Problems: Extracting the complete Method Call Hierarchy of a complicated Java method.

[oriana19993926782]

I'm attempting to prompt chatGPT to generate unit tests for a method within the apache/shardingsphere-elasticjob project.

For this purpose, I need to

1. Extract the body of the target method, which I've termed "Method A".
2. Get the complete call hierarchy or call graph of "Method A", as it might invoke another method, "Method B", which could further call "Method C", and so on. This is crucial for chatGPT to understand the exact behavior of "Method A". (Method A B C here are self-defined methods, not from some third-party library)

The body of my target method setUpJobConfiguration(String, JobConfiguration) is down below, where you can see it further calls
Non-static methods: checkConflict Job(String, JobConfiguration) (for example)
Static methods: YamlEngine.marshal, JobConfigurationPOJO.fromJobConfiguration

public JobConfiguration setUpJobConfiguration(final String jobClassName, final JobConfiguration jobConfig) {
    checkConflictJob(jobClassName, jobConfig);
    if (!jobNodeStorage.isJobNodeExisted(ConfigurationNode.ROOT) || jobConfig.isOverwrite()) {
        jobNodeStorage.replaceJobNode(ConfigurationNode.ROOT, YamlEngine.marshal(JobConfigurationPOJO.fromJobConfiguration(jobConfig)));
        jobNodeStorage.replaceJobRootNode(jobClassName);
        return jobConfig;
    }
    return load(false);
}

Two big problems I ran into that cause me only getting an incomplete call graph:

1. Unable to capture the specific overridden implementation method body of almost every interface method invoked within this method body, directly or indirectly. This also means I'm unable to further navigate the call graph within these implementation bodies.

Example Interface method:void persist(String key, String value);

Example overridden implementation on this interface method:

  @Override
    public void persist(final String key, final String value) {
        try {
            if (!isExisted(key)) {
                client.create().creatingParentsIfNeeded().withMode(CreateMode.PERSISTENT).forPath(key, value.getBytes(StandardCharsets.UTF_8));
            } else {update(key, value);}
        //CHECKSTYLE:OFF
        } catch (final Exception ex) {
        //CHECKSTYLE:ON
            RegExceptionHandler.handleException(ex);
        }}

2. Unable to get the body of self-defined static methods, such as_YamlEngine.marshal_, JobConfigurationPOJO.fromJobConfiguration. (Note that when I deleted 'and overriddenMethod.fromSource()' and 'overriddenMethod.getBody(),' from my query, I can see these two static methods in results, but I still cannot get the body of these two static methods, which are not from some third-party compiled library.)

Below is the codeQL query I wrote to achieve my purpose. I'm currently stuck here and greatly appreciate guidance on these challenges and any enhancements that can be made to my query:

import java

// Predicate to determine if a 'caller' either directly or indirectly calls another 'callee'.
predicate callsRecursively(Callable caller, Callable callee) {
  caller.polyCalls(callee)

  or 
  // Checks if 'caller' calls an interface method and 'callee' is a potential target of that call.
  exists(Callable interfaceMethod, MethodAccess ma |
    caller.polyCalls(interfaceMethod) and
    ma.getMethod() = interfaceMethod and
    ma.getAChildExpr*().(MethodAccess).getMethod() = callee 
  )
  
  or
  // Recursively checks if 'caller' indirectly calls 'callee' through some other method 'mid'.
  exists(Callable mid|
    caller.polyCalls(mid) and
    callsRecursively(mid, callee)
  )
}

predicate hasMethodCall(Expr expr) {
  expr instanceof MethodAccess
  or
  exists(Expr child |
    child = expr.getAChildExpr() and
    hasMethodCall(child))
}


class TargetMethod extends Method {
  TargetMethod() {
    this.getName() = "setUpJobConfiguration" 
  }
}

from TargetMethod tm, Method calledMethod, Method overriddenMethod
where
callsRecursively(tm, calledMethod) and
// Check if 'overriddenMethod' either overrides 'calledMethod' or is the same as 'calledMethod'
// and ensure that 'overriddenMethod' is not abstract
(overriddenMethod.overrides*(calledMethod) and not overriddenMethod.isAbstract() or overriddenMethod = calledMethod) 
// Ensure that 'overriddenMethod' are from source code and not from compiled libraries
and overriddenMethod.fromSource()
  
select
  tm.getSignature(), 
  tm.getBody(), 
  "This is a target method",
  calledMethod,
  overriddenMethod,
  overriddenMethod.getBody(),
  "This method is called directly or indirectly in the target method."

[aibaars]

I tried running the query on apache/shardingshere but got no results because that code base does not contain a method named setUpJobConfiguration.

The missing body of YamlEngine.marshal is strange; that class exists in the repository. However, it seems to be missing somehow from the CodeQL database. Instead CodeQL reports /home/runner/work/shardingsphere/shardingsphere/infra/util/target/classes/org/apache/shardingsphere/infra/util/yaml/YamlEngine.class as the path of the YamlEngine class. Class files do not contain the source code of the method body, so that explains why getBody does not return anything. It looks like CodeQL did not "see" the YamlEngine.java source file. This typically happens when a source tree contains build artefacts from a previous compilation run. In such cases the Java compiler may re-use existing class files and skip the source files. As a result CodeQL won't analyse the source file. In case of apache/shardingsphere, I don't understand how this could happen. The project seems to run mvn clean install which should get rid of any .class files.

Looking at the documentation it looks like polyCalls already includes methods that override other methods. I think the Method overriddenMethod should not be needed. You can probably replace caller.callsRecursively(callee) with caller.polyCalls+(callee).

I think the following block is not doing what you need; it finds cases where a caller calls some other method and then finds all accesses to that other method anywhere in the code.

  // Checks if 'caller' calls an interface method and 'callee' is a potential target of that call.
  exists(Callable interfaceMethod, MethodAccess ma |
    caller.polyCalls(interfaceMethod) and
    ma.getMethod() = interfaceMethod and
    ma.getAChildExpr*().(MethodAccess).getMethod() = callee 
  )

[oriana19993926782]

Really really sorry, I've made a stupid mistake. It should be apache/shardingsphere-elasticjob instead of apache/shardingsphere. That's why you got 0 results running my query. I have already corrected this mistake in my question.

[aibaars]

I think the problem is in the CodeQL database . Some source files are missing and methods resolve to class files instead. For example the definition of the persist method is found in RegistryCenter.java, but the implementations are found only in class files ZookeeperRegistryCenter.class and ZookeeperRegistryCenterTestUtil.class

from Method m, string f
where m.getName() = "persist" and f = m.getFile().getRelativePath()
select m, f

#	m	f
1	persist	elasticjob-infra/elasticjob-registry-center/elasticjob-registry-center-api/src/main/java/org/apache/shardingsphere/elasticjob/reg/base/RegistryCenter.java
2	persist	elasticjob-infra/elasticjob-registry-center/elasticjob-regitry-center-provider/elasticjob-registry-center-zookeeper-curator/target/classes/org/apache/shardingsphere/elasticjob/reg/zookeeper/ZookeeperRegistryCenter.class
3	persist	elasticjob-infra/elasticjob-registry-center/elasticjob-regitry-center-provider/elasticjob-registry-center-zookeeper-curator/target/test-classes/org/apache/shardingsphere/elasticjob/reg/zookeeper/util/ZookeeperRegistryCenterTestUtil.class

The same thing happens for the marshal method.

#	m	f
1	marshal	elasticjob-infra/elasticjob-infra-common/target/classes/org/apache/shardingsphere/elasticjob/infra/yaml/YamlEngine.class

I have no idea why the database is incomplete, but it explains why you aren't able to get the bodies of those methods. We'll need to investigate this further.

[aibaars]

The cause of the incomplete database is that CodeQL did not support Lombok and skips any files that import lombok annotations. Support for Lombok is experimental and can be enabled by setting the environment variable CODEQL_EXTRACTOR_JAVA_RUN_ANNOTATION_PROCESSORS. You can create a complete database with the CodeQL CLI as follows:

CODEQL_EXTRACTOR_JAVA_RUN_ANNOTATION_PROCESSORS=true codeql database create -ljava -s shardingsphere-elasticjob shardingsphere-elasticjob-database

[oriana19993926782]

Thank you so so so so much💗. This is very very helpful!!!!!! I have used CodeQl CLI to create the database, run my query, and have seen the implementations with methods invoked within and static methods as well.

Also, I took your advice that I could just use polyCalls+. I feel a bit silly for having gone the long way and writing a callsRecursively predicate 😂. My query looks much neater now. But it is a little strange that even though this polyCalls+ works like magic, there doesn't seem to be any documentation on it, only on polyCalls.

On another note, is there a way to enable this CODEQL_EXTRACTOR_JAVA_RUN_ANNOTATION_PROCESSORS=true feature if I want to add this database from Github in VS Code instead? Or does it has to be created using CodeQL CLI? Also, I could not find documentation regarding this CODEQL_EXTRACTOR_JAVA_RUN_ANNOTATION_PROCESSORS neither. If you come across it, could you kindly share the link with me?

[aibaars]

The + in polyCalls+ is a QL feature that means apply polyCalls 1 or more times. There is also * which means zero or more times. You can use * and + on other predicates too. It's a short way of writing a simple recursion.

CODEQL_EXTRACTOR_JAVA_RUN_ANNOTATION_PROCESSORS is an undocumented feature flag. It will probably be removed once Lombok support is no longer experimental. Most databases downloaded from GitHub via VSCode will likely work with Lombok in a few weeks time.

[aibaars]

@oriana19993926782 The database for apache/shardingsphere-elasticjob has been updated and should contain the previously missing Java sources files now.