Javac not running annotation processors

Question

Problem

I'm revisiting a maven project that I haven't touched in at least a year. I'm pretty sure it was compiling successfully when I left it (there was still a working jar in the target directory), but now compilation fails because generated classes are missing.

What I tried

• Verified there are indeed no sources in target/generated-sources/annotations (none present)
• Ensured javac wasn't invoked with -proc:none (it wasn't)
• Ran mvn clean dependency:unpack-dependencies -Dmdep.useSubDirectoryPerArtifact=true to ensure the expected dependencies are on the classpath and that they contain a valid META-INF/services/javax.annotation.processing.Processor entry (yes there were multiple, including org.immutables.processor.ProxyProcessor )
• Followed these steps from How to make sure javac is using an annotation processor and troubleshoot when it is not:
  1. Take the javac options from the maven debug log
  2. Remove -nowarn option and add -verbose -XprintRounds -XprintProcessorInfo -Xlint -J-verbose. NOTE: I also had to add the relative path to the main class, or javac would complain there are no sources.
  3. Verify if class of processor is loaded (not present)
  4. Look for at least one processing round log (none present)
  5. Make sure annotations are present in the list of the round log (no round log)
• Added -proc:only option to javac command
• Also added -processor org.immutables.processor.ProxyProcessor to javac command (processor now loaded, but still no rounds printed and no classes generated)
• Changed the processor option to -processor org.immutables.processor.ProxyProcessorXXX to see if it would make a difference (it did, it now printed a warning stating that processing was requested with proc:only, but no processors were found)
• Checked to see if that warning appeared if I didn't use the -processor option, which should make javac detect processors from class path (it didn't show the warning, which suggests it's detecting processors yet the log doesn't show any sign of it)
• Tried above after switching from JDK1.8 to JDK9 (using jenv)
• Installed Zulu 1.7 and tried above after switching to it (using jenv)

Possible factors

• I did an entirely clean install of macOS
• I ran macOS Sierra before, currently High Sierra
• Before I didn't have JDK9 installed
• I switched to using Gradle a long time ago, so I might have forgotten an important detail about maven

Command options

This is the javac command I'm currently using:

javac -d ./target/classes -classpath ./target/classes:$HOME/.m2/repository/com/google/dagger/dagger/2.4/dagger-2.4.jar:$HOME/.m2/repository/com/google/dagger/dagger-compiler/2.4/dagger-compiler-2.4.jar:$HOME/.m2/repository/com/google/auto/factory/auto-factory/1.0-beta3/auto-factory-1.0-beta3.jar:$HOME/.m2/repository/org/immutables/builder/2.3.9/builder-2.3.9.jar:$HOME/.m2/repository/org/immutables/value/2.3.9/value-2.3.9.jar:<LONG LIST OF DIRECT AND TRANSITIVE DEPENDENCIES HERE> -sourcepath ./src/main/java: -s ./target/generated-sources/annotations -verbose -XprintRounds -XprintProcessorInfo -Xlint -J-verbose -processor org.immutables.processor.ProxyProcessor -proc:only ./src/main/java/com/mycompany/myproject/Main.java

NOTE: for better readability, I put each argument on a new line, replaced absolute paths with $HOME/ and ./ and I omitted most dependencies.

Question

What am I missing here? Any suggestions or pointers would be greatly appreciated.

Answer 1

javac appears to completely and silently skip annotation processing for some parse failures.

For me a duplicated annotation caused by a merge mistake caused thousands of failures because my annotation processors didn't run - so the real error got completely lost.

@Data @AllArgsConstructor @NoArgsConstructor @EqualsAndHashCode(callSuper = false)
@EqualsAndHashCode(callSuper = false)

Answer 2

We had the same issue on a git branch, while the main branch built fine. The root cause turned out to be a .java source with a duplicate definition of a private void function. This duplicate was caused by a merge from the main branch. At first we did not notice it, because the merge did not cause conflicts. The problem occurred with Amazon Corretto 8 and 11 compilers. After removing the duplicate private void function, the java compiler started generating sources as before.

Answer 3

As far as I can remember maven generates a marker file or marker directory (starting with a dot) for generated sources. But it's also a long time ago...