Changing annotation processing to use single Context for all rounds of annotation processing

[Werner Dietl]

Hi Jan,

thanks for the heads-up!
The Checker Framework compiles with these changes, but there is one
unit test failure:

https://code.google.com/p/checker-framework/source/browse/checkers/tests/framework/UnimportedExtends2.java

now fails with:

    [junit] Testcase:
run[framework/UnimportedExtends2](tests.FrameworkTest): Caused an
ERROR
    [junit] null
    [junit] java.lang.NullPointerException
    [junit] at com.sun.tools.javac.main.JavaCompiler.complete(JavaCompiler.java:795)
    [junit] at com.sun.tools.javac.processing.JavacProcessingEnvironment$ImplicitCompleter.complete(JavacProcessingEnvironment.java:1307)
    [junit] at com.sun.tools.javac.code.Symbol.complete(Symbol.java:560)
    [junit] at com.sun.tools.javac.code.Symbol$ClassSymbol.complete(Symbol.java:1030)
    [junit] at com.sun.tools.javac.code.Type$ClassType.complete(Type.java:913)
    [junit] at com.sun.tools.javac.code.Type$ClassType.getTypeArguments(Type.java:831)
    [junit] at com.sun.tools.javac.code.Printer.visitClassType(Printer.java:223)
    [junit] at com.sun.tools.javac.code.Printer.visitClassType(Printer.java:52)
    [junit] at com.sun.tools.javac.code.Type$ClassType.accept(Type.java:763)
    [junit] at com.sun.tools.javac.code.Printer.visit(Printer.java:136)
    [junit] at com.sun.tools.javac.util.AbstractDiagnosticFormatter.formatArgument(AbstractDiagnosticFormatter.java:193)
    [junit] at com.sun.tools.javac.util.AbstractDiagnosticFormatter.formatArguments(AbstractDiagnosticFormatter.java:161)
    [junit] at com.sun.tools.javac.util.BasicDiagnosticFormatter.formatMessage(BasicDiagnosticFormatter.java:111)
    [junit] at com.sun.tools.javac.util.BasicDiagnosticFormatter.formatMessage(BasicDiagnosticFormatter.java:67)
    [junit] at com.sun.tools.javac.util.AbstractDiagnosticFormatter.formatArgument(AbstractDiagnosticFormatter.java:179)
    [junit] at com.sun.tools.javac.util.AbstractDiagnosticFormatter.formatArguments(AbstractDiagnosticFormatter.java:161)
    [junit] at com.sun.tools.javac.util.BasicDiagnosticFormatter.formatMessage(BasicDiagnosticFormatter.java:111)
    [junit] at com.sun.tools.javac.util.BasicDiagnosticFormatter.formatMeta(BasicDiagnosticFormatter.java:184)
    [junit] at com.sun.tools.javac.util.BasicDiagnosticFormatter.formatDiagnostic(BasicDiagnosticFormatter.java:100)
    [junit] at com.sun.tools.javac.util.AbstractDiagnosticFormatter.format(AbstractDiagnosticFormatter.java:117)
    [junit] at com.sun.tools.javac.util.AbstractDiagnosticFormatter.format(AbstractDiagnosticFormatter.java:69)
    [junit] at com.sun.tools.javac.util.JCDiagnostic.toString(JCDiagnostic.java:537)
    [junit] at com.sun.tools.javac.api.ClientCodeWrapper$DiagnosticSourceUnwrapper.toString(ClientCodeWrapper.java:600)
    [junit] at checkers.util.test.CheckerTest.assertDiagnostics(CheckerTest.java:286)
    [junit] at checkers.util.test.CheckerTest.checkTestResult(CheckerTest.java:200)
    [junit] at checkers.util.test.CheckerTest.runTest(CheckerTest.java:134)
    [junit] at checkers.util.test.CheckerTest.test(CheckerTest.java:65)
    [junit] at checkers.util.test.ParameterizedCheckerTest.run(ParameterizedCheckerTest.java:25)

(Note that I applied your patch against jsr308-langtools, so some of
these line numbers might be off. JavaCompiler.java is unchanged.)

I didn't see a performance difference when running the unit tests.

Best regards,
cu, WMD.

On Thu, Mar 6, 2014 at 4:22 PM, Jan Lahoda <jan.lahoda at oracle.com> wrote:
> Hi Werner and others,
>
> Currently, most of javac's services and data get cleared between annotation
> processing rounds. Lately, I've been working on a patch that would allow to
> keep as much as possible across rounds. The advantages include faster
> compilation (as ClassSymbols from classpath don't need to be read again and
> again for each annotation processing round) and cleaner javac internals.
>
> My current patch is available here:
> http://cr.openjdk.java.net/~jlahoda/2206926/webrev.01/
>
> The patch should be applicable to current jdk9/dev/langtools and current
> type-annotations/langtools.
>
> Any comments on the patch, or practical experiences running the Checker
> Framework (or any other annotation processor/framework) on a javac with this
> patch are very welcome and would be greatly appreciated.
>
> Thanks,
>     Jan



-- 
http://www.google.com/profiles/wdietl