[Enhanced Switch][Null] Inconsistent nullness propagation

[srikanth-sankaran]

Found by code inspection:

With null analysis enabled, we report two warnings in foo(Object) Null comparison always yields false: The variable o cannot be null at this location but none in foo(X) (with redundant null check set to warning)

import org.eclipse.jdt.annotation.NonNull;

public class X {
	static @NonNull Object foo(Object o) {
		switch (o) {
			case String s -> {
				if (o == null) {
					System.out.println("o cannot be null at all!");
				}
				System.out.println();
			}
			default -> {
				if (o == null) {
					System.out.println("o cannot be null at all!");
				}
				System.out.println();	
			}
		}
		return new Object();
	}
	
	static @NonNull Object foo(X o) {
		switch (o) {
			case X s  -> {
				if (o == null) {
					System.out.println("o cannot be null at all!");
				}
				System.out.println(s);
			}
		}
		return new Object();
	}
}

This is because of the incorrect asymmetry in org.eclipse.jdt.internal.compiler.ast.SwitchStatement.analyseCode(BlockScope, FlowContext, FlowInfo) between a regular case and a default.

[srikanth-sankaran]

This is because of the incorrect asymmetry in org.eclipse.jdt.internal.compiler.ast.SwitchStatement.analyseCode(BlockScope, FlowContext, FlowInfo) between a regular case and a default.

Not entirely sure of this. Need to study in detail.

At the outset it does look like there is some inconsistent behavior. Cause may be different than hypothesized.

[srikanth-sankaran]

The first commit from #3382 also this issue - basically the isNullHostile method's wrong result in the presence of total patterns was inhibiting a diagnostic from being reported.