False Negative: DoubleCheckedLockingWithInitRace.ql misses initialization races once the double-checked pattern is split across helpers or early returns.

先阅读 Likely Bugs/Concurrency/DoubleCheckedLockingWithInitRace.ql 以及 PosCase1_Var3.java 和 PosCase1_Var5.java 示例。跟踪查询如何处理辅助调用和提前返回，然后验证两种初始化竞争变体都被标记出来。

False Negative: DoubleCheckedLockingWithInitRace.ql misses initialization races once the double-checked pattern is split across helpers or early returns.

Version
codeql 2.24.3

Checker

• Checker id: Likely Bugs/Concurrency/DoubleCheckedLockingWithInitRace.ql
• Checker description: This checker detects a potential race condition in double-checked locking patterns where a field assignment inside a synchronized block may be visible to other threads before subsequent side-effect statements are executed.

Description of the false negative

These samples still publish f before the rest of the initialization work is complete. One variant moves the synchronized initialization into a helper, and the other rewrites the fast path as an early return, but the initialization race is the same.

Affected test cases

PosCase1_Var3.java

The helper call only hides the same double-checked-locking race where publication happens before later side effects.

// Double-checked locking with assignment to another field after the field assignment should be flagged as potential race condition.
package scensct.var.pos;

public class PosCase1_Var3 {
    private Object f;
    private Object otherField;

    public Object getF() {
        if (f == null) {
            initField();
        }
        return f;
    }

    private void initField() {
        synchronized (this) {
            if (f == null) {
                f = new Object();
                otherField = new Object();
            }
        }
    }
}

PosCase1_Var5.java

This still publishes the initialized object before a later field assignment completes, which is the race the query is meant to catch.

// Double-checked locking with assignment to another field after the field assignment should be flagged as potential race condition.
package scensct.var.pos;

public class PosCase1_Var5 {
    private Object f;
    private Object otherField;

    public Object getF() {
        if (f != null) {
            return f;
        }
        synchronized (this) {
            if (f == null) {
                f = new Object();
                otherField = new Object();
            }
            return f;
        }
    }
}

Cause analysis

The query appears too tied to one inline statement ordering pattern for double-checked initialization. Once the synchronized block is extracted into a helper or the fast path is expressed as an early return, it stops recognizing that f becomes visible before the remaining side effects complete.

That is a real concurrency gap. Initialization races often survive exactly this kind of refactoring.

References

None known.