Update to Tycho 4.0.9 #2360

laeubi · 2024-09-19T10:05:54Z

No description provided.

HeikoKlare · 2024-09-20T10:23:36Z

The update to Tycho 4.0.9 broke the SWT builds. For some reason, tests within a fragment are not compiled against classes in that fragment anymore but seem to be compiled against the classes from the fragment for the current platform. This results in compilation failures for the win32 tests (within the win32 fragment) on MacOS and Linux, e.g., in our Jenkins builds such as https://ci.eclipse.org/releng/job/eclipse.platform.swt/job/PR-1478/2/pipeline-console/?selected-node=214

[ERROR] Failed to execute goal org.eclipse.tycho:tycho-compiler-plugin:4.0.9:testCompile (default-testCompile) on project org.eclipse.swt.win32.win32.x86_64: Compilation failure: Compilation failure: 
11:34:44  [ERROR] /home/jenkins/agent/workspace/eclipse.platform.swt_PR-1478/eclipse.platform.swt/bundles/org.eclipse.swt/Eclipse SWT Tests/win32/org/eclipse/swt/graphics/GCWin32Tests.java:[35] 
11:34:44  [ERROR] 	scaledGcNativeZoom.complete(event.gc.getGCData().nativeZoom);
11:34:44  [ERROR] 	                                                 ^^^^^^^^^^
...

@laeubi do you know about any change in Tycho that might have produced this regression?

HeikoKlare · 2024-09-20T12:44:50Z

Just found that with the new Tycho version, the classpath for the testCompile execution does not only contain the current fragment (like before), but also contains jars of all the other fragments from the Maven repository:

[INFO] --- tycho-compiler:4.0.9:testCompile (default-testCompile) @ org.eclipse.swt.win32.win32.aarch64 ---
...
[DEBUG] Classpath: [/Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/bundles/org.eclipse.swt/.
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.cocoa.macosx.x86_64/3.128.0.v20240917-1259/org.eclipse.swt.cocoa.macosx.x86_64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.gtk.linux.ppc64le/3.128.0.v20240917-1259/org.eclipse.swt.gtk.linux.ppc64le-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.gtk.linux.aarch64/3.128.0.v20240917-1259/org.eclipse.swt.gtk.linux.aarch64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.cocoa.macosx.aarch64/3.128.0.v20240917-1259/org.eclipse.swt.cocoa.macosx.aarch64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.gtk.linux.x86_64/3.128.0.v20240917-1259/org.eclipse.swt.gtk.linux.x86_64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/binaries/org.eclipse.swt.win32.win32.aarch64/target/classes
...

vs

[INFO] --- tycho-compiler:4.0.8:testCompile (default-testCompile) @ org.eclipse.swt.win32.win32.aarch64 ---
...
[DEBUG] Classpath: [/Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/bundles/org.eclipse.swt/.
 /Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/binaries/org.eclipse.swt.win32.win32.aarch64/target/classes
...

laeubi · 2024-09-20T13:16:27Z

Its hard to tell without a testcase and given all test for Tycho complete without a problem I would suggest to at best
provide an integration-test to demonstrate the issue.

do you know about any change in Tycho that might have produced this regression?

In general if there is no test there is no regression, just relying on undefined behavior, especially as the problem seem to not arise in the usual integration builds.

HeikoKlare · 2024-09-20T14:24:17Z

I was just asking whether there is an idea of which change might be the cause, because it would be easier if one could simply take a look at specific changes and their effects to find the cause than considering the issue as rather a black box with quite some effort to find potential reasons and writing a test case. If not, that's of course fine.

In general if there is no test there is no regression, just relying on undefined behavior, especially as the problem seem to not arise in the usual integration builds.

If that is really the attitude, then I am not sure why you have ever agreed on adding tests to fragments (eclipse-platform/eclipse.platform.swt#1203), i.e. to add code that is "relying on undefined behavior".

laeubi · 2024-09-20T15:40:16Z

I was just asking whether there is an idea of which change might be the cause

As said all test run successful (including full platform integration build) so I have no real clue, I'm also not aware of any changes in that area, as its all mostly dependency updates and some additional feature, so nothing "dramatic", maybe eclipse-tycho/tycho#3928 could be something to check but thats just a guess because it slightly changes dependency resolution in very special cases. Maybe also its just an ordering issue of the classpath but I can't spot a specific changes in that area...

If that is really the attitude, then I am not sure why you have ever agreed on adding tests to fragments

This use-case is official supported and the corresponding tests seem to pass (but sadly not covering the special situation of SWT).

HannesWell · 2024-09-20T17:39:35Z

With regards to eclipse-platform/eclipse.platform.swt#1478, I also see similar warnings that fail the build in PDE now, but they are actually false positives since the mentioned filters actually have to stay forever and in the IDE API-tools also complain if I remove them.
I also see API-errors about removed types in the platform and platform UI repo build and sporadically see them in Equinox. So something is wrong any where. It is maybe in Tycho or the API-tools themself or it is due to an infrastructure problem.

I'll try to investigate it tomorrow (of course help is always welcome).

HeikoKlare · 2024-09-21T13:49:19Z

maybe eclipse-tycho/tycho#3928 could be something to check but thats just a guess because it slightly changes dependency resolution in very special cases

Good catch, thank you! 🥇
I have just built the tycho-4.0.x branch with and without that change reverted. Reverting it makes the build succeed again.

I guess the indeterministic consequences (integration builds working while standalone SWT builds fail) might be because of dependency ordering. But in general, the change leads to further fragments being placed on the classpath.

This is the complete classpath (during compile / testCompile) with the mentioned change:

[DEBUG] Classpath: [/Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/bundles/org.eclipse.swt/.
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.cocoa.macosx.x86_64/3.128.0.v20240917-1259/org.eclipse.swt.cocoa.macosx.x86_64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.gtk.linux.ppc64le/3.128.0.v20240917-1259/org.eclipse.swt.gtk.linux.ppc64le-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.gtk.linux.aarch64/3.128.0.v20240917-1259/org.eclipse.swt.gtk.linux.aarch64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.cocoa.macosx.aarch64/3.128.0.v20240917-1259/org.eclipse.swt.cocoa.macosx.aarch64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.swt.gtk.linux.x86_64/3.128.0.v20240917-1259/org.eclipse.swt.gtk.linux.x86_64-3.128.0.v20240917-1259.jar
 /Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/binaries/org.eclipse.swt.win32.win32.aarch64/target/classes
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-api/5.11.0/junit-jupiter-api-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-engine/5.11.0/junit-jupiter-engine-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-migrationsupport/5.11.0/junit-jupiter-migrationsupport-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-params/5.11.0/junit-jupiter-params-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-commons/1.11.0/junit-platform-commons-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-engine/1.11.0/junit-platform-engine-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-launcher/1.11.0/junit-platform-launcher-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-runner/1.11.0/junit-platform-runner-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-suite-api/1.11.0/junit-platform-suite-api-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/vintage/junit-vintage-engine/5.11.0/junit-vintage-engine-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/opentest4j/opentest4j/1.3.0/opentest4j-1.3.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/apiguardian/apiguardian-api/1.1.2/apiguardian-api-1.1.2.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.junit/4.13.2.v20230809-1000/org.junit-4.13.2.v20230809-1000.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/apache/servicemix/bundles/org.apache.servicemix.bundles.hamcrest/1.3_1/org.apache.servicemix.bundles.hamcrest-1.3_1.jar[+**/*:?**/*]
 /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.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.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.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.versioning/1.1.2/org.osgi.annotation.versioning-1.1.2.jar[+org/osgi/annotation/versioning/*:?**/*]
 /Users/heikoklare/.m2/repository/org/eclipse/pde/org.eclipse.pde.api.tools.annotations/1.4.0/org.eclipse.pde.api.tools.annotations-1.4.0.jar[+org/eclipse/pde/api/tools/annotations/*:?**/*]
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.osgi.services/3.12.100.v20240327-0645/org.eclipse.osgi.services-3.12.100.v20240327-0645.jar[+org/osgi/service/component/annotations/*:?**/*]]

And this is without it:

[DEBUG] Classpath: [/Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/bundles/org.eclipse.swt/.
 /Users/heikoklare/dev/eclipse/platform-aggregator/git/eclipse.platform.swt/binaries/org.eclipse.swt.win32.win32.aarch64/target/classes
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-api/5.11.0/junit-jupiter-api-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-engine/5.11.0/junit-jupiter-engine-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-migrationsupport/5.11.0/junit-jupiter-migrationsupport-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/jupiter/junit-jupiter-params/5.11.0/junit-jupiter-params-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-commons/1.11.0/junit-platform-commons-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-engine/1.11.0/junit-platform-engine-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-launcher/1.11.0/junit-platform-launcher-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-runner/1.11.0/junit-platform-runner-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/platform/junit-platform-suite-api/1.11.0/junit-platform-suite-api-1.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/junit/vintage/junit-vintage-engine/5.11.0/junit-vintage-engine-5.11.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/opentest4j/opentest4j/1.3.0/opentest4j-1.3.0.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/apiguardian/apiguardian-api/1.1.2/apiguardian-api-1.1.2.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.junit/4.13.2.v20230809-1000/org.junit-4.13.2.v20230809-1000.jar[+**/*:?**/*]
 /Users/heikoklare/.m2/repository/org/apache/servicemix/bundles/org.apache.servicemix.bundles.hamcrest/1.3_1/org.apache.servicemix.bundles.hamcrest-1.3_1.jar[+**/*:?**/*]
 /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.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.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.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.versioning/1.1.2/org.osgi.annotation.versioning-1.1.2.jar[+org/osgi/annotation/versioning/*:?**/*]
 /Users/heikoklare/.m2/repository/org/eclipse/pde/org.eclipse.pde.api.tools.annotations/1.4.0/org.eclipse.pde.api.tools.annotations-1.4.0.jar[+org/eclipse/pde/api/tools/annotations/*:?**/*]
 /Users/heikoklare/.m2/repository/p2/osgi/bundle/org.eclipse.osgi.services/3.12.100.v20240327-0645/org.eclipse.osgi.services-3.12.100.v20240327-0645.jar[+org/osgi/service/component/annotations/*:?**/*]]

The difference between these two are exactly the other fragments of the host of the fragment currently being built.

laeubi · 2024-09-22T11:17:28Z

Looking at the output I see that multiple fragments are resolved but actually their platform filter should not match, this has not changed but seem to contribute to the problem here.

HeikoKlare · 2024-09-23T11:58:25Z

Yes, the change resulted in these additional fragments being resolved (from the m2 repo). The causing change is the change in TargetPlatformFactoryImpl#getPreliminaryReactorProjectUIs() from

project.getDependencyMetadata(DependencyMetadataType.INITIAL);

to

project.getDependencyMetadata(DependencyMetadataType.SEED);

I cannot say anything about the reasons for this change leading to additional fragments on the classpath. I tried to provide a Tycho test case for some time now, but so far I was not able to write a proper regression test.

laeubi · 2024-09-23T12:40:40Z

I tried to provide a Tycho test case for some time now, but so far I was not able to write a proper regression test.

My guess currently is that they are pulled in by the I-Build repository what contains the fragments of the previous build (as enabled by the -Pbuild-individual-bundles) that would also explain why it is not a problem in the integration builds itself.

HeikoKlare · 2024-09-23T13:22:29Z

My guess currently is that they are pulled in by the I-Build repository what contains the fragments of the previous build (as enabled by the -Pbuild-individual-bundles) that would also explain why it is not a problem in the integration builds itself.

I can confirm this with my latest experiments: stripping down the SWT builds results in the fragments disappearing from the classpath once I remove the I-Build repo (provided via the build-individual-bundles profile).

I was not able to generate a perfect reproduction, but now I at least one that shows the problem: when you reference a P2 repo in your build that contains a fragment for the host of a fragment currently being built, that other fragment will be put on the classpath such that you (illegally) compile against it. The test case is provided via: eclipse-tycho/tycho#4293

Update to Tycho 4.0.9

000c493

akurtakov merged commit 47c6b46 into eclipse-platform:master Sep 19, 2024
4 checks passed

HeikoKlare mentioned this pull request Sep 20, 2024

Remove obsolete API filter for GCData and Display on MacOS eclipse-platform/eclipse.platform.swt#1478

Open

This was referenced Sep 20, 2024

Remove short-named equinox.launcher.cocoa.macosx fragment eclipse-equinox/equinox#622

Merged

Build swt failed on v4967r3 eclipse-platform/eclipse.platform.swt#1481

Closed

HannesWell mentioned this pull request Sep 22, 2024

Temporarily use 2024-06 API-Tools for API checks in verification builds #2374

Merged

HeikoKlare mentioned this pull request Sep 23, 2024

Add test case for invalid compilation against fragments from P2 repo eclipse-tycho/tycho#4293

Draft

HannesWell mentioned this pull request Sep 23, 2024

[Build] Temporarily use tycho-apitools:verify mojo in version 4.0.8 #2380

Merged

akurtakov mentioned this pull request Sep 23, 2024

Add dependency to o.e.test to o.e.swt.tests eclipse-platform/eclipse.platform.swt#1475

Merged

This was referenced Sep 25, 2024

Temporarily revert to Tycho 4.0.8 for GitHub Actions builds eclipse-platform/eclipse.platform.swt#1487

Merged

Switch back GitHub Actions workflow to use current Tycho version eclipse-platform/eclipse.platform.swt#1488

Draft

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update to Tycho 4.0.9 #2360

Update to Tycho 4.0.9 #2360

laeubi commented Sep 19, 2024

HeikoKlare commented Sep 20, 2024

HeikoKlare commented Sep 20, 2024

laeubi commented Sep 20, 2024

HeikoKlare commented Sep 20, 2024

laeubi commented Sep 20, 2024

HannesWell commented Sep 20, 2024

HeikoKlare commented Sep 21, 2024 •

edited

Loading

laeubi commented Sep 22, 2024 •

edited

Loading

HeikoKlare commented Sep 23, 2024

laeubi commented Sep 23, 2024 •

edited

Loading

HeikoKlare commented Sep 23, 2024

Update to Tycho 4.0.9 #2360

Update to Tycho 4.0.9 #2360

Conversation

laeubi commented Sep 19, 2024

HeikoKlare commented Sep 20, 2024

HeikoKlare commented Sep 20, 2024

laeubi commented Sep 20, 2024

HeikoKlare commented Sep 20, 2024

laeubi commented Sep 20, 2024

HannesWell commented Sep 20, 2024

HeikoKlare commented Sep 21, 2024 • edited Loading

laeubi commented Sep 22, 2024 • edited Loading

HeikoKlare commented Sep 23, 2024

laeubi commented Sep 23, 2024 • edited Loading

HeikoKlare commented Sep 23, 2024

HeikoKlare commented Sep 21, 2024 •

edited

Loading

laeubi commented Sep 22, 2024 •

edited

Loading

laeubi commented Sep 23, 2024 •

edited

Loading