Avoid forcing infos of some symbols in InfoTransforms

[odersky]

Provides the "mayChange" hook to exclude symbols from being completed
prior to a transformInfo.

[odersky]

@DarkDimius @smarter Can you check whether this fixes the CyclicReferences you were seeing in #200?

[DarkDimius]

Didn't help. I've changed a travis config to use JVM 1.8 in this branch.

[odersky]

In fact, it did help! The cyclic reference occurs now somewhere else, when we try to find the abstract terms in RefChecks. I think what this hints at is that JDK 8 has some complex cyclic relationships which JDK 7 did not have, and which trigger the Cyclic Reference errors. I could not see anything obvious where we could avoid them in the stack trace yet. Will keep looking. But this seems to be a tough one.

[odersky]

Should we switch to jdk8 for all branches?

[DarkDimius]

I guess so. If we say that our main platform is Java8 so we should test on it.
We can actually configure travis to use multiple jdk's. I propose to test both on java7 and java8. WDYT?

[odersky]

On Thu, Nov 13, 2014 at 10:36 AM, Dmitry Petrashko <notifications@github.com

wrote:

I guess so. If we say that our main platform is Java8 so we should test on
it.
We can actually configure travis to use multiple jdk's. I propose to test
both on java7 and java8. WDYT?

As long as Travis lets us, this is an option. But since we will generate
code for Java 8 only, why test on Java 7?

• Martin

—
Reply to this email directly or view it on GitHub
#223 (comment).

Martin Odersky
EPFL

[DarkDimius]

The problem fixed in this PR shows that different JKDs could be used as to show errors in compiler.