Buid I20231208-1800 is unstable

[iloveeclipse]

See

• https://download.eclipse.org/eclipse/downloads/drops4/I20231208-1800/
• https://download.eclipse.org/eclipse/downloads/drops4/I20231208-1800/buildlogs/comparatorlogs/buildtimeComparatorUnanticipated.log.txt

Looks like most bundles from the platform repository were affected.

I assume this is caused by eclipse-platform/eclipse.platform#965 that changed build settings for compilation with maven.

The most intriguing question is: which settings were used before???

I will check that later.

[iloveeclipse]

which settings were used before???

@laeubi : I would be interested in your opinion regarding this question. Is that maven/tycho/ecj who defines defaults if nothing is specified? Or do we have some default "profile" or whatever that is used in such case?

[laeubi]

First, ecj defaults are used (not sure where these are defined) then compiler args are used (e.g.

eclipse.platform.releng.aggregator/eclipse-platform-parent/pom.xml

Lines 473 to 507 in 888de34

	<plugin>
	<groupId>org.eclipse.tycho</groupId>
	<artifactId>tycho-compiler-plugin</artifactId>
	<version>${tycho.version}</version>
	<dependencies>
	<dependency>
	<groupId>org.eclipse.jdt</groupId>
	<artifactId>ecj</artifactId>
	<version>${cbi-ecj-version}</version>
	</dependency>
	</dependencies>
	<configuration>
	<compilerArgs>
	<args>${code.ignoredWarnings}</args>
	<args>-verbose</args>
	<args>-inlineJSR</args>
	<args>-enableJavadoc</args>
	<args>-encoding</args>
	<args>${project.build.sourceEncoding}</args>
	<args>-proceedOnError</args>
	<!-- Use this form, for Tycho 22 or less.
	<args>-log</args>
	<args>${project.build.directory}/@dot.xml</args>
	-->
	</compilerArgs>
	<!-- Use this form for Tycho 23 or greater -->
	<log>xml</log>
	<logDirectory>${project.build.directory}/compilelogs</logDirectory>
	<showWarnings>true</showWarnings>
	<excludeResources>
	<exclude>**/package.html</exclude>
	</excludeResources>
	<useProjectSettings>false</useProjectSettings>
	</configuration>
	</plugin>

), what then now is overridden here to use the project specific settings file, see https://tycho.eclipseprojects.io/doc/latest/tycho-compiler-plugin/compile-mojo.html#useProjectSettings if file ${project.basedir}/.settings/org.eclipse.jdt.core.prefs can be found by tycho.

[HannesWell]

@iloveeclipse is it correct that unstable builds are not added to the i-builds composite and therefore I cannot use the methods added in eclipse-equinox/equinox#424 in eclipse-platform/eclipse.platform#972?

[iloveeclipse]

Exact. Until we have a stable build we can't move forward. Makes sense.

[iloveeclipse]

Ome can revert eclipse-platform/eclipse.platform#965 as a quick fix.

[HannesWell]

Exact. Until we have a stable build we can't move forward. Makes sense.

Absolutely. Just wanted to confirm it. :)

Ome can revert eclipse-platform/eclipse.platform#965 as a quick fix.

Yes. I although I wonder what can be configured in the settings that only changes the generated byte-code?

Looking for example at
https://github.com/eclipse-platform/eclipse.platform/blob/a5789fbe6b8787def560cc10ed086c50f3662980/runtime/bundles/org.eclipse.core.jobs/.settings/org.eclipse.jdt.core.prefs#L12-L15

I would say it is probably on of

`inlineJsrBytecode=enabled`
`methodParameters=do not generate`
`codegen.unusedLocal=preserve`

[akurtakov]

unusedLocal=preserve looks wrong to me and my gut feeling is that this is the reason why POP is replaced by ASTORE

[iloveeclipse]

I haven't investigated this deeper, so no proposal from my side here.

[HannesWell]

Looking at the build logs of a PR from eclipse.platform I see so it is reproducible (but doesn't fail the build)

10:22:25.107 [WARNING] MavenProject: org.eclipse.platform:org.eclipse.core.jobs:3.15.200-SNAPSHOT @ /home/jenkins/agent/workspace/eclipse.platform_PR-971/runtime/bundles/org.eclipse.core.jobs/.polyglot.META-INF: baseline and build artifacts have same version but different contents
   no-classifier: different
      org/eclipse/core/internal/jobs/ImplicitJobs.class: different
      org/eclipse/core/internal/jobs/InternalWorker.class: different
      org/eclipse/core/internal/jobs/JobListeners.class: different
      org/eclipse/core/internal/jobs/JobManager.class: different
      org/eclipse/core/internal/jobs/LockManager$LockState.class: different
      org/eclipse/core/internal/jobs/LockManager.class: different
      org/eclipse/core/internal/jobs/OrderedLock.class: different
      org/eclipse/core/internal/jobs/ThreadJob.class: different
      org/eclipse/core/internal/jobs/Worker.class: different
      org/eclipse/core/internal/jobs/WorkerPool.class: different

I'll try out a few options and check the results in a PR.
In general we could consider to fail the build in such cases by setting -Dtycho.p2.baselineMode to a corresponding value.

[akurtakov]

What is the status here? If there is no fix in the next couple of hours I'll partially revert the patch causing it to not change compiler options.

[HannesWell]

I cannot help on this issue any more before this evening.
For me there is some input/knowledge missing at the moment.
But since a soonish resultion is desired anybody else has to take action based on what has been said already.

[akurtakov]

We can not afford continuing without a build - that's for sure. Improving/changing/etc. is always welcome and should come when this missing input/knowledge is gathered and the changes are better needed and executed in more integral way.

[jukzi]

good morning, ill take a look now

[jukzi]

The only "codgen" options that ecj changes in "org.eclipse.jdt.internal.compiler.batch.Main.initializeWarnings(String)" (when using parameter -properties {file}) seems to be org.eclipse.jdt.core.compiler.codegen.unusedLocal.
As far as i understand the unusedLocal is by default "optimize out" when not parameter is given but defaults to "preserve" when jdt UI or ecjs -properties is used

[srikanth-sankaran]

unusedLocal=preserve looks wrong to me and my gut feeling is that this is the reason why POP is replaced by ASTORE

See eclipse-platform/eclipse.platform#973 (comment)

[iloveeclipse]

There is now a jface error: https://download.eclipse.org/eclipse/downloads/drops4/I20231211-0340/buildlogs/comparatorlogs/buildtimeComparatorUnanticipated.log.txt

[iloveeclipse]

OK, I see Alex created #1655 for that.

[akurtakov]

I have opened #1655 as it's api_description file this time so most likely different issue.