Memory leak in compiler due to duplicated entries in typeCache map

[azinneera]

Description:

$subject was found when we analyzed the heap usage of the jballerina-tools distribution built from 2.0.0-alpha5-dev branch (commit: 8241385) for the following scenario:

Adding and deleting newlines to a source file in googleapis.calendar module [1]. Some of the dependencies were not there in the local cache so the package showed diagnostics for the unresolved modules (since the LS is run offline).

When we kept adding/deleting newlines we found out that the retained bytes increased significantly for org.wso2.ballerinalang.compiler.util.Name [2]. When analyzing further with @pubudu91 we figured out that the memory leak is due to duplicated entries in the typeCache map resulting in unnecessarily held BSymbol objects.

Steps to reproduce:

Duplication of entries can be reproduced with ballerina-lang as follows:

1. Write the below test case in io.ballerina.projects.test.TestBuildProject [3]:

@Test
    public void testMemoryLeak() {
        BuildProject project = BuildProject.load(Paths.get("<path-to-module-ballerinax-googleapis.calendar>/calendar"));
        project.currentPackage().getCompilation();
        for (int i = 0; i < 10; i++) {
            DocumentId documentId = project.currentPackage().getDefaultModule().documentIds().stream().findFirst().get();
            String sourceCode = project.currentPackage().getDefaultModule().document(documentId).syntaxTree().toSourceCode();
            project.currentPackage().getDefaultModule().document(documentId).modify().withContent(sourceCode).apply();
            project.currentPackage().getCompilation();
        }
    }

2. Debug and evalute TypesFactory.getInstance(project.projectEnvironmentContext().getService(CompilerContext.class)).typeCache for each iteration and you can observe it growing

Affected Versions:
At least Alpha 5.4

OS, DB, other environment details and versions:
Tested on Ubuntu 20.04

[1] https://github.com/ballerina-platform/module-ballerinax-googleapis.calendar
[2] https://github.com/ballerina-platform/ballerina-lang/blob/master/compiler/ballerina-lang/src/main/java/org/wso2/ballerinalang/compiler/util/Names.java
[3] https://github.com/ballerina-platform/ballerina-lang/blob/master/project-api/project-api-test/src/test/java/io/ballerina/projects/test/TestBuildProject.java

[pubudu91]

When working on fixing this issue by implementing equals() and hashCode() for BType and BSymbol and their sub classes, faced an issue with cyclic types. Basically attempting to implement hashCode() for them resulted in stack over flow errors. Even if we got around the stack over flow issue, there's the concern about correctness of using mutable fields of a class in the hash code computation. Therefore, we decided not to go down this path. Instead, we'll remove the type cache for now and see.