padlock-source 0.4.0

// padlock-source/src/frontends/rust.rs
//
// Extracts struct layouts from Rust source using syn + the Visit API.
// Sizes are approximated from type names using the target arch config.
// Only repr(C) / repr(packed) / plain structs are handled; generics are opaque.

use padlock_core::arch::ArchConfig;
use padlock_core::ir::{AccessPattern, Field, StructLayout, TypeInfo};
use quote::ToTokens;
use syn::{visit::Visit, Fields, ItemStruct, Type};

// ── attribute guard extraction ────────────────────────────────────────────────

/// Extract a lock guard name from field attributes.
///
/// Recognised forms:
/// - `#[lock_protected_by = "mu"]`
/// - `#[protected_by = "mu"]`
/// - `#[guarded_by("mu")]` or `#[guarded_by(mu)]`
/// - `#[pt_guarded_by("mu")]` or `#[pt_guarded_by(mu)]` (pointer variant)
pub fn extract_guard_from_attrs(attrs: &[syn::Attribute]) -> Option<String> {
    for attr in attrs {
        let path = attr.path();
        // Name-value form: #[lock_protected_by = "mu"] / #[protected_by = "mu"]
        if path.is_ident("lock_protected_by") || path.is_ident("protected_by") {
            if let syn::Meta::NameValue(nv) = &attr.meta {
                if let syn::Expr::Lit(syn::ExprLit {
                    lit: syn::Lit::Str(s),
                    ..
                }) = &nv.value
                {
                    return Some(s.value());
                }
            }
        }
        // List form: #[guarded_by("mu")] / #[guarded_by(mu)] / #[pt_guarded_by(...)]
        if path.is_ident("guarded_by") || path.is_ident("pt_guarded_by") {
            // Try string literal first
            if let Ok(s) = attr.parse_args::<syn::LitStr>() {
                return Some(s.value());
            }
            // Fall back to bare identifier
            if let Ok(id) = attr.parse_args::<syn::Ident>() {
                return Some(id.to_string());
            }
        }
    }
    None
}

// ── type resolution ───────────────────────────────────────────────────────────

fn rust_type_size_align(ty: &Type, arch: &'static ArchConfig) -> (usize, usize, TypeInfo) {
    match ty {
        Type::Path(tp) => {
            let name = tp
                .path
                .segments
                .last()
                .map(|s| s.ident.to_string())
                .unwrap_or_default();
            let (size, align) = primitive_size_align(&name, arch);
            (size, align, TypeInfo::Primitive { name, size, align })
        }
        Type::Ptr(_) | Type::Reference(_) => {
            let s = arch.pointer_size;
            (s, s, TypeInfo::Pointer { size: s, align: s })
        }
        Type::Array(arr) => {
            let (elem_size, elem_align, elem_ty) = rust_type_size_align(&arr.elem, arch);
            let count = array_len_from_expr(&arr.len);
            let size = elem_size * count;
            (
                size,
                elem_align,
                TypeInfo::Array {
                    element: Box::new(elem_ty),
                    count,
                    size,
                    align: elem_align,
                },
            )
        }
        _ => {
            let s = arch.pointer_size;
            (
                s,
                s,
                TypeInfo::Opaque {
                    name: "(unknown)".into(),
                    size: s,
                    align: s,
                },
            )
        }
    }
}

fn primitive_size_align(name: &str, arch: &'static ArchConfig) -> (usize, usize) {
    match name {
        "bool" | "u8" | "i8" => (1, 1),
        "u16" | "i16" => (2, 2),
        "u32" | "i32" | "f32" => (4, 4),
        "u64" | "i64" | "f64" => (8, 8),
        "u128" | "i128" => (16, 16),
        "usize" | "isize" => (arch.pointer_size, arch.pointer_size),
        "char" => (4, 4), // Rust char is a Unicode scalar (4 bytes)
        // x86 SSE / AVX / AVX-512 SIMD types (via std::arch or target_feature)
        "__m64" => (8, 8),
        "__m128" | "__m128d" | "__m128i" => (16, 16),
        "__m256" | "__m256d" | "__m256i" => (32, 32),
        "__m512" | "__m512d" | "__m512i" => (64, 64),
        // Rust portable SIMD / packed types (std::simd, packed_simd)
        "f32x4" | "i32x4" | "u32x4" => (16, 16),
        "f64x2" | "i64x2" | "u64x2" => (16, 16),
        "f32x8" | "i32x8" | "u32x8" => (32, 32),
        "f64x4" | "i64x4" | "u64x4" => (32, 32),
        "f32x16" | "i32x16" | "u32x16" => (64, 64),
        _ => (arch.pointer_size, arch.pointer_size),
    }
}

fn array_len_from_expr(expr: &syn::Expr) -> usize {
    if let syn::Expr::Lit(syn::ExprLit {
        lit: syn::Lit::Int(n),
        ..
    }) = expr
    {
        n.base10_parse::<usize>().unwrap_or(0)
    } else {
        0
    }
}

// ── struct repr detection ─────────────────────────────────────────────────────

fn is_packed(attrs: &[syn::Attribute]) -> bool {
    attrs
        .iter()
        .any(|a| a.path().is_ident("repr") && a.to_token_stream().to_string().contains("packed"))
}

fn simulate_rust_layout(
    name: String,
    fields: &[(String, Type)],
    packed: bool,
    arch: &'static ArchConfig,
) -> StructLayout {
    let mut offset = 0usize;
    let mut struct_align = 1usize;
    let mut out_fields: Vec<Field> = Vec::new();

    for (fname, ty) in fields {
        let (size, align, type_info) = rust_type_size_align(ty, arch);
        let effective_align = if packed { 1 } else { align };

        if effective_align > 0 {
            offset = offset.next_multiple_of(effective_align);
        }
        struct_align = struct_align.max(effective_align);

        out_fields.push(Field {
            name: fname.clone(),
            ty: type_info,
            offset,
            size,
            align: effective_align,
            source_file: None,
            source_line: None,
            access: AccessPattern::Unknown,
        });
        offset += size;
    }

    if !packed && struct_align > 0 {
        offset = offset.next_multiple_of(struct_align);
    }

    StructLayout {
        name,
        total_size: offset,
        align: struct_align,
        fields: out_fields,
        source_file: None,
        source_line: None,
        arch,
        is_packed: packed,
        is_union: false,
    }
}

// ── visitor ───────────────────────────────────────────────────────────────────

struct StructVisitor {
    arch: &'static ArchConfig,
    layouts: Vec<StructLayout>,
}

impl<'ast> Visit<'ast> for StructVisitor {
    fn visit_item_struct(&mut self, node: &'ast ItemStruct) {
        syn::visit::visit_item_struct(self, node); // recurse into nested items

        let name = node.ident.to_string();
        let packed = is_packed(&node.attrs);

        // Collect (field_name, type, optional_guard)
        let fields: Vec<(String, Type, Option<String>)> = match &node.fields {
            Fields::Named(nf) => nf
                .named
                .iter()
                .map(|f| {
                    let fname = f.ident.as_ref().map(|i| i.to_string()).unwrap_or_default();
                    let guard = extract_guard_from_attrs(&f.attrs);
                    (fname, f.ty.clone(), guard)
                })
                .collect(),
            Fields::Unnamed(uf) => uf
                .unnamed
                .iter()
                .enumerate()
                .map(|(i, f)| {
                    let guard = extract_guard_from_attrs(&f.attrs);
                    (format!("_{i}"), f.ty.clone(), guard)
                })
                .collect(),
            Fields::Unit => vec![],
        };

        let name_ty: Vec<(String, Type)> = fields
            .iter()
            .map(|(n, t, _)| (n.clone(), t.clone()))
            .collect();
        let mut layout = simulate_rust_layout(name, &name_ty, packed, self.arch);
        layout.source_line = Some(node.ident.span().start().line as u32);

        // Apply explicit guard annotations; these take precedence over the
        // heuristic type-name pass in concurrency.rs (which skips non-Unknown fields).
        for (i, (_, _, guard)) in fields.iter().enumerate() {
            if let Some(g) = guard {
                layout.fields[i].access = AccessPattern::Concurrent {
                    guard: Some(g.clone()),
                    is_atomic: false,
                };
            }
        }

        self.layouts.push(layout);
    }
}

// ── public API ────────────────────────────────────────────────────────────────

pub fn parse_rust(source: &str, arch: &'static ArchConfig) -> anyhow::Result<Vec<StructLayout>> {
    let file: syn::File = syn::parse_str(source)?;
    let mut visitor = StructVisitor {
        arch,
        layouts: Vec::new(),
    };
    visitor.visit_file(&file);
    Ok(visitor.layouts)
}

// ── tests ─────────────────────────────────────────────────────────────────────

#[cfg(test)]
mod tests {
    use super::*;
    use padlock_core::arch::X86_64_SYSV;

    #[test]
    fn parse_simple_struct() {
        let src = "struct Foo { a: u8, b: u64, c: u32 }";
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        assert_eq!(layouts.len(), 1);
        let l = &layouts[0];
        assert_eq!(l.name, "Foo");
        assert_eq!(l.fields.len(), 3);
        assert_eq!(l.fields[0].size, 1); // u8
        assert_eq!(l.fields[1].size, 8); // u64
        assert_eq!(l.fields[2].size, 4); // u32
    }

    #[test]
    fn layout_includes_padding() {
        // u8 then u64: 7 bytes padding inserted
        let src = "struct T { a: u8, b: u64 }";
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        let l = &layouts[0];
        assert_eq!(l.fields[0].offset, 0);
        assert_eq!(l.fields[1].offset, 8); // u64 aligned to 8
        assert_eq!(l.total_size, 16);
        let gaps = padlock_core::ir::find_padding(l);
        assert_eq!(gaps[0].bytes, 7);
    }

    #[test]
    fn multiple_structs_parsed() {
        let src = "struct A { x: u32 } struct B { y: u64 }";
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        assert_eq!(layouts.len(), 2);
    }

    #[test]
    fn packed_struct_no_padding() {
        let src = "#[repr(packed)] struct P { a: u8, b: u64 }";
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        let l = &layouts[0];
        assert!(l.is_packed);
        assert_eq!(l.fields[1].offset, 1); // no padding, b immediately after a
        let gaps = padlock_core::ir::find_padding(l);
        assert!(gaps.is_empty());
    }

    #[test]
    fn pointer_field_uses_arch_size() {
        let src = "struct S { p: *const u8 }";
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        assert_eq!(layouts[0].fields[0].size, 8); // 64-bit pointer
    }

    // ── attribute guard extraction ─────────────────────────────────────────────

    #[test]
    fn lock_protected_by_attr_sets_guard() {
        let src = r#"
struct Cache {
    #[lock_protected_by = "mu"]
    readers: u64,
    mu: u64,
}
"#;
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        let readers = &layouts[0].fields[0];
        assert_eq!(readers.name, "readers");
        if let AccessPattern::Concurrent { guard, .. } = &readers.access {
            assert_eq!(guard.as_deref(), Some("mu"));
        } else {
            panic!("expected Concurrent, got {:?}", readers.access);
        }
    }

    #[test]
    fn guarded_by_string_attr_sets_guard() {
        let src = r#"
struct S {
    #[guarded_by("lock")]
    value: u32,
}
"#;
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        if let AccessPattern::Concurrent { guard, .. } = &layouts[0].fields[0].access {
            assert_eq!(guard.as_deref(), Some("lock"));
        } else {
            panic!("expected Concurrent");
        }
    }

    #[test]
    fn guarded_by_ident_attr_sets_guard() {
        let src = r#"
struct S {
    #[guarded_by(mu)]
    count: u64,
}
"#;
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        if let AccessPattern::Concurrent { guard, .. } = &layouts[0].fields[0].access {
            assert_eq!(guard.as_deref(), Some("mu"));
        } else {
            panic!("expected Concurrent");
        }
    }

    #[test]
    fn protected_by_attr_sets_guard() {
        let src = r#"
struct S {
    #[protected_by = "lock_a"]
    x: u64,
}
"#;
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        if let AccessPattern::Concurrent { guard, .. } = &layouts[0].fields[0].access {
            assert_eq!(guard.as_deref(), Some("lock_a"));
        } else {
            panic!("expected Concurrent");
        }
    }

    #[test]
    fn different_guards_on_same_cache_line_is_false_sharing() {
        // readers and writers are at offsets 0 and 8 — same cache line (line 0).
        // They have different explicit guards → confirmed false sharing.
        let src = r#"
struct HotPath {
    #[lock_protected_by = "mu_a"]
    readers: u64,
    #[lock_protected_by = "mu_b"]
    writers: u64,
}
"#;
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        assert!(padlock_core::analysis::false_sharing::has_false_sharing(
            &layouts[0]
        ));
    }

    #[test]
    fn same_guard_on_same_cache_line_is_not_false_sharing() {
        let src = r#"
struct Safe {
    #[lock_protected_by = "mu"]
    a: u64,
    #[lock_protected_by = "mu"]
    b: u64,
}
"#;
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        assert!(!padlock_core::analysis::false_sharing::has_false_sharing(
            &layouts[0]
        ));
    }

    #[test]
    fn unannotated_field_stays_unknown() {
        let src = "struct S { x: u64 }";
        let layouts = parse_rust(src, &X86_64_SYSV).unwrap();
        assert!(matches!(
            layouts[0].fields[0].access,
            AccessPattern::Unknown
        ));
    }
}