Add test case for invalid compilation against fragments from P2 repo #4293
Conversation
HeikoKlare
force-pushed
the
fragment-unmatchingp2
branch
from
b0f3710
to
5f5314b
Compare
Test Results 603 files 603 suites 4h 37m 19s ⏱️ For more details on these failures, see this check. Results for commit c742b40. ♻️ This comment has been updated with latest results. |
|
Thanks Heiko, I'll try to look into this. |
|
Thank you, Christoph! That would be really great. I am afraid my knowledge in Tycho is not deep enough to provide an according enhancement soon, at least not a proper one. |
| @@ -12,6 +12,9 @@ | |||
| *******************************************************************************/ | |||
| package org.eclipse.tycho.test.eeProfile; | |||
|
|
|||
| import static org.junit.jupiter.api.Assertions.assertThrows; | |||
|
|
|||
| import org.apache.bcel.verifier.exc.VerificationException; | |||
There was a problem hiding this comment.
bcel. VerificationException ? Are you sure? I would assume this has to be org.apache.maven.it.VerificationException.
There was a problem hiding this comment.
Yes, that's wrong. It has to be org.apache.maven.it.VerificationException. I will have a look.
There was a problem hiding this comment.
I've corrected the import and the test still fails as expected.
As an additional information for investigation: when executed locally with debug output, you see the unexpected classpath entry for the unmatching fragment when compiling the gtk.linux.x86 one:
[DEBUG] Classpath: [/Users/heikoklare/dev/eclipse/tycho/git/tycho/tycho-its/projects/eeProfile.resolution.fragments.unmatchinginp2/compiler.fragments.unmatchinginp2/.
/Users/heikoklare/.m2/repository/p2/osgi/bundle/compiler.fragments.unmatchinginp2.unmatching/1.0.0.202409231244/compiler.fragments.unmatchinginp2.unmatching-1.0.0.202409231244.jar
/Users/heikoklare/dev/eclipse/tycho/git/tycho/tycho-its/projects/eeProfile.resolution.fragments.unmatchinginp2/compiler.fragments.unmatchinginp2.gtk.linux.x86/target/classes
/Users/heikoklare/.m2/repository/org/osgi/org.osgi.service.cm/1.6.1/org.osgi.service.cm-1.6.1.jar[+org/osgi/service/cm/annotations/*:?**/*]
/Users/heikoklare/.m2/repository/org/osgi/org.osgi.annotation.bundle/1.1.1/org.osgi.annotation.bundle-1.1.1.jar[+org/osgi/annotation/bundle/*:?**/*]
/Users/heikoklare/.m2/repository/org/osgi/org.osgi.annotation.versioning/1.1.2/org.osgi.annotation.versioning-1.1.2.jar[+org/osgi/annotation/versioning/*:?**/*]
/Users/heikoklare/.m2/repository/org/osgi/org.osgi.service.metatype.annotations/1.4.1/org.osgi.service.metatype.annotations-1.4.1.jar[+org/osgi/service/metatype/annotations/*:?**/*]
/Users/heikoklare/.m2/repository/org/osgi/org.osgi.service.event/1.4.1/org.osgi.service.event-1.4.1.jar[+org/osgi/service/event/annotations/*:?**/*]
/Users/heikoklare/.m2/repository/org/osgi/org.osgi.service.component.annotations/1.5.1/org.osgi.service.component.annotations-1.5.1.jar[+org/osgi/service/component/annotations/*:?**/*]]
When compiling a fragment for whose hose a fragment exists in a P2 repo referenced by the build (even if the platform filter does not match), the classes from that fragment are put on the classpath, such that code in the currently compiled fragment is illegally compiled against that fragment code. This change adds a test case that demonstrates the behavior via a fragment that uses code from another fragment, which is must not depend on.
HeikoKlare
force-pushed
the
fragment-unmatchingp2
branch
from
5f5314b
to
c742b40
Compare
|
@HeikoKlare I see the unmatching on the classpath in the debug output but the compile does not fails, is this expected? the integration test seem to indicate that a compile error is about to be expected, also if I compare 4.0.8 debug output with 4.0.9 debug output I get the same result regarding the classpath output. |
Yes (depending what "expected" refers to): the code for the test is written in a way that compilation should fail when fragments are considered correctly. I.e., there is a class in the fragment-to-be-built that depends on a class in the unmatching fragment. Currently, the compilation succeeds, but actually a failure is expected. The test case is written in a way that it expects a compilation failure, which is why the test fails due to the compilation succeeding.
This issue seems to be existent longer than 4.0.9. The change in 4.0.9 (#3928) just changed the way in which dependencies from P2 repos are resolved/considered, such that this existent (undetected) issue now leads to a visible issue / derivation of behavior. At least this is my current understanding of the problem. I was not able to provide a test for the exact issue happening with 4.0.9 (such that the test succeeds with 4.0.8). It would be more complicated to define and I expect the issue shown with this test to be the root cause for the issue appearing with 4.0.9, as the actual difference between 4.0.8 and 4.0.9 builds is that with 4.0.9 wrong fragments appear on the classpath (like in this test) while they were not present on the classpath with 4.0.8. |
|
Thanks for the clarification. |
|
@HeikoKlare I now added this testcase together with a fix to this PR: |
When compiling a fragment for whose hose a fragment exists in a P2 repo referenced by the build (even if the platform filter does not match), the classes from that fragment are put on the classpath, such that code in the currently compiled fragment is illegally compiled against that fragment code.
This change adds a test case that demonstrates the behavior via a fragment that uses code from another fragment, which is must not depend on.
See the discussion in eclipse-platform/eclipse.platform.releng.aggregator#2360 (comment)
Interestingly, this test also fails with 4.0.8 while the SWT build issues this is supposed to be a regression test for did not appear before 4.0.9. Still, the situation in this test is, in my understanding, unexpected, and should cover the case we have in SWT as well.