semanticdb-javac: exceptions in reportException() use Kind.ERROR, breaking compilation for projects with partial classpaths (e.g. Spark)

[rpalcolea]

When semanticdb-javac encounters an exception during the ANALYZE phase (e.g. a CompletionFailure for a missing class), it calls reportException(), which reports the failure as Diagnostic.Kind.ERROR. This causes javac to exit with a non-zero status, failing the entire compilation — even though the catch block comment explicitly says "we don't want to stop the compilation".

Root cause

SemanticdbTaskListener.java lines 80–93 — the intent is correct, the implementation is not:

} catch (Throwable ex) {
    // Catch exceptions because we don't want to stop the compilation even if this
    // plugin has a bug. We report the full stack trace because it's helpful for bug reports.
    Throwable throwable = ex;
    if (e.getSourceFile() != null) {
        throwable = new CompilationUnitException(
            String.valueOf(e.getSourceFile().toUri().toString()), throwable);
    }
    this.reportException(throwable, e);  // ← ends up as Kind.ERROR → build fails
}

SemanticdbTaskListener.java lines 102–108 — reportException calls reporter.error():

private void reportException(Throwable exception, TaskEvent e) {
    ByteArrayOutputStream baos = new ByteArrayOutputStream();
    PrintWriter pw = new PrintWriter(baos);
    exception.printStackTrace(pw);
    pw.close();
    reporter.error(baos.toString(), e.getCompilationUnit(), e.getCompilationUnit());
    //      ^^^^^ Kind.ERROR → javac exits non-zero → build fails
}

SemanticdbReporter.java lines 54–61 — error() uses Kind.ERROR:

public void error(String message, Tree tree, CompilationUnitTree root) {
    trees.printMessage(
        Diagnostic.Kind.ERROR,   // ← this is what fails the build
        String.format("semanticdb-javac: %s", message), tree, root);
}

Concrete trigger

Projects that use Apache Spark (or any library with anonymous inner classes compiled from Scala) have types like org.apache.spark.sql.types.DataType$1 on the Scala classpath but not on the Java compile classpath. When semanticdb-javac tries to resolve these during deep type analysis, it throws:

semanticdb-javac: CompilationUnitException: file:/path/to/BoundReflectFunction.java
  Caused by: CompletionFailure: class file for org.apache.spark.sql.types.DataType$1 not found

1 error
FAILURE: Build failed with an exception.
* What went wrong: Execution failed for task ':compileJava'.

The build works fine without semanticdb-javac. Removing it restores the build.

Proposed fix

Add a warning() method to SemanticdbReporter and update reportException() to use it instead of error(). Intentional error paths (e.g. -no-relative-path:error mode) are left unchanged.

// SemanticdbReporter.java
+  public void warning(String message, Tree tree, CompilationUnitTree root) {
+    trees.printMessage(
+        Diagnostic.Kind.WARNING, String.format("semanticdb-javac: %s", message), tree, root);
+  }

// SemanticdbTaskListener.java — reportException()
-    reporter.error(baos.toString(), e.getCompilationUnit(), e.getCompilationUnit());
+    reporter.warning(baos.toString(), e.getCompilationUnit(), e.getCompilationUnit());

Also change SemanticdbReporter.exception() (the Throwable overload) from Kind.ERROR to Kind.WARNING for consistency.

The full stack trace is still reported — useful for bug reports — but as a warning. The build succeeds with partial semanticdb output rather than aborting entirely.

[linear]

OSS-24208