reflectable 0.5.0 copy "reflectable: ^0.5.0" to clipboard
reflectable: ^0.5.0 copied to clipboard

outdatedDart 1 only

This package allows programmers to reduce certain usages of dynamic reflection to a transformer generated specialization with the same behavior. The generated code does not use dynamic reflection and [...]

Reflectable #

A listing of known limitations is given at the end of this page.

Introduction #

This package provides support for reflection which may be tailored to cover certain reflective features and omit others, thus reducing the resource requirements at runtime.

The core idea is that the desired level of support for reflection is specified explicitly and statically, and any usage of reflection at runtime must stay within the boundaries thus specified; otherwise a NoSuchCapabilityError is thrown. In return for statically confining the required reflection support, programs can be smaller. Other than that, using this package and using dart:mirrors is very similar.

The core concept in the reflection support specification is that of a reflectable capability. For a more detailed discussion about capabilities, please consult the reflectable capability design document. On this page we just use a couple of simple special cases.

The resource benefits obtained by using this package are established by transformation. That is, this package includes a pub transformer. The transformer receives a specification of which programs to transform (see the description of entry_points: below), and it uses certain elements in the program itself to decide how to transform it. As a result, the transformed program will contain generated code providing the requested level of support for reflection. If a quick turn-around is more important than space economy, it is also possible to run the code without transformation. In this case, the behavior of the program is identical, but it is implemented by delegation to dart:mirrors, which means that the execution may be costly in terms of space.

In other words, the package can be used in two modes: transformed mode and untransformed mode.

For both modes, your code will not depend directly on dart:mirrors. In the untransformed mode there is an indirect dependency because the support for reflective features is implemented by delegation to mirror objects from dart:mirrors. In the transformed mode there is no dependency on dart:mirrors at all, because the reflectable transformer generates specialized, static code that supports the required reflective features.

The use of dynamic reflection is supported if and only if the usage is covered by the set of capabilities specified.

In general, reflection is provided via a subclass of the class Reflectable (we use the term reflector to designate an instance of such a subclass).

class Reflector extends Reflectable {
  const Reflector() : super(capability1, capability2, ...);
}

Reflection is disabled by default, and it is enabled by specifying reflectable capabilities. With Reflector we have specified that capability1, capability2, and so on must be supported. The main case for using this is annotating a given class A with a reflector. This implies that the specified level of reflection support for that reflector should be provided for the class A and its instances.

@Reflector()
class A {
  ...
}

Only classes covered by a reflector R and their instances can be accessed reflectively using R, and that access is constrained by the capabilities passed as the superinitializer in the class of R. The basic case is when R is used as an annotation of a given class (as is the case with A above), but a class C can also be covered by a reflector R because a supertype A of C is annotated by a reflector which specifies that subtypes of A are covered. There are several other indirect mechanisms in addition to subtype based coverage, as described in the capability design document.

As a result, the available universe of reflection related operations is so well-known at compile time that it is possible to specialize the support for reflection to a purely static form that satisfies the requirements of the given annotations and capabilities.

Usage #

Here is a simple usage example. Assume that we have the following code in the file web/main.dart:

import 'package:reflectable/reflectable.dart';

// Annotate with this class to enable reflection.
class Reflector extends Reflectable {
  const Reflector()
      : super(invokingCapability); // Request the capability to invoke methods.
}

const reflector = const Reflector();

@reflector // This annotation enables reflection on A.
class A {
  final int a;
  A(this.a);
  greater(int x) => x > a;
  lessEqual(int x) => x <= a;
}

main() {
  A x = new A(10);
  // Reflect upon [x] using the const instance of the reflector:
  InstanceMirror instanceMirror = reflector.reflect(x);
  int weekday = new DateTime.now().weekday;
  // On Fridays we test if 3 is greater than 10, on other days if it is less
  // than or equal.
  String methodName = weekday == DateTime.FRIDAY ? "greater" : "lessEqual";
  // Reflectable invocation:
  print(instanceMirror.invoke(methodName, [3]));
}

You can directly run this code in untransformed mode, for instance using the Dart VM. To avoid the dependency on dart:mirrors and save space, you can transform the code using the transformer in this package.

In order to do this, add the following to your 'pubspec.yaml':

dependencies:
  reflectable: any

transformers:
- reflectable:
    entry_points:
      - web/main.dart # The path to your main file
    formatted: true # Optional.

Now run pub build --mode=debug web to perform the transformation. This will rename the file web/main.dart and generate a new file build/web/main.dart. That file contains the data needed for reflection, and a main function that will initialize the reflection framework before running the original main. When you run this file, it is important that the package-root is set to build/web/packages, because the reflectable package itself is transformed to a version that uses the generated data, instead of using dart:mirrors.

Some elements in this scenario are optional: In the pub build.. command, you could omit the final argument when it is equal to web, because that is the default. You could also omit --mode=debug if you do not wish to inspect or use the generated Dart code (typically, that would be because you only need the final JavaScript output). In 'pubspec.yaml', the formatted option can be omitted, in which case the output from the transformer will skip the formatting step (saving time and space, but putting most of the generated code into one very long line). Finally, in 'pubspec.yaml' you can use one more option: 'suppress_warnings'; e.g., you could specify 'suppress_warnings: missing_entry_point' in order to suppress the warnings given in the situation where an entry point has been specified in 'pubspec.yaml', but no corresponding library (aka 'asset') has been provided by pub to the transformer.

For a more advanced example, you could look at serialize_test.dart and its library, where the base of a serialization framework is implemented; or you could look at meta_reflectors_test.dart and the libraries it imports, which illustrates how reflectable can be used to dynamically make a choice among several kinds of reflection, and how to eliminate several kinds of static dependencies among libraries.

Comparison with dart:mirrors with MirrorsUsed annotations #

A lot of the overhead of dart:mirrors can be mitigated by using a suitable MirrorsUsed annotation, which has a similar purpose as the capabilities of reflectable. So why do we have both?

First, on some Dart platforms there is no support for reflection at all, in which case it is necessary to use an approach that offers execution without any built-in reflection mechanism, and the reflectable transformer does just that.

Even when there is support for reflection, it makes sense to use reflectable. In particular, the concepts and mechanisms embedded in the reflectable capabilities represent a significant further development and generalization of the approach used with MirrorsUsed, which means that the level of reflection support can be tailored more precisely.

Another advantage of the reflectable approach is that the same restrictions are implemented both for running untransformed code (on top of dart:mirrors), and for running transformed code. The behaviour will thus be the same on Dartium and with code that is transformed and then compiled with Dart2js.

Another advantage is that whereas the interaction between two different MirrorsUsed annotations often is hard to predict, reflectable allows for several separate mirror-systems (reflectors) with different capabilities. This means that for some reflection targets used by one sub-system of the program (say, a serializer) detailed information can be generated. Another sub-system with simpler requirements could then be supported by a much smaller amount of information.

Known limitations #

Several parts of the library have not yet been implemented. In particular, the following parts are still incomplete:

  • Reflection on functions/closures. We do not have the required primitives to support this feature, so it is expected to remain unsupported for a while.

  • Private declarations. There is currently almost no support for reflection on private declarations, as this would require special support from the runtime for accessing private names from other libraries. As an example of a case where there is some support, library mirrors can deliver class mirrors for private classes, and instanceMembers includes public members inherited from private superclasses. But in the vast majority of situations, private declarations are not supported.

  • uri's of libraries. The transformer framework does not give us access to good uri's of libraries, so these are currently only partially supported: A unique uri containing the name given in the library directive (if any) is generated for each library; this means that equality tests will work, and the toString() of a uri will be somewhat human readable. But this kind of uri does not give any information about the location of a corresponding file on disk.

  • Type arguments of generic types are only supported in the simple cases. E.g., when it is known that a given list of actual type arguments is empty then the empty list is returned. However, when a parameterized type has a non-trivial list of actual type arguments then returning the actual type arguments would require runtime support that does not currently exist.

  • The mirror methods delegate, isAssignableTo, isSubtypeOf, and libraryDependencies have not yet been implemented with transformed code.

  • When an attempt to invoke a method, getter, or setter fails, it is sometimes not possible to distinguish the situation where the method is missing (in which case noSuchMethod should be invoked) and the situation where the method is present, but the requested capabilities do not provide access to it (in which case a NoSuchCapabilityError should be thrown).

Feature requests and bug reports #

Please file feature requests and bugs using the github issue tracker for this repository.

154
likes
0
pub points
95%
popularity

Publisher

unverified uploader

This package allows programmers to reduce certain usages of dynamic reflection to a transformer generated specialization with the same behavior. The generated code does not use dynamic reflection and thus improves the performance of (esp. dart2js) generated code.

Homepage

License

unknown (LICENSE)

Dependencies

analyzer, barback, code_transformers, dart_style, logging

More

Packages that depend on reflectable