Comment 1 Dani Megert 2011-04-15 05:06:26 EDT

Problem is that
    ToolProvider.getSystemJavaCompiler();
returns 'null'.

According to Kim, the SDK is installed and tools.jar in place.

Comment 2 Dani Megert 2011-04-15 05:30:44 EDT

I just tried on our Suse Linux machine using jdk-7-ea-bin-b137-linux-i586 and the tests pass.

Comment 3 Olivier Thomann 2011-04-15 10:23:38 EDT

When I try to run a small program using this machine that prints out the java system compiler, and it works perfectly fine.
The best I can do for now is to skip the test when the system java compiler is null. This can potentially happen so we might want to add protection against this anyway.

Comment 4 Dani Megert 2011-04-15 10:26:35 EDT

Kim, can you spin another build once that's done? We've fixed most of the issues.

Comment 5 Kim Moir 2011-04-15 10:45:25 EDT

Sure, I'm running a test build for another bug.  I'll rerun those tests with Java 7 once the zips are available.

Comment 6 Olivier Thomann 2011-04-15 13:37:02 EDT

Annotation processors don't behave exactly the same way between javac 1.7 and javac 1.6.
I'll try to come up with a reasonable support for both.

Comment 7 Olivier Thomann 2011-04-15 13:48:19 EDT

Created attachment 193394 [details]
Proposed fix

Comment 8 Olivier Thomann 2011-04-15 13:48:32 EDT

Released for 3.7M7.

Comment 9 Dani Megert 2011-04-18 02:37:35 EDT

(In reply to comment #6)
> Annotation processors don't behave exactly the same way between javac 1.7 and
> javac 1.6.
> I'll try to come up with a reasonable support for both.

What's the difference? Anything that needs more attention?

Any clue why the tests get 'null' for getSystemJavaCompiler()? Aren't the tests run with Ant? Maybe some problem there?

Comment 10 Dani Megert 2011-04-18 02:37:46 EDT

(In reply to comment #5)
> Sure, I'm running a test build for another bug.  I'll rerun those tests with
> Java 7 once the zips are available.

Kim, I haven't seen the build. BTW: please, make sure that code coverage is disabled.

Comment 11 Kim Moir 2011-04-18 07:51:39 EDT

I had code coverage enabled in the previous build so I'll have to run them again. Are there specific test suites that you'd like me to run so instead of running all the tests again (4-5 hours)

Comment 12 Dani Megert 2011-04-18 07:53:37 EDT

(In reply to comment #11)
> I had code coverage enabled in the previous build so I'll have to run them
> again. Are there specific test suites that you'd like me to run so instead of
> running all the tests again (4-5 hours)

I'd like to run all, so that I can look at one page and get the full picture.

Comment 13 Kim Moir 2011-04-18 08:29:01 EDT

Okay, I'm rerunning the tests on Linux for N20110417-2000 which was a green build so any failures should be quite evident.  I'll regenerate the page once they have completed.

Comment 14 Olivier Thomann 2011-04-18 09:02:27 EDT

(In reply to comment #9)
> (In reply to comment #6)
> > Annotation processors don't behave exactly the same way between javac 1.7 and
> > javac 1.6.
> > I'll try to come up with a reasonable support for both.
> 
Most differences are around the handling of error type. This is loosely spec'd and therefore it is difficult to know exactly what is supposed to be returned.
 
> What's the difference? Anything that needs more attention?
jsr269 doesn't really have specs. The specs are the javadoc of the corresponding types.

> Any clue why the tests get 'null' for getSystemJavaCompiler()? Aren't the tests
> run with Ant? Maybe some problem there?
No, but since this is possible I prefer to handle the cases inside the test suite. I believe that now the compiler.tool and compiler.apt tests are more resilient.

Comment 15 Jay Arthanareeswaran 2011-04-25 13:23:20 EDT

Verified for 3.7M7 using code inspection.