alban-auzeill / jdt-core-ast-for-batch-of-compilation-units-bug

Reproduce a bug in jdt-core when parsing a batch of compilation units

Reproduce a bug in jdt-core when parsing a batch of compilation units

Reproduce the eclipse-jdt/eclipse.jdt.core issues/2050

Goal

Parsing a batch of 3 java files, A.java, B.java, C.java, using ASTParser.createASTs() method with a classpath that contains one missing class file to fully resolve B.java semantics.

Expected output

The createASTs() method should callback the FileASTRequestor 3 times, one for each file. Problems in the semantics should not prevent B.java to be parsed and the following files should also be parsed.

Actual output

The createASTs() method calls the FileASTRequestor only for A.java and silently ignore B.java and C.java files.

Steps to reproduce

Run:

mvn clean package

This will:

• prepare the binary files to reproduce the bug
  • module1/src/main/java/org/example/Module1.java -> module1/target/classes/org/example/Module1.class and Module1$Inner.class
  • module2/src/main/java/org/example/Module2.java -> module2/target/classes/org/example/Module2.class
• run module4-parser/src/test/java/org/example/ParserTest.java that reproduces the bug
  • check_accept_files_with_valid_semantic use a full classpath to resolve B.java semantics and is able to parse all 3 files
  • check_accept_files_with_missing_semantic use a classpath with a missing class file and is able to parse only A.java

The fact that the check_accept_files_with_missing_semantic test pass highlights the bug because of this assertion

containsExactlyInAnyOrder("A.java");

should be

containsExactlyInAnyOrder("A.java", "B.java", "C.java");

Workaround

Using org.eclipse.jdt:org.eclipse.jdt.core:3.35.0 instead of org.eclipse.jdt:org.eclipse.jdt.core:3.36.0 here fixes the issue. Running mvn clean package, the check_accept_files_with_missing_semantic will fail as expected.

Investigation

During B.java semantic resolution, the ProblemHandler is called while referenceContext is null which cause a AbortCompilation exception line 161. This exception may be related to issues/1915 but it's not because of an exception during semantic resolution that the parsing of the batch should silently stop. Stopping in the middle of a batch should not append, or should append by throwing an exception during the ASTParser.createASTs() call.