Type derive (simpler)

[cstamas]

Instead to "smear" this feature across Maven and Resolver classes, for start let's keep it "confined" with single class: the TypeDeriver.

Later we can see where to go further with it.

This PR also includes bugfix, where Maven DefaultType implements ArtifactType, while in reality it does not (violates contract by returning null when no classifier present).

Supersedes #11373

[gnodet]

What's the benefit of this decoupling? It does not seem necessary.

[cstamas]

The decoupling is a must as we figured: this class violates (maven or resolver, dependending how you look at it) contract for Type: classifier returns null instead of "" (resolver).

[desruisseaux]

Thanks. Just tried with the processor-type integration test in the compiler plugin. I still have the dependency on the classpath, but I can try to run in the debugger during the weekend.

[cstamas]

@desruisseaux that is weird, as I have locally (master of m-compiler-p):

[INFO] --- invoker:3.9.1:integration-test (integration-test) @ maven-compiler-plugin ---
[INFO] Running 4 setup jobs:
[INFO] Building: setup_annotation-verify-plugin/pom.xml
[INFO]           setup_annotation-verify-plugin/pom.xml ........... SUCCESS (5.192 s)
[INFO] Building: setup_jar_automodule/pom.xml
[INFO]           setup_jar_automodule/pom.xml ..................... SUCCESS (1.464 s)
[INFO] Building: setup_jar_module/pom.xml
[INFO]           setup_jar_module/pom.xml ......................... SUCCESS (1.657 s)
[INFO] Building: setup_x/pom.xml
[INFO]           setup_x/pom.xml .................................. SUCCESS (1.813 s)
[INFO] Setup done.
[INFO] Building: processor-type/pom.xml
[INFO]           processor-type/pom.xml ........................... SUCCESS (1.481 s)

Invoked as:

[cstamas@angeleyes maven-compiler-plugin (master)]$ mvn -v
Apache Maven 4.1.0-SNAPSHOT (4ff85c235b1ecee29c6a71b16d9cf40dfb841c2d)
Maven home: /home/cstamas/.sdkman/candidates/maven/latest-master
Java version: 21.0.8, vendor: Eclipse Adoptium, runtime: /home/cstamas/.sdkman/candidates/java/21.0.8-tem
Default locale: en_US, platform encoding: UTF-8
OS name: "linux", version: "6.17.6-300.fc43.x86_64", arch: "amd64", family: "unix"
[cstamas@angeleyes maven-compiler-plugin (master)]$ mvn clean install -P run-its -Dinvoker.test=setup_*,processor-type

[cstamas]

Aha, I think I see it now:

[cstamas@angeleyes processor-type (master)]$ cat annotation-user/target/javac.args 
--release 17
--module-version 1.0-SNAPSHOT
--processor-path "../annotation-processor/target/classes"
--class-path "../annotation-processor-dep/target/classes"
--source-path "src/main/java"
-d "target/classes"
"src/main/java/user/SimpleAnnotation.java"
"src/main/java/user/SimpleObject.java"
[cstamas@angeleyes processor-type (master)]$ 

[desruisseaux]

Yes, the integration test passes because the line that cause a failure is commented out. If you open the following file:

target/it/processor-type/annotation-processor/src/main/java/processor/SimpleAnnotationProcessor.java

then uncomment the second line below:

        // TODO: pending https://github.com/apache/maven/pull/11373
        // dependency.AnnotationProcessorDependency.foo();

then, we get the test failure. You can also check that if the value on --class-path is moved to --processor-path, it works.

[cstamas]

Blocked by apache/maven-resolver#1648

[cstamas]

@desruisseaux if this build passes OK, you will have a Maven build (available as GH CI artifact here: https://github.com/apache/maven/actions/runs/19105278503?pr=11380) w/ resolver fix incorporated (will use 2.0.14-SNAPSHOT resolver)

[cstamas]

Using that Maven along with latest Toolbox you can inspect the tree output from Maven and how types are derived.

[desruisseaux]

Thanks! Will try this weekend. No worry for the github actions, I always build Maven core locally anyway.

[cstamas]

Funnily, GH Actions revealed several "leaks" (UT and IT bubbled up to checkout root), so I am just using this as a good test to fix all those leaks 😄

[desruisseaux]

I cannot really evaluate this part of Maven code, but I tested with the Maven Compiler Plugin and confirm that it worked. I was able to re-enable a processor dependency test which was commented-out.

[cstamas]

Note: this PR and feature depends on latest (not yet released) Resolver 2.0.14! Will revisit once it is out.