @@ fn enum_repr(enum_: &syn::ItemEnum) -> TokenStream2 {
 #[allow(clippy::missing_panics_doc)]
 pub fn ast(_args: TokenStream, input: TokenStream) -> TokenStream {
     let input = syn::parse_macro_input!(input as syn::Item);
-
-    let repr = match input {
-        syn::Item::Enum(ref enum_) => enum_repr(enum_),
-        syn::Item::Struct(_) => quote!(#[repr(C)]),
+    let (repr, item) = match input {
+        syn::Item::Enum(enum_) => (enum_repr(&enum_), syn::Item::Enum(enum_)),
+        syn::Item::Struct(mut struct_) => {
+            let id = struct_.ident.to_string();
+            // if we have field ordering data for this type use it to reorder.
+            if let Some(data) = generated::ast_field_order_data::get(id.as_str()) {
+                reorder_fields(&mut struct_, data);
+            };
+            (quote!(#[repr(C)]), syn::Item::Struct(struct_))
+        }
 
         _ => {
             unreachable!()
@@ -38,7 +50,7 @@ pub fn ast(_args: TokenStream, input: TokenStream) -> TokenStream {
     let expanded = quote! {
         #[derive(::oxc_ast_macros::Ast)]
         #repr
-        #input
+        #item
     };
     TokenStream::from(expanded)
 }

crates/oxc_ast/src/lib.rs

@@ -0,0 +1,58 @@
+use proc_macro2::TokenStream;
+use quote::format_ident;
+use syn::{Field, Fields, FieldsNamed, FieldsUnnamed, ItemStruct};
+
+const DUMMY: &str = "DUMMY";
+
+#[inline]
+pub fn reorder_fields(ty: &mut ItemStruct, data: &[u8]) {
+    let (Fields::Named(FieldsNamed { named: fields, .. })
+    | Fields::Unnamed(FieldsUnnamed { unnamed: fields, .. })) = &mut ty.fields
+    else {
+        debug_assert!(false, "Entered unreachable code!");
+        // SAFETY: We don't generate any ordering data for empty structs, And the debug assertions
+        // are on in CI runs; The debug assertion above would ensure any possible mistake gets caught
+        // by tests early on in the PR's life span. This allows us to avoid a branch here.
+        unsafe { std::hint::unreachable_unchecked() }
+    };
+
+    // TODO: We can replace this with uninitialized memory, It might be faster if we use one byte
+    // to check if a field is placeholder or not and keep the rest of the bytes uninitialized as we
+    // never read them. I'm not sure if it is safe with a mutable reference or not but I believe it
+    // would be safe with exclusive ownership of the field.
+    let mut pick = Field {
+        attrs: Vec::default(),
+        vis: syn::Visibility::Inherited,
+        mutability: syn::FieldMutability::None,
+        ident: Some(format_ident!("{DUMMY}")),
+        colon_token: None,
+        ty: syn::Type::Verbatim(TokenStream::default()),
+    };
+    // TODO: use bit array here?
+    let mut is_ordered = vec![false; fields.len()];
+    let mut target;
+    // Best case O(n), Worst case O(2n)
+    for i in 0..fields.len() {
+        if is_ordered[i] {
+            continue;
+        }
+
+        let field = &mut fields[i];
+        // `pick` the first unordered field
+        pick = std::mem::replace(field, pick);
+        // capture its ideal position
+        target = data[i];
+
+        // while we have something in our `pick`
+        while pick.ident.as_ref().is_some_and(|it| it != DUMMY) {
+            // select the slot of target position
+            let field = &mut fields[target as usize];
+            // put the picked field in the target slot and pick the previous item
+            pick = std::mem::replace(field, pick);
+            // mark the field as ordered
+            is_ordered[target as usize] = true;
+            // capture the ideal position of our new pick
+            target = data[target as usize];
+        }
+    }
+}

crates/oxc_ast_macros/Cargo.toml

@@ -0,0 +1,90 @@
+use itertools::Itertools;
+use proc_macro2::TokenStream;
+use quote::quote;
+use syn::ItemStruct;
+
+use crate::{layout::Layout, output, schema::RType, CodegenCtx, Generator, GeneratorOutput};
+
+use super::{define_generator, generated_header};
+
+define_generator! {
+    pub struct AstFieldOrder;
+}
+
+impl Generator for AstFieldOrder {
+    fn name(&self) -> &'static str {
+        stringify!(AstFieldOrder)
+    }
+
+    fn generate(&mut self, ctx: &CodegenCtx) -> GeneratorOutput {
+        let orders_64 = ctx
+            .ty_table
+            .iter()
+            .filter(|ty| matches!(&*ty.borrow(), RType::Struct(s) if !s.item.fields.is_empty()))
+            .map(|ty| {
+                let RType::Struct(ty) = &*ty.borrow() else { unreachable!() };
+                generate_orders(&ty.item, &ty.meta.layout_64)
+            });
+        let orders_32 = ctx
+            .ty_table
+            .iter()
+            .filter(|ty| matches!(&*ty.borrow(), RType::Struct(s) if !s.item.fields.is_empty()))
+            .map(|ty| {
+                let RType::Struct(ty) = &*ty.borrow() else { unreachable!() };
+                generate_orders(&ty.item, &ty.meta.layout_64)
+            });
+        let header = generated_header!();
+        GeneratorOutput::Stream((
+            output(crate::AST_MACROS_CRATE, "ast_field_order_data.rs"),
+            quote! {
+                #header
+                use lazy_static::lazy_static;
+                use rustc_hash::FxHashMap;
+
+                endl!();
+
+                pub fn get(ident: &str) -> Option<&[u8]> {
+
+                    #[cfg(not(any(target_pointer_width = "64", target_pointer_width = "32")))]
+                    std::compile_error!(
+                        "Platforms with pointer width other than 64 or 32 bit are not supported"
+                    );
+                    #[cfg(target_pointer_width = "64")]
+                    lazy_static! {
+                        static ref DATA: FxHashMap<&'static str, &'static [u8]> =
+                            FxHashMap::from_iter([#(#orders_64),*]);
+                    }
+                    #[cfg(target_pointer_width = "32")]
+                    lazy_static! {
+                        static ref DATA: FxHashMap<&'static str, &'static [u8]> =
+                            FxHashMap::from_iter([#(#orders_32),*]);
+                    }
+
+
+                    DATA.get(ident).copied()
+                }
+            },
+        ))
+    }
+}
+
+fn generate_orders(ty: &ItemStruct, layout: &Layout) -> Option<TokenStream> {
+    let ident = &ty.ident.to_string();
+    let Layout::Layout(layout) = layout else { panic!("Layout isn't determined yet!") };
+    let offsets = layout.offsets();
+    if let Some(offsets) = offsets {
+        let orders = offsets
+            .iter()
+            .zip(ty.fields.iter().enumerate())
+            .sorted_by(|a, b| Ord::cmp(a.0, b.0))
+            .map(|(_, fi)| fi)
+            .enumerate()
+            .sorted_by(|a, b| Ord::cmp(&a.1 .0, &b.1 .0))
+            .map(|it| {
+                u8::try_from(it.0).expect("We have no AST type with enough fields to exhaust `u8`.")
+            });
+        Some(quote!((#ident, &[#(#orders),*][..])))
+    } else {
+        None
+    }
+}

crates/oxc_ast_macros/src/generated/ast_field_order_data.rs

@@ -1,5 +1,6 @@
 mod assert_layouts;
 mod ast_builder;
+mod ast_field_order;
 mod ast_kind;
 mod impl_get_span;
 mod visit;
@@ -42,6 +43,7 @@ pub(crate) use insert;
 
 pub use assert_layouts::AssertLayouts;
 pub use ast_builder::AstBuilderGenerator;
+pub use ast_field_order::AstFieldOrder;
 pub use ast_kind::AstKindGenerator;
 pub use impl_get_span::ImplGetSpanGenerator;
 pub use visit::{VisitGenerator, VisitMutGenerator};

crates/oxc_ast_macros/src/lib.rs

@@ -1,5 +1,4 @@
 const AST_CRATE: &str = "crates/oxc_ast";
-#[allow(dead_code)]
 const AST_MACROS_CRATE: &str = "crates/oxc_ast_macros";
 
 mod defs;
@@ -22,7 +21,7 @@ use syn::parse_file;
 
 use defs::TypeDef;
 use generators::{
-    AssertLayouts, AstBuilderGenerator, AstKindGenerator, Generator, VisitGenerator,
+    AssertLayouts, AstBuilderGenerator, AstFieldOrder, AstKindGenerator, Generator, VisitGenerator,
     VisitMutGenerator,
 };
 use schema::{Module, REnum, RStruct, RType, Schema};
@@ -265,6 +264,7 @@ fn main() -> std::result::Result<(), Box<dyn std::error::Error>> {
         .pass(CalcLayout)
         .pass(BuildSchema)
         .gen(AssertLayouts)
+        .gen(AstFieldOrder)
         .gen(AstKindGenerator)
         .gen(AstBuilderGenerator)
         .gen(ImplGetSpanGenerator)

crates/oxc_ast_macros/src/reorder_fields.rs

@@ -164,7 +164,16 @@ fn calc_struct_layout(ty: &mut RStruct, ctx: &CodegenCtx) -> Result<PlatformLayo
     fn with_padding(
         layouts: &[KnownLayout],
     ) -> std::result::Result<KnownLayout, std::alloc::LayoutError> {
-        let layouts = layouts.iter().enumerate();
+        // reorder fields
+        let layouts = layouts.iter().enumerate().sorted_by(|a, b| {
+            let (ia, a) = a;
+            let (ib, b) = b;
+            if b.size() == a.size() {
+                Ord::cmp(&ia, &ib)
+            } else {
+                Ord::cmp(&b.size(), &a.size())
+            }
+        });
         let mut offsets = vec![0; layouts.len()];
         let mut output = std::alloc::Layout::from_size_align(0, 1)?;
         let mut niches = 0;