Back out the fixes for #2158 and #3623 to to understand status quo ante

[srikanth-sankaran]

• Backs out only code changes (note: not a full reversal, some peripheral changes have been left in)
• As tests are not backed out, these would fail.

[srikanth-sankaran]

Hmm, I am surprised by a green test run from github for commit 1 - I expected at least the following tests to fail:

org.eclipse.jdt.core.tests.model.ExternalAnnotations9Test.testGH3623_withNullAnnoatations()
org.eclipse.jdt.core.tests.model.ExternalAnnotations9Test.testGH3623_nullAnnotationsNotEnabled()
org.eclipse.jdt.core.tests.compiler.regression.NullTypeAnnotationTest.testGH2158()

[srikanth-sankaran]

Hmm, I am surprised by a green test run from github for commit 1 - I expected at least the following tests to fail:

org.eclipse.jdt.core.tests.model.ExternalAnnotations9Test.testGH3623_withNullAnnoatations() org.eclipse.jdt.core.tests.model.ExternalAnnotations9Test.testGH3623_nullAnnotationsNotEnabled() org.eclipse.jdt.core.tests.compiler.regression.NullTypeAnnotationTest.testGH2158()

It gets curiouser and curiouser!

I have pushed the regression test (alone, without code changes) from #4682 here and I expect it to pass also - as it passes locally. This is completely counterintuitive!!!

So it would seem backing out the fixes for #2158 and #3623 solves #2158 and #3623 and #4494 - no explanations yet. I ran into this scenario while simply preparing the grounds for understanding status quo ante #2158 and #3623

@stephan-herrmann - FYI. You can hold off on tracking this until I call out.

[stephan-herrmann]

I have some news for you here: if additionally you'd backout #4554 (4.38 M3), then you'd see org.eclipse.jdt.core.tests.compiler.regression.NullTypeAnnotationTest.testGH2158() failing again.

Method Scope.areSignificantlyDifferent(TypeBinding, TypeBinding) is intended to filter out differences caused only be "equivalent" but different wildcard bindings. Apparently that method now also masks a problem of unexpectedly different ids. I can indeed observe two different PTBs with different ids, while all their ingredients (type, arguments ...) have the same ids pairwise. I.e., !areSignificantlyDifferent() and also TypeBinding.notEquals() and even with no wildcards in the picture.

HTH

[srikanth-sankaran]

I have some news for you here: if additionally you'd backout #4554 (4.38 M3), then you'd see org.eclipse.jdt.core.tests.compiler.regression.NullTypeAnnotationTest.testGH2158() failing again.

This saves me lots of time in digging - thanks.