Commit `05947ea870`

Andrew Kelley <andrew@ziglang.org>

2022-03-30 07:19:06

stage2: implement `@intToError` with safety

This commit introduces a new AIR instruction `cmp_lt_errors_len`. It's specific to this use case for two reasons: * The total number of errors is not stable during semantic analysis; it can only be reliably checked when flush() is called. So the backend that is lowering the instruction must emit a relocation of some kind and then populate it during flush(). * The fewer AIR instructions in memory, the better for compiler performance, so we squish complex meanings into AIR tags without hesitation. The instruction is implemented only in the LLVM backend so far. It does this by creating a simple function which is gutted and re-populated with each flush(). AstGen now uses ResultLoc.coerced_ty for `@intToError` and Sema does the coercion.

master

1 parent 83617ea

Changed files (13)

src

arch

aarch64

arm

riscv64

wasm

x86_64

codegen

test

behavior

cast.zig

@@ -570,7 +570,9 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {
             .cmp_gte => try self.airCmp(inst, .gte),
             .cmp_gt  => try self.airCmp(inst, .gt),
             .cmp_neq => try self.airCmp(inst, .neq),
+
             .cmp_vector => try self.airCmpVector(inst),
+            .cmp_lt_errors_len => try self.airCmpLtErrorsLen(inst),
 
             .bool_and        => try self.airBinOp(inst),
             .bool_or         => try self.airBinOp(inst),
@@ -2660,6 +2662,14 @@ fn airCmpVector(self: *Self, inst: Air.Inst.Index) !void {
     return self.fail("TODO implement airCmpVector for {}", .{self.target.cpu.arch});
 }
 
+fn airCmpLtErrorsLen(self: *Self, inst: Air.Inst.Index) !void {
+    const un_op = self.air.instructions.items(.data)[inst].un_op;
+    const operand = try self.resolveInst(un_op);
+    _ = operand;
+    const result: MCValue = if (self.liveness.isUnused(inst)) .dead else return self.fail("TODO implement airCmpLtErrorsLen for {}", .{self.target.cpu.arch});
+    return self.finishAir(inst, result, .{ un_op, .none, .none });
+}
+
 fn airDbgStmt(self: *Self, inst: Air.Inst.Index) !void {
     const dbg_stmt = self.air.instructions.items(.data)[inst].dbg_stmt;

@@ -567,7 +567,9 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {
             .cmp_gte => try self.airCmp(inst, .gte),
             .cmp_gt  => try self.airCmp(inst, .gt),
             .cmp_neq => try self.airCmp(inst, .neq),
+
             .cmp_vector => try self.airCmpVector(inst),
+            .cmp_lt_errors_len => try self.airCmpLtErrorsLen(inst),
 
             .bool_and        => try self.airBinOp(inst),
             .bool_or         => try self.airBinOp(inst),
@@ -3063,6 +3065,14 @@ fn airCmpVector(self: *Self, inst: Air.Inst.Index) !void {
     return self.fail("TODO implement airCmpVector for {}", .{self.target.cpu.arch});
 }
 
+fn airCmpLtErrorsLen(self: *Self, inst: Air.Inst.Index) !void {
+    const un_op = self.air.instructions.items(.data)[inst].un_op;
+    const operand = try self.resolveInst(un_op);
+    _ = operand;
+    const result: MCValue = if (self.liveness.isUnused(inst)) .dead else return self.fail("TODO implement airCmpLtErrorsLen for {}", .{self.target.cpu.arch});
+    return self.finishAir(inst, result, .{ un_op, .none, .none });
+}
+
 fn airDbgStmt(self: *Self, inst: Air.Inst.Index) !void {
     const dbg_stmt = self.air.instructions.items(.data)[inst].dbg_stmt;

@@ -537,7 +537,9 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {
             .cmp_gte => try self.airCmp(inst, .gte),
             .cmp_gt  => try self.airCmp(inst, .gt),
             .cmp_neq => try self.airCmp(inst, .neq),
+
             .cmp_vector => try self.airCmpVector(inst),
+            .cmp_lt_errors_len => try self.airCmpLtErrorsLen(inst),
 
             .bool_and        => try self.airBoolOp(inst),
             .bool_or         => try self.airBoolOp(inst),
@@ -1799,6 +1801,14 @@ fn airCmpVector(self: *Self, inst: Air.Inst.Index) !void {
     return self.fail("TODO implement airCmpVector for {}", .{self.target.cpu.arch});
 }
 
+fn airCmpLtErrorsLen(self: *Self, inst: Air.Inst.Index) !void {
+    const un_op = self.air.instructions.items(.data)[inst].un_op;
+    const operand = try self.resolveInst(un_op);
+    _ = operand;
+    const result: MCValue = if (self.liveness.isUnused(inst)) .dead else return self.fail("TODO implement airCmpLtErrorsLen for {}", .{self.target.cpu.arch});
+    return self.finishAir(inst, result, .{ un_op, .none, .none });
+}
+
 fn airDbgStmt(self: *Self, inst: Air.Inst.Index) !void {
     const dbg_stmt = self.air.instructions.items(.data)[inst].dbg_stmt;

@@ -1319,7 +1319,9 @@ fn genInst(self: *Self, inst: Air.Inst.Index) !WValue {
         .cmp_lte => self.airCmp(inst, .lte),
         .cmp_lt => self.airCmp(inst, .lt),
         .cmp_neq => self.airCmp(inst, .neq),
+
         .cmp_vector => self.airCmpVector(inst),
+        .cmp_lt_errors_len => self.airCmpLtErrorsLen(inst),
 
         .array_elem_val => self.airArrayElemVal(inst),
         .array_to_slice => self.airArrayToSlice(inst),
@@ -2267,6 +2269,16 @@ fn airCmpVector(self: *Self, inst: Air.Inst.Index) InnerError!WValue {
     return self.fail("TODO implement airCmpVector for wasm", .{});
 }
 
+fn airCmpLtErrorsLen(self: *Self, inst: Air.Inst.Index) InnerError!WValue {
+    if (self.liveness.isUnused(inst)) return WValue{ .none = {} };
+
+    const un_op = self.air.instructions.items(.data)[inst].un_op;
+    const operand = try self.resolveInst(un_op);
+
+    _ = operand;
+    return self.fail("TODO implement airCmpLtErrorsLen for wasm", .{});
+}
+
 fn airBr(self: *Self, inst: Air.Inst.Index) InnerError!WValue {
     const br = self.air.instructions.items(.data)[inst].br;
     const block = self.blocks.get(br.block_inst).?;

@@ -693,7 +693,9 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {
             .cmp_gte => try self.airCmp(inst, .gte),
             .cmp_gt  => try self.airCmp(inst, .gt),
             .cmp_neq => try self.airCmp(inst, .neq),
+
             .cmp_vector => try self.airCmpVector(inst),
+            .cmp_lt_errors_len => try self.airCmpLtErrorsLen(inst),
 
             .bool_and        => try self.airBoolOp(inst),
             .bool_or         => try self.airBoolOp(inst),
@@ -3818,6 +3820,14 @@ fn airCmpVector(self: *Self, inst: Air.Inst.Index) !void {
     return self.fail("TODO implement airCmpVector for {}", .{self.target.cpu.arch});
 }
 
+fn airCmpLtErrorsLen(self: *Self, inst: Air.Inst.Index) !void {
+    const un_op = self.air.instructions.items(.data)[inst].un_op;
+    const operand = try self.resolveInst(un_op);
+    _ = operand;
+    const result: MCValue = if (self.liveness.isUnused(inst)) .dead else return self.fail("TODO implement airCmpLtErrorsLen for {}", .{self.target.cpu.arch});
+    return self.finishAir(inst, result, .{ un_op, .none, .none });
+}
+
 fn airDbgStmt(self: *Self, inst: Air.Inst.Index) !void {
     const dbg_stmt = self.air.instructions.items(.data)[inst].dbg_stmt;
     const payload = try self.addExtra(Mir.DbgLineColumn{

@@ -1767,7 +1767,8 @@ fn genBody(f: *Function, body: []const Air.Inst.Index) error{ AnalysisFail, OutO
             .cmp_eq  => try airEquality(f, inst, "((", "=="),
             .cmp_neq => try airEquality(f, inst, "!((", "!="),
 
-            .cmp_vector => return f.fail("TODO: C backend: implement binary op for tag '{s}'", .{@tagName(Air.Inst.Tag.cmp_vector)}),
+            .cmp_vector => return f.fail("TODO: C backend: implement cmp_vector", .{}),
+            .cmp_lt_errors_len => return f.fail("TODO: C backend: implement cmp_lt_errors_len", .{}),
 
             // bool_and and bool_or are non-short-circuit operations
             .bool_and        => try airBinOp(f, inst, " & "),

@@ -440,8 +440,38 @@ pub const Object = struct {
         error_name_table_ptr_global.setInitializer(error_name_table_ptr);
     }
 
+    fn genCmpLtErrorsLenFunction(object: *Object, comp: *Compilation) !void {
+        // If there is no such function in the module, it means the source code does not need it.
+        const llvm_fn = object.llvm_module.getNamedFunction(lt_errors_fn_name) orelse return;
+        const mod = comp.bin_file.options.module.?;
+        const errors_len = mod.global_error_set.count();
+
+        // Delete previous implementation. We replace it with every flush() because the
+        // total number of errors may have changed.
+        while (llvm_fn.getFirstBasicBlock()) |bb| {
+            bb.deleteBasicBlock();
+        }
+
+        const builder = object.context.createBuilder();
+
+        const entry_block = object.context.appendBasicBlock(llvm_fn, "Entry");
+        builder.positionBuilderAtEnd(entry_block);
+        builder.clearCurrentDebugLocation();
+
+        // Example source of the following LLVM IR:
+        // fn __zig_lt_errors_len(index: u16) bool {
+        //     return index < total_errors_len;
+        // }
+
+        const lhs = llvm_fn.getParam(0);
+        const rhs = lhs.typeOf().constInt(errors_len, .False);
+        const is_lt = builder.buildICmp(.ULT, lhs, rhs, "");
+        _ = builder.buildRet(is_lt);
+    }
+
     pub fn flushModule(self: *Object, comp: *Compilation) !void {
         try self.genErrorNameTable(comp);
+        try self.genCmpLtErrorsLenFunction(comp);
 
         if (self.di_builder) |dib| {
             // When lowering debug info for pointers, we emitted the element types as
@@ -3457,7 +3487,9 @@ pub const FuncGen = struct {
                 .cmp_lt  => try self.airCmp(inst, .lt),
                 .cmp_lte => try self.airCmp(inst, .lte),
                 .cmp_neq => try self.airCmp(inst, .neq),
+
                 .cmp_vector => try self.airCmpVector(inst),
+                .cmp_lt_errors_len => try self.airCmpLtErrorsLen(inst),
 
                 .is_non_null     => try self.airIsNonNull(inst, false, false, .NE),
                 .is_non_null_ptr => try self.airIsNonNull(inst, true , false, .NE),
@@ -3738,6 +3770,16 @@ pub const FuncGen = struct {
         return self.cmp(lhs, rhs, vec_ty, cmp_op);
     }
 
+    fn airCmpLtErrorsLen(self: *FuncGen, inst: Air.Inst.Index) !?*const llvm.Value {
+        if (self.liveness.isUnused(inst)) return null;
+
+        const un_op = self.air.instructions.items(.data)[inst].un_op;
+        const operand = try self.resolveInst(un_op);
+        const llvm_fn = try self.getCmpLtErrorsLenFunction();
+        const args: [1]*const llvm.Value = .{operand};
+        return self.builder.buildCall(llvm_fn, &args, args.len, .Fast, .Auto, "");
+    }
+
     fn cmp(
         self: *FuncGen,
         lhs: *const llvm.Value,
@@ -6392,6 +6434,25 @@ pub const FuncGen = struct {
         return fn_val;
     }
 
+    fn getCmpLtErrorsLenFunction(self: *FuncGen) !*const llvm.Value {
+        if (self.dg.object.llvm_module.getNamedFunction(lt_errors_fn_name)) |llvm_fn| {
+            return llvm_fn;
+        }
+
+        // Function signature: fn (anyerror) bool
+
+        const ret_llvm_ty = try self.dg.llvmType(Type.bool);
+        const anyerror_llvm_ty = try self.dg.llvmType(Type.anyerror);
+        const param_types = [_]*const llvm.Type{anyerror_llvm_ty};
+
+        const fn_type = llvm.functionType(ret_llvm_ty, &param_types, param_types.len, .False);
+        const llvm_fn = self.dg.object.llvm_module.addFunction(lt_errors_fn_name, fn_type);
+        llvm_fn.setLinkage(.Internal);
+        llvm_fn.setFunctionCallConv(.Fast);
+        self.dg.addCommonFnAttributes(llvm_fn);
+        return llvm_fn;
+    }
+
     fn airErrorName(self: *FuncGen, inst: Air.Inst.Index) !?*const llvm.Value {
         if (self.liveness.isUnused(inst)) return null;
 
@@ -7663,3 +7724,5 @@ const AnnotatedDITypePtr = enum(usize) {
         return @truncate(u1, @enumToInt(self)) != 0;
     }
 };
+
+const lt_errors_fn_name = "__zig_lt_errors_len";

@@ -637,6 +637,15 @@ pub const Inst = struct {
         /// Uses the `pl_op` field, payload represents the index of the target memory.
         wasm_memory_grow,
 
+        /// Returns `true` if and only if the operand, an integer with
+        /// the same size as the error integer type, is less than the
+        /// total number of errors in the Module.
+        /// Result type is always `bool`.
+        /// Uses the `un_op` field.
+        /// Note that the number of errors in the Module cannot be considered stable until
+        /// flush().
+        cmp_lt_errors_len,
+
         pub fn fromCmpOp(op: std.math.CompareOperator) Tag {
             return switch (op) {
                 .lt => .cmp_lt,
@@ -928,6 +937,7 @@ pub fn typeOfIndex(air: Air, inst: Air.Inst.Index) Type {
         .cmp_gte,
         .cmp_gt,
         .cmp_neq,
+        .cmp_lt_errors_len,
         .is_null,
         .is_non_null,
         .is_null_ptr,
@@ -936,9 +946,9 @@ pub fn typeOfIndex(air: Air, inst: Air.Inst.Index) Type {
         .is_non_err,
         .is_err_ptr,
         .is_non_err_ptr,
-        => return Type.initTag(.bool),
+        => return Type.bool,
 
-        .const_ty => return Type.initTag(.type),
+        .const_ty => return Type.type,
 
         .alloc,
         .ret_ptr,

@@ -7250,7 +7250,7 @@ fn builtinCall(
 
         .ptr_to_int            => return simpleUnOp(gz, scope, rl, node, .none,                               params[0], .ptr_to_int),
         .error_to_int          => return simpleUnOp(gz, scope, rl, node, .none,                               params[0], .error_to_int),
-        .int_to_error          => return simpleUnOp(gz, scope, rl, node, .{ .ty = .u16_type },                params[0], .int_to_error),
+        .int_to_error          => return simpleUnOp(gz, scope, rl, node, .{ .coerced_ty = .u16_type },        params[0], .int_to_error),
         .compile_error         => return simpleUnOp(gz, scope, rl, node, .{ .ty = .const_slice_u8_type },     params[0], .compile_error),
         .set_eval_branch_quota => return simpleUnOp(gz, scope, rl, node, .{ .coerced_ty = .u32_type },        params[0], .set_eval_branch_quota),
         .enum_to_int           => return simpleUnOp(gz, scope, rl, node, .none,                               params[0], .enum_to_int),

@@ -393,6 +393,7 @@ fn analyzeInst(
         .ceil,
         .round,
         .trunc_float,
+        .cmp_lt_errors_len,
         => {
             const operand = inst_datas[inst].un_op;
             return trackOperands(a, new_set, inst, main_tomb, .{ operand, .none, .none });

@@ -166,6 +166,7 @@ const Writer = struct {
             .ceil,
             .round,
             .trunc_float,
+            .cmp_lt_errors_len,
             => try w.writeUnOp(s, inst),
 
             .breakpoint,

@@ -5640,7 +5640,7 @@ fn zirErrorToInt(sema: *Sema, block: *Block, inst: Zir.Inst.Index) CompileError!
     const operand_src: LazySrcLoc = .{ .node_offset_builtin_call_arg0 = inst_data.src_node };
     const op = sema.resolveInst(inst_data.operand);
     const op_coerced = try sema.coerce(block, Type.anyerror, op, operand_src);
-    const result_ty = Type.initTag(.u16);
+    const result_ty = Type.u16;
 
     if (try sema.resolveMaybeUndefVal(block, src, op_coerced)) |val| {
         if (val.isUndef()) {
@@ -5665,32 +5665,31 @@ fn zirIntToError(sema: *Sema, block: *Block, inst: Zir.Inst.Index) CompileError!
     const inst_data = sema.code.instructions.items(.data)[inst].un_node;
     const src = inst_data.src();
     const operand_src: LazySrcLoc = .{ .node_offset_builtin_call_arg0 = inst_data.src_node };
-
-    const op = sema.resolveInst(inst_data.operand);
+    const uncasted_operand = sema.resolveInst(inst_data.operand);
+    const operand = try sema.coerce(block, Type.u16, uncasted_operand, operand_src);
     const target = sema.mod.getTarget();
 
-    if (try sema.resolveDefinedValue(block, operand_src, op)) |value| {
-        const int = value.toUnsignedInt(target);
+    if (try sema.resolveDefinedValue(block, operand_src, operand)) |value| {
+        const int = try sema.usizeCast(block, operand_src, value.toUnsignedInt(target));
         if (int > sema.mod.global_error_set.count() or int == 0)
             return sema.fail(block, operand_src, "integer value {d} represents no error", .{int});
         const payload = try sema.arena.create(Value.Payload.Error);
         payload.* = .{
             .base = .{ .tag = .@"error" },
-            .data = .{ .name = sema.mod.error_name_list.items[@intCast(usize, int)] },
+            .data = .{ .name = sema.mod.error_name_list.items[int] },
         };
         return sema.addConstant(Type.anyerror, Value.initPayload(&payload.base));
     }
     try sema.requireRuntimeBlock(block, src);
     if (block.wantSafety()) {
-        return sema.fail(block, src, "TODO: get max errors in compilation", .{});
-        // const is_gt_max = @panic("TODO get max errors in compilation");
-        // try sema.addSafetyCheck(block, is_gt_max, .invalid_error_code);
+        const is_lt_len = try block.addUnOp(.cmp_lt_errors_len, operand);
+        try sema.addSafetyCheck(block, is_lt_len, .invalid_error_code);
     }
     return block.addInst(.{
         .tag = .bitcast,
         .data = .{ .ty_op = .{
             .ty = Air.Inst.Ref.anyerror_type,
-            .operand = op,
+            .operand = operand,
         } },
     });
 }

@@ -360,7 +360,11 @@ test "expected [*c]const u8, found [*:0]const u8" {
 }
 
 test "explicit cast from integer to error type" {
-    if (builtin.zig_backend != .stage1) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_wasm) return error.SkipZigTest;
+    if (builtin.zig_backend == .stage2_x86_64) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_c) return error.SkipZigTest; // TODO
+    if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest;
+    if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; // TODO
 
     try testCastIntToErr(error.ItBroke);
     comptime try testCastIntToErr(error.ItBroke);

Commit 05947ea870

Commit `05947ea870`