DLS_DEAD_LOCAL_STORE false-positive in switch

[boris-petrov]

switch (x) {
    case Map<?, ?> m -> return 1;
    default -> return 2;
}

This leads to a DLS_DEAD_LOCAL_STORE error for the m variable. That's a false-positive because Java doesn't allow the variable name to be omitted.

[gtoison]

This might be solved if this JEP 456: "Unnamed Variables & Patterns" is implemented: https://openjdk.org/jeps/456

[boris-petrov]

@gtoison yes, that's true, once Java 22 is available, the current behavior will be correct. But until then it isn't. 😄

[gtoison]

@gtoison yes, that's true, once Java 22 is available, the current behavior will be correct. But until then it isn't. 😄

... and once everybody has switched to Java 22 😄
It also remains to be seen how SpotBugs will analyse the bytecode produced with unamed variables

[boris-petrov]

I tried SpotBugs 4.8.4 which I thought contains the fix for this but it didn't fix any of my cases. Am I missing something or is this still not released?

[gtoison]

Hum, I thought this was fixed by #2828 ...
The test cases in https://github.com/gtoison/spotbugs/blob/b5893ac7a825413dd6085135318ff7a5a7c49e49/spotbugsTestCases/src/java21/Issue2782.java seem to be handled correctly so maybe your code is a bit different?
What compiler (and version) are you using? I suppose that different compiler might be outputting different bytecode here.

The change in #2828 looks for a CHECKCAST followed by an ASTORE at the beginning of a switch case

[boris-petrov]

% java --version
openjdk 21.0.2 2024-01-16
OpenJDK Runtime Environment (build 21.0.2+13)
OpenJDK 64-Bit Server VM (build 21.0.2+13, mixed mode, sharing)

As for the code that doesn't work - as I said, none of my cases (around 10 or so) have been fixed... but the original one I've given here seems to work which is strange. 😄 Here are a couple more:

return switch (value) {
	case Map<?, ?> map -> map.size();
	case Set<?> set -> set.size();
	case List<?> list -> list.size();
	case Collection<?> collection ->
		throw new UnsupportedOperationException("Error " + value.getClass());
	case null, default -> 0;
};

Gives the warning on collection.

private static String getValueType(Object value) {
	return switch (value) {
		case List<?> list -> "list";
		case Map<?, ?> map -> "map";
		case Number number -> "number";
		case null, default -> "other";
	};
}

Gives the warning on all list, map and number.

[gtoison]

Thanks for the samples, I have added them in #2937 and they are flagged indeed
I'd have to investigate why they are not when running the test on my computer

[maxl2287]

Updating on 4.8.5.0 is removing this DLS for such switches.

[gtoison]

You might run into false positives with 4.8.5 because I had not realized that the bytecode instruction for a switch is either LOOKUPSWITCH or TABLESWITCH. Your compiler might use TABLESWITCH in some cases.
In 4.8.5 only LOOKUPSWITCH is handled, once released the fix in #2937 should handle both.