Summary
-------

Redesign the `javac` annotations pipeline to better address the
requirements of annotations and tools that process annotations.


Goals
-----

  - The compiler should behave correctly with regard to annotations and
    type annotations: the emitted class files should have
    correctly-formatted attributes for all kinds of annotations.

  - Compile-time reflection (`javax.lang.model` and
    `javax.annotation.processing`) should properly handle all annotations
    in signature-visible positions.

  - Run-time reflection (Core Reflection) should work properly with
    annotations in class files.

  - The [Checkers Framework][1] must continue to work.

  - The Java Compiler API should continue to work as designed.


Non-Goals
---------

No new language features or APIs will be added. It is not a goal to
update support for annotations in related tools such as `javadoc` and
`javap`.


Motivation
----------

Java SE 8 introduced two new annotation features,
[Repeating Annotations (JEP 120)][jep120] and
[Annotations on Java Types (JSR 308 / JEP 104)][jep104].
In addition [Lambda Expressions (JSR 335 / JEP 126)][jep126] added new
syntactic position for annotations. These features, none of which existed
when the `javac` annotation pipeline was initially conceived, can be
combined together, leading to patterns such as:

    Function<String, String> fss = (@Anno @Anno String s) -> s;

The existing annotation pipeline could not handle such cases out of the
box; as a result, the original design has been stretched in order to
accommodate the new use cases, leading to a brittle and hard-to-maintain
implementation. The goal of this work is to replace this aging
architecture with a new one which supports the new use cases in a more
straightforward fashion, leading to more-correct and maintainable code.


Description
-----------

Refactor the `javac` annotation pipeline. This should not be externally
noticeable except where we fix bugs and improve correctness. The first
step is to improve testing coverage so we can measure and evaluate our
exit criteria. After that follows a series of incremental
refactorings. This work will be done in the OpenJDK
[Annotations Pipeline 2.0 Project][2].

The `javadoc` tool has some related issues with regards to type
annotations. `javadoc` is, however, undergoing significant work as part of
the [Javadoc.Next Project](http://openjdk.java.net/projects/javadoc-next/).
Part of that work includes converting `javadoc` to use the
`javax.lang.model` API instead of the older `com.sun.javadoc`
API. It is, therefore, not a goal of this project to work on `javadoc` to
ensure that annotations, including type annotations, are presented
correctly. It is expected that, as part of the JavaDoc.Next Project,
`javadoc` will be enhanced to leverage the updates to the
`javax.lang.model` API that are a goal of this project.


Testing
-------

There is already good coverage of a substantial set of end-to-end use
cases of annotations. This includes JCK and langtools regression tests.
A big part of this work is to develop further tests to ensure the
measurability of the success metrics.

We will create tests that exercise the intersection of the new features
in Java SE 8, as mentioned above.


[1]: http://types.cs.washington.edu/checker-framework/
[2]: http://openjdk.java.net/projects/anno-pipeline/
[jep104]: http://openjdk.java.net/jeps/104
[jep120]: http://openjdk.java.net/jeps/120
[jep126]: http://openjdk.java.net/jeps/126