stage2: Implement explicit backing integers for packed structs

Now the backing integer of a packed struct type may be explicitly
specified with e.g. `packed struct(u32) { ... }`.

Author: ifreund

lib/std/builtin.zig

@@ -294,6 +294,8 @@ pub const Type = union(enum) {
     /// therefore must be kept in sync with the compiler implementation.
     pub const Struct = struct {
         layout: ContainerLayout,
+        /// Only valid if layout is .Packed
+        backing_integer: ?type = null,
         fields: []const StructField,
         decls: []const Declaration,
         is_tuple: bool,

lib/std/zig/Ast.zig

@@ -2967,7 +2967,7 @@ pub const Node = struct {
         /// Same as ContainerDeclTwo except there is known to be a trailing comma
         /// or semicolon before the rbrace.
         container_decl_two_trailing,
-        /// `union(lhs)` / `enum(lhs)`. `SubRange[rhs]`.
+        /// `struct(lhs)` / `union(lhs)` / `enum(lhs)`. `SubRange[rhs]`.
         container_decl_arg,
         /// Same as container_decl_arg but there is known to be a trailing
         /// comma or semicolon before the rbrace.

lib/std/zig/parse.zig

@@ -3356,16 +3356,18 @@ const Parser = struct {
     }
 
     /// Caller must have already verified the first token.
+    /// ContainerDeclAuto <- ContainerDeclType LBRACE container_doc_comment? ContainerMembers RBRACE
+    ///
     /// ContainerDeclType
-    ///     <- KEYWORD_struct
+    ///     <- KEYWORD_struct (LPAREN Expr RPAREN)?
+    ///      / KEYWORD_opaque
     ///      / KEYWORD_enum (LPAREN Expr RPAREN)?
     ///      / KEYWORD_union (LPAREN (KEYWORD_enum (LPAREN Expr RPAREN)? / Expr) RPAREN)?
-    ///      / KEYWORD_opaque
     fn parseContainerDeclAuto(p: *Parser) !Node.Index {
         const main_token = p.nextToken();
         const arg_expr = switch (p.token_tags[main_token]) {
-            .keyword_struct, .keyword_opaque => null_node,
-            .keyword_enum => blk: {
+            .keyword_opaque => null_node,
+            .keyword_struct, .keyword_enum => blk: {
                 if (p.eatToken(.l_paren)) |_| {
                     const expr = try p.expectExpr();
                     _ = try p.expectToken(.r_paren);

lib/std/zig/parser_test.zig

@@ -3064,6 +3064,13 @@ test "zig fmt: struct declaration" {
         \\    c: u8,
         \\};
         \\
+        \\const Ps = packed struct(u32) {
+        \\    a: u1,
+        \\    b: u2,
+        \\
+        \\    c: u29,
+        \\};
+        \\
         \\const Es = extern struct {
         \\    a: u8,
         \\    b: u8,

src/AstGen.zig

@@ -152,6 +152,7 @@ pub fn generate(gpa: Allocator, tree: Ast) Allocator.Error!Zir {
         0,
         tree.containerDeclRoot(),
         .Auto,
+        0,
     )) |struct_decl_ref| {
         assert(refToIndex(struct_decl_ref).? == 0);
     } else |err| switch (err) {
@@ -4224,15 +4225,18 @@ fn structDeclInner(
     node: Ast.Node.Index,
     container_decl: Ast.full.ContainerDecl,
     layout: std.builtin.Type.ContainerLayout,
+    backing_int_node: Ast.Node.Index,
 ) InnerError!Zir.Inst.Ref {
     const decl_inst = try gz.reserveInstructionIndex();
 
-    if (container_decl.ast.members.len == 0) {
+    if (container_decl.ast.members.len == 0 and backing_int_node == 0) {
         try gz.setStruct(decl_inst, .{
             .src_node = node,
             .layout = layout,
             .fields_len = 0,
             .decls_len = 0,
+            .backing_int_ref = .none,
+            .backing_int_body_len = 0,
             .known_non_opv = false,
             .known_comptime_only = false,
         });
@@ -4267,6 +4271,34 @@ fn structDeclInner(
     };
     defer block_scope.unstack();
 
+    const scratch_top = astgen.scratch.items.len;
+    var backing_int_body_inst: []Zir.Inst.Index = &[0]Zir.Inst.Index{};
+    defer astgen.scratch.items.len = scratch_top;
+    const backing_int_ref: Zir.Inst.Ref = blk: {
+        if (backing_int_node != 0) {
+            if (layout != .Packed) {
+                return astgen.failNode(backing_int_node, "non-packed struct does not support backing integer type", .{});
+            } else {
+                const backing_int_ref = try typeExpr(&block_scope, &namespace.base, backing_int_node);
+                if (!block_scope.isEmpty()) {
+                    if (!block_scope.endsWithNoReturn()) {
+                        _ = try block_scope.addBreak(.break_inline, decl_inst, backing_int_ref);
+                    }
+
+                    const body = block_scope.instructionsSlice();
+                    const old_scratch_len = astgen.scratch.items.len;
+                    try astgen.scratch.ensureUnusedCapacity(gpa, countBodyLenAfterFixups(astgen, body));
+                    appendBodyWithFixupsArrayList(astgen, &astgen.scratch, body);
+                    backing_int_body_inst = astgen.scratch.items[old_scratch_len..];
+                    block_scope.instructions.items.len = block_scope.instructions_top;
+                }
+                break :blk backing_int_ref;
+            }
+        } else {
+            break :blk .none;
+        }
+    };
+
     const decl_count = try astgen.scanDecls(&namespace, container_decl.ast.members);
     const field_count = @intCast(u32, container_decl.ast.members.len - decl_count);
 
@@ -4379,6 +4411,8 @@ fn structDeclInner(
         .layout = layout,
         .fields_len = field_count,
         .decls_len = decl_count,
+        .backing_int_ref = backing_int_ref,
+        .backing_int_body_len = @intCast(u32, backing_int_body_inst.len),
         .known_non_opv = known_non_opv,
         .known_comptime_only = known_comptime_only,
     });
@@ -4387,7 +4421,9 @@ fn structDeclInner(
     const decls_slice = wip_members.declsSlice();
     const fields_slice = wip_members.fieldsSlice();
     const bodies_slice = astgen.scratch.items[bodies_start..];
-    try astgen.extra.ensureUnusedCapacity(gpa, decls_slice.len + fields_slice.len + bodies_slice.len);
+    try astgen.extra.ensureUnusedCapacity(gpa, backing_int_body_inst.len +
+        decls_slice.len + fields_slice.len + bodies_slice.len);
+    astgen.extra.appendSliceAssumeCapacity(backing_int_body_inst);
     astgen.extra.appendSliceAssumeCapacity(decls_slice);
     astgen.extra.appendSliceAssumeCapacity(fields_slice);
     astgen.extra.appendSliceAssumeCapacity(bodies_slice);
@@ -4583,9 +4619,7 @@ fn containerDecl(
                 else => unreachable,
             } else std.builtin.Type.ContainerLayout.Auto;
 
-            assert(container_decl.ast.arg == 0);
-
-            const result = try structDeclInner(gz, scope, node, container_decl, layout);
+            const result = try structDeclInner(gz, scope, node, container_decl, layout, container_decl.ast.arg);
             return rvalue(gz, rl, result, node);
         },
         .keyword_union => {
@@ -11234,14 +11268,16 @@ const GenZir = struct {
         src_node: Ast.Node.Index,
         fields_len: u32,
         decls_len: u32,
+        backing_int_ref: Zir.Inst.Ref,
+        backing_int_body_len: u32,
         layout: std.builtin.Type.ContainerLayout,
         known_non_opv: bool,
         known_comptime_only: bool,
     }) !void {
         const astgen = gz.astgen;
         const gpa = astgen.gpa;
 
-        try astgen.extra.ensureUnusedCapacity(gpa, 4);
+        try astgen.extra.ensureUnusedCapacity(gpa, 6);
         const payload_index = @intCast(u32, astgen.extra.items.len);
 
         if (args.src_node != 0) {
@@ -11254,6 +11290,12 @@ const GenZir = struct {
         if (args.decls_len != 0) {
             astgen.extra.appendAssumeCapacity(args.decls_len);
         }
+        if (args.backing_int_ref != .none) {
+            astgen.extra.appendAssumeCapacity(args.backing_int_body_len);
+            if (args.backing_int_body_len == 0) {
+                astgen.extra.appendAssumeCapacity(@enumToInt(args.backing_int_ref));
+            }
+        }
         astgen.instructions.set(inst, .{
             .tag = .extended,
             .data = .{ .extended = .{
@@ -11262,6 +11304,7 @@ const GenZir = struct {
                     .has_src_node = args.src_node != 0,
                     .has_fields_len = args.fields_len != 0,
                     .has_decls_len = args.decls_len != 0,
+                    .has_backing_int = args.backing_int_ref != .none,
                     .known_non_opv = args.known_non_opv,
                     .known_comptime_only = args.known_comptime_only,
                     .name_strategy = gz.anon_name_strategy,

src/Module.zig

@@ -895,6 +895,11 @@ pub const Struct = struct {
     zir_index: Zir.Inst.Index,
 
     layout: std.builtin.Type.ContainerLayout,
+    /// If the layout is not packed, this is the noreturn type.
+    /// If the layout is packed, this is the backing integer type of the packed struct.
+    /// Whether zig chooses this type or the user specifies it, it is stored here.
+    /// This will be set to the noreturn type until status is `have_layout`.
+    backing_int_ty: Type = Type.initTag(.noreturn),
     status: enum {
         none,
         field_types_wip,
@@ -1025,27 +1030,15 @@ pub const Struct = struct {
 
     pub fn packedFieldBitOffset(s: Struct, target: Target, index: usize) u16 {
         assert(s.layout == .Packed);
-        assert(s.haveFieldTypes());
+        assert(s.haveLayout());
         var bit_sum: u64 = 0;
         for (s.fields.values()) |field, i| {
             if (i == index) {
                 return @intCast(u16, bit_sum);
             }
             bit_sum += field.ty.bitSize(target);
         }
-        return @intCast(u16, bit_sum);
-    }
-
-    pub fn packedIntegerBits(s: Struct, target: Target) u16 {
-        return s.packedFieldBitOffset(target, s.fields.count());
-    }
-
-    pub fn packedIntegerType(s: Struct, target: Target, buf: *Type.Payload.Bits) Type {
-        buf.* = .{
-            .base = .{ .tag = .int_unsigned },
-            .data = s.packedIntegerBits(target),
-        };
-        return Type.initPayload(&buf.base);
+        unreachable; // index out of bounds
     }
 };