/** * Boostrapper for math calls that may overflow * * @param lookup lookup * @param name name of operation * @param type method type * @param programPoint program point to bind to callsite * @return callsite for a math intrinsic node */ public static CallSite mathBootstrap( final Lookup lookup, final String name, final MethodType type, final int programPoint) { final MethodHandle mh; switch (name) { case "iadd": mh = JSType.ADD_EXACT.methodHandle(); break; case "isub": mh = JSType.SUB_EXACT.methodHandle(); break; case "imul": mh = JSType.MUL_EXACT.methodHandle(); break; case "idiv": mh = JSType.DIV_EXACT.methodHandle(); break; case "irem": mh = JSType.REM_EXACT.methodHandle(); break; case "ineg": mh = JSType.NEGATE_EXACT.methodHandle(); break; default: throw new AssertionError("unsupported math intrinsic"); } return new ConstantCallSite( MH.insertArguments(mh, mh.type().parameterCount() - 1, programPoint)); }
/** This class houses bootstrap method for invokedynamic instructions generated by compiler. */ public final class Bootstrap { /** Reference to the seed boostrap function */ public static final Call BOOTSTRAP = staticCallNoLookup( Bootstrap.class, "bootstrap", CallSite.class, Lookup.class, String.class, MethodType.class, int.class); private static final MethodHandleFunctionality MH = MethodHandleFactory.getFunctionality(); private static final MethodHandle VOID_TO_OBJECT = MH.constant(Object.class, ScriptRuntime.UNDEFINED); /** * The default dynalink relink threshold for megamorphism is 8. In the case of object fields only, * it is fine. However, with dual fields, in order to get performance on benchmarks with a lot of * object instantiation and then field reassignment, it can take slightly more relinks to become * stable with type changes swapping out an entire property map and making a map guard fail. Since * we need to set this value statically it must work with possibly changing optimistic types and * dual fields settings. A higher value does not seem to have any other negative performance * implication when running with object-only fields, so we choose a higher value here. * * <p>See for example octane.gbemu, run with --log=fields:warning to study megamorphic behavior */ private static final int NASHORN_DEFAULT_UNSTABLE_RELINK_THRESHOLD = 16; // do not create me!! private Bootstrap() {} private static final DynamicLinker dynamicLinker; static { final DynamicLinkerFactory factory = new DynamicLinkerFactory(); final NashornBeansLinker nashornBeansLinker = new NashornBeansLinker(); factory.setPrioritizedLinkers( new NashornLinker(), new NashornPrimitiveLinker(), new NashornStaticClassLinker(), new BoundCallableLinker(), new JavaSuperAdapterLinker(), new JSObjectLinker(nashornBeansLinker), new BrowserJSObjectLinker(nashornBeansLinker), new ReflectionCheckLinker()); factory.setFallbackLinkers(nashornBeansLinker, new NashornBottomLinker()); factory.setSyncOnRelink(true); factory.setPrelinkFilter( new GuardedInvocationFilter() { @Override public GuardedInvocation filter( final GuardedInvocation inv, final LinkRequest request, final LinkerServices linkerServices) { final CallSiteDescriptor desc = request.getCallSiteDescriptor(); return OptimisticReturnFilters.filterOptimisticReturnValue(inv, desc) .asType(linkerServices, desc.getMethodType()); } }); factory.setAutoConversionStrategy( new MethodTypeConversionStrategy() { @Override public MethodHandle asType(final MethodHandle target, final MethodType newType) { return unboxReturnType(target, newType); } }); factory.setInternalObjectsFilter(NashornBeansLinker.createHiddenObjectFilter()); final int relinkThreshold = Options.getIntProperty( "nashorn.unstable.relink.threshold", NASHORN_DEFAULT_UNSTABLE_RELINK_THRESHOLD); if (relinkThreshold > -1) { factory.setUnstableRelinkThreshold(relinkThreshold); } // Linkers for any additional language runtimes deployed alongside Nashorn will be picked up by // the factory. factory.setClassLoader(Bootstrap.class.getClassLoader()); dynamicLinker = factory.createLinker(); } /** * Returns if the given object is a "callable" * * @param obj object to be checked for callability * @return true if the obj is callable */ public static boolean isCallable(final Object obj) { if (obj == ScriptRuntime.UNDEFINED || obj == null) { return false; } return obj instanceof ScriptFunction || isJSObjectFunction(obj) || BeansLinker.isDynamicMethod(obj) || obj instanceof BoundCallable || isFunctionalInterfaceObject(obj) || obj instanceof StaticClass; } /** * Returns true if the given object is a strict callable * * @param callable the callable object to be checked for strictness * @return true if the obj is a strict callable, false if it is a non-strict callable. * @throws ECMAException with {@code TypeError} if the object is not a callable. */ public static boolean isStrictCallable(final Object callable) { if (callable instanceof ScriptFunction) { return ((ScriptFunction) callable).isStrict(); } else if (isJSObjectFunction(callable)) { return ((JSObject) callable).isStrictFunction(); } else if (callable instanceof BoundCallable) { return isStrictCallable(((BoundCallable) callable).getCallable()); } else if (BeansLinker.isDynamicMethod(callable) || callable instanceof StaticClass || isFunctionalInterfaceObject(callable)) { return false; } throw notFunction(callable); } private static ECMAException notFunction(final Object obj) { return typeError("not.a.function", ScriptRuntime.safeToString(obj)); } private static boolean isJSObjectFunction(final Object obj) { return obj instanceof JSObject && ((JSObject) obj).isFunction(); } /** * Returns if the given object is a dynalink Dynamic method * * @param obj object to be checked * @return true if the obj is a dynamic method */ public static boolean isDynamicMethod(final Object obj) { return BeansLinker.isDynamicMethod( obj instanceof BoundCallable ? ((BoundCallable) obj).getCallable() : obj); } /** * Returns if the given object is an instance of an interface annotated with * java.lang.FunctionalInterface * * @param obj object to be checked * @return true if the obj is an instance of @FunctionalInterface interface */ public static boolean isFunctionalInterfaceObject(final Object obj) { return !JSType.isPrimitive(obj) && (NashornBeansLinker.getFunctionalInterfaceMethodName(obj.getClass()) != null); } /** * Create a call site and link it for Nashorn. This version of the method conforms to the * invokedynamic bootstrap method expected signature and is referenced from Nashorn generated * bytecode as the bootstrap method for all invokedynamic instructions. * * @param lookup MethodHandle lookup. Ignored as Nashorn only uses public lookup. * @param opDesc Dynalink dynamic operation descriptor. * @param type Method type. * @param flags flags for call type, trace/profile etc. * @return CallSite with MethodHandle to appropriate method or null if not found. */ public static CallSite bootstrap( final Lookup lookup, final String opDesc, final MethodType type, final int flags) { return dynamicLinker.link(LinkerCallSite.newLinkerCallSite(lookup, opDesc, type, flags)); } /** * Boostrapper for math calls that may overflow * * @param lookup lookup * @param name name of operation * @param type method type * @param programPoint program point to bind to callsite * @return callsite for a math intrinsic node */ public static CallSite mathBootstrap( final Lookup lookup, final String name, final MethodType type, final int programPoint) { final MethodHandle mh; switch (name) { case "iadd": mh = JSType.ADD_EXACT.methodHandle(); break; case "isub": mh = JSType.SUB_EXACT.methodHandle(); break; case "imul": mh = JSType.MUL_EXACT.methodHandle(); break; case "idiv": mh = JSType.DIV_EXACT.methodHandle(); break; case "irem": mh = JSType.REM_EXACT.methodHandle(); break; case "ineg": mh = JSType.NEGATE_EXACT.methodHandle(); break; default: throw new AssertionError("unsupported math intrinsic"); } return new ConstantCallSite( MH.insertArguments(mh, mh.type().parameterCount() - 1, programPoint)); } /** * Returns a dynamic invoker for a specified dynamic operation using the public lookup. You can * use this method to create a method handle that when invoked acts completely as if it were a * Nashorn-linked call site. An overview of available dynamic operations can be found in the <a * href="https://github.com/szegedi/dynalink/wiki/User-Guide-0.6">Dynalink User Guide</a>, but * we'll show few examples here: * * <ul> * <li>Get a named property with fixed name: * <pre> * MethodHandle getColor = Boostrap.createDynamicInvoker("dyn:getProp:color", Object.class, Object.class); * Object obj = ...; // somehow obtain the object * Object color = getColor.invokeExact(obj); * </pre> * <li>Get a named property with variable name: * <pre> * MethodHandle getProperty = Boostrap.createDynamicInvoker("dyn:getElem", Object.class, Object.class, String.class); * Object obj = ...; // somehow obtain the object * Object color = getProperty.invokeExact(obj, "color"); * Object shape = getProperty.invokeExact(obj, "shape"); * MethodHandle getNumProperty = Boostrap.createDynamicInvoker("dyn:getElem", Object.class, Object.class, int.class); * Object elem42 = getNumProperty.invokeExact(obj, 42); * </pre> * <li>Set a named property with fixed name: * <pre> * MethodHandle setColor = Boostrap.createDynamicInvoker("dyn:setProp:color", void.class, Object.class, Object.class); * Object obj = ...; // somehow obtain the object * setColor.invokeExact(obj, Color.BLUE); * </pre> * <li>Set a property with variable name: * <pre> * MethodHandle setProperty = Boostrap.createDynamicInvoker("dyn:setElem", void.class, Object.class, String.class, Object.class); * Object obj = ...; // somehow obtain the object * setProperty.invokeExact(obj, "color", Color.BLUE); * setProperty.invokeExact(obj, "shape", Shape.CIRCLE); * </pre> * <li>Call a function on an object; two-step variant. This is the actual variant used by * Nashorn-generated code: * <pre> * MethodHandle findFooFunction = Boostrap.createDynamicInvoker("dyn:getMethod:foo", Object.class, Object.class); * Object obj = ...; // somehow obtain the object * Object foo_fn = findFooFunction.invokeExact(obj); * MethodHandle callFunctionWithTwoArgs = Boostrap.createDynamicInvoker("dyn:call", Object.class, Object.class, Object.class, Object.class, Object.class); * // Note: "call" operation takes a function, then a "this" value, then the arguments: * Object foo_retval = callFunctionWithTwoArgs.invokeExact(foo_fn, obj, arg1, arg2); * </pre> * <li>Call a function on an object; single-step variant. Although Nashorn doesn't use this * variant and never emits any INVOKEDYNAMIC instructions with {@code dyn:getMethod}, it * still supports this standard Dynalink operation: * <pre> * MethodHandle callFunctionFooWithTwoArgs = Boostrap.createDynamicInvoker("dyn:callMethod:foo", Object.class, Object.class, Object.class, Object.class); * Object obj = ...; // somehow obtain the object * Object foo_retval = callFunctionFooWithTwoArgs.invokeExact(obj, arg1, arg2); * </pre> * </ul> * * Few additional remarks: * * <ul> * <li>Just as Nashorn works with any Java object, the invokers returned from this method can * also be applied to arbitrary Java objects in addition to Nashorn JavaScript objects. * <li>For invoking a named function on an object, you can also use the {@link InvokeByName} * convenience class. * <li>For Nashorn objects {@code getElem}, {@code getProp}, and {@code getMethod} are handled * almost identically, since JavaScript doesn't distinguish between different kinds of * properties on an object. Either can be used with fixed property name or a variable * property name. The only significant difference is handling of missing properties: {@code * getMethod} for a missing member will link to a potential invocation of {@code * __noSuchMethod__} on the object, {@code getProp} for a missing member will link to a * potential invocation of {@code __noSuchProperty__}, while {@code getElem} for a missing * member will link to an empty getter. * <li>In similar vein, {@code setElem} and {@code setProp} are handled identically on Nashorn * objects. * <li>There's no rule that the variable property identifier has to be a {@code String} for * {@code getProp/setProp} and {@code int} for {@code getElem/setElem}. You can declare * their type to be {@code int}, {@code double}, {@code Object}, and so on regardless of the * kind of the operation. * <li>You can be as specific in parameter types as you want. E.g. if you know that the receiver * of the operation will always be {@code ScriptObject}, you can pass {@code * ScriptObject.class} as its parameter type. If you happen to link to a method that expects * different types, (you can use these invokers on POJOs too, after all, and end up linking * with their methods that have strongly-typed signatures), all necessary conversions * allowed by either Java or JavaScript will be applied: if invoked methods specify either * primitive or wrapped Java numeric types, or {@code String} or {@code boolean/Boolean}, * then the parameters might be subjected to standard ECMAScript {@code ToNumber}, {@code * ToString}, and {@code ToBoolean} conversion, respectively. Less obviously, if the * expected parameter type is a SAM type, and you pass a JavaScript function, a proxy object * implementing the SAM type and delegating to the function will be passed. Linkage can * often be optimized when linkers have more specific type information than "everything can * be an object". * <li>You can also be as specific in return types as you want. For return types any necessary * type conversion available in either Java or JavaScript will be automatically applied, * similar to the process described for parameters, only in reverse direction: if you * specify any either primitive or wrapped Java numeric type, or {@code String} or {@code * boolean/Boolean}, then the return values will be subjected to standard ECMAScript {@code * ToNumber}, {@code ToString}, and {@code ToBoolean} conversion, respectively. Less * obviously, if the return type is a SAM type, and the return value is a JavaScript * function, a proxy object implementing the SAM type and delegating to the function will be * returned. * </ul> * * @param opDesc Dynalink dynamic operation descriptor. * @param rtype the return type for the operation * @param ptypes the parameter types for the operation * @return MethodHandle for invoking the operation. */ public static MethodHandle createDynamicInvoker( final String opDesc, final Class<?> rtype, final Class<?>... ptypes) { return createDynamicInvoker(opDesc, MethodType.methodType(rtype, ptypes)); } /** * Returns a dynamic invoker for a specified dynamic operation using the public lookup. Similar to * {@link #createDynamicInvoker(String, Class, Class...)} but with an additional parameter to set * the call site flags of the dynamic invoker. * * @param opDesc Dynalink dynamic operation descriptor. * @param flags the call site flags for the operation * @param rtype the return type for the operation * @param ptypes the parameter types for the operation * @return MethodHandle for invoking the operation. */ public static MethodHandle createDynamicInvoker( final String opDesc, final int flags, final Class<?> rtype, final Class<?>... ptypes) { return bootstrap( MethodHandles.publicLookup(), opDesc, MethodType.methodType(rtype, ptypes), flags) .dynamicInvoker(); } /** * Returns a dynamic invoker for a specified dynamic operation using the public lookup. Similar to * {@link #createDynamicInvoker(String, Class, Class...)} but with return and parameter types * composed into a method type in the signature. See the discussion of that method for details. * * @param opDesc Dynalink dynamic operation descriptor. * @param type the method type for the operation * @return MethodHandle for invoking the operation. */ public static MethodHandle createDynamicInvoker(final String opDesc, final MethodType type) { return bootstrap(MethodHandles.publicLookup(), opDesc, type, 0).dynamicInvoker(); } /** * Binds any object Nashorn can use as a [[Callable]] to a receiver and optionally arguments. * * @param callable the callable to bind * @param boundThis the bound "this" value. * @param boundArgs the bound arguments. Can be either null or empty array to signify no arguments * are bound. * @return a bound callable. * @throws ECMAException with {@code TypeError} if the object is not a callable. */ public static Object bindCallable( final Object callable, final Object boundThis, final Object[] boundArgs) { if (callable instanceof ScriptFunction) { return ((ScriptFunction) callable).createBound(boundThis, boundArgs); } else if (callable instanceof BoundCallable) { return ((BoundCallable) callable).bind(boundArgs); } else if (isCallable(callable)) { return new BoundCallable(callable, boundThis, boundArgs); } throw notFunction(callable); } /** * Creates a super-adapter for an adapter, that is, an adapter to the adapter that allows * invocation of superclass methods on it. * * @param adapter the original adapter * @return a new adapter that can be used to invoke super methods on the original adapter. */ public static Object createSuperAdapter(final Object adapter) { return new JavaSuperAdapter(adapter); } /** * If the given class is a reflection-specific class (anything in {@code java.lang.reflect} and * {@code java.lang.invoke} package, as well a {@link Class} and any subclass of {@link * ClassLoader}) and there is a security manager in the system, then it checks the {@code * nashorn.JavaReflection} {@code RuntimePermission}. * * @param clazz the class being tested * @param isStatic is access checked for static members (or instance members) */ public static void checkReflectionAccess(final Class<?> clazz, final boolean isStatic) { ReflectionCheckLinker.checkReflectionAccess(clazz, isStatic); } /** * Returns the Nashorn's internally used dynamic linker's services object. Note that in code that * is processing a linking request, you will normally use the {@code LinkerServices} object passed * by whatever top-level linker invoked the linking (if the call site is in Nashorn-generated * code, you'll get this object anyway). You should only resort to retrieving a linker services * object using this method when you need some linker services (e.g. type converter method * handles) outside of a code path that is linking a call site. * * @return Nashorn's internal dynamic linker's services object. */ public static LinkerServices getLinkerServices() { return dynamicLinker.getLinkerServices(); } /** * Takes a guarded invocation, and ensures its method and guard conform to the type of the call * descriptor, using all type conversions allowed by the linker's services. This method is used by * Nashorn's linkers as a last step before returning guarded invocations. Most of the code used to * produce the guarded invocations does not make an effort to coordinate types of the methods, and * so a final type adjustment before a guarded invocation is returned to the aggregating linker is * the responsibility of the linkers themselves. * * @param inv the guarded invocation that needs to be type-converted. Can be null. * @param linkerServices the linker services object providing the type conversions. * @param desc the call site descriptor to whose method type the invocation needs to conform. * @return the type-converted guarded invocation. If input is null, null is returned. If the input * invocation already conforms to the requested type, it is returned unchanged. */ static GuardedInvocation asTypeSafeReturn( final GuardedInvocation inv, final LinkerServices linkerServices, final CallSiteDescriptor desc) { return inv == null ? null : inv.asTypeSafeReturn(linkerServices, desc.getMethodType()); } /** * Adapts the return type of the method handle with {@code explicitCastArguments} when it is an * unboxing conversion. This will ensure that nulls are unwrapped to false or 0. * * @param target the target method handle * @param newType the desired new type. Note that this method does not adapt the method handle * completely to the new type, it only adapts the return type; this is allowed as per {@link * DynamicLinkerFactory#setAutoConversionStrategy(MethodTypeConversionStrategy)}, which is * what this method is used for. * @return the method handle with adapted return type, if it required an unboxing conversion. */ private static MethodHandle unboxReturnType(final MethodHandle target, final MethodType newType) { final MethodType targetType = target.type(); final Class<?> oldReturnType = targetType.returnType(); final Class<?> newReturnType = newType.returnType(); if (TypeUtilities.isWrapperType(oldReturnType)) { if (newReturnType.isPrimitive()) { // The contract of setAutoConversionStrategy is such that the difference between newType and // targetType // can only be JLS method invocation conversions. assert TypeUtilities.isMethodInvocationConvertible(oldReturnType, newReturnType); return MethodHandles.explicitCastArguments( target, targetType.changeReturnType(newReturnType)); } } else if (oldReturnType == void.class && newReturnType == Object.class) { return MethodHandles.filterReturnValue(target, VOID_TO_OBJECT); } return target; } }