We Keep Coding

swagger-maven-plugin triggers Javadoc warning: element value must be a constant expression (but it is!) in Java annotation

Javadoc (via Maven) is giving me the following error in one my Java JAX-RS interface method signatures:
error: element value must be a constant expression
Here is my JAX-RS interface:
public interface FooResource {
#Consumes(APPLICATION_FORM_URLENCODED_UTF_8)
public void bar();
}
Javdoc gives the error for #Consumes. Here is the definition for APPLICATION_FORM_URLENCODED_UTF_8, which appears in MyAppConstants in the same project:
public static final String APPLICATION_FORM_URLENCODED_UTF_8 =
APPLICATION_FORM_URLENCODED + ";" + CHARSET_PARAMETER + "=UTF-8";
And here is the definition of APPLICATION_FORM_URLENCODED, which appears in javax.ws.rs.core.MediaType:
public final static String APPLICATION_FORM_URLENCODED = "application/x-www-form-urlencoded";
And here is the definition of CHARSET_PARAMETER, which also appears in javax.ws.rs.core.MediaType:
public static final String CHARSET_PARAMETER = "charset";
Now I ask you: what about APPLICATION_FORM_URLENCODED_UTF_8 is not constant at compile time?
The error message didn't say that I have to provide a literal. It said I had to provide a constant. So what about this is not a constant?
(I could almost swear that this worked at one time but suddenly stopped working.)
Update: Found cause, but still don't understand.
For some reason, merely including the swagger-maven-plugin in the POM will trigger this problem! My code doesn't change at all, but as soon as I add the following dependency, suddenly I get Javadoc warnings for my existing code!!!
<dependency>
<groupId>com.github.kongchen</groupId>
<artifactId>swagger-maven-plugin</artifactId>
<version>3.1.5</version>
</dependency>
How can a single dependency make Javadoc work differently on a code file? What is swagger-maven-plugin doing?

My best guess is that this happens because swagger-maven-plugin transitively (via io.swagger:swagger-core:1.5.13) an old version of the JAX-RS specification in javax.ws.rs:jsr311-api:1.1.1. Note that the JAX-RS 2 artifact ID is javax.ws.rs-api, Maven doesn't realize that they are different versions of the same JAR, and pulls them both in as dependencies. I can only guess that javax.ws.rs:jsr311-api in fact does not use constants for the variables in question. In any case, when I threw out swagger-maven-plugin and pulled in io.swagger:swagger-annotations (which was all I needed in this project for documentation), the problem went away.
See https://github.com/kongchen/swagger-maven-plugin/issues/543.

Way to test Metrics.Counter value as done with Aggregator

With Aggregators being removed as of Beam 2, is there a preferred way to test Counter value, like I did with Aggregators:
assertThat(tester.getAggregatorValue(fn.success)).isEqualTo(1);
assertThat(tester.getAggregatorValue(fn.failure)).isEqualTo(0);
assertThat(tester.getAggregatorValue(fn.empty)).isEqualTo(0);
Where tester is an instance od DoFnTester class and fn instance of my DoFn implementation, with aggregators success, failure and empty being final package-private visible fields.

This is similar to a question asked on the user#beam.apache.org list recently.
The answer there suggested installing a MetricsContainerImpl before running the test, and interacting with that, like Beam does for its existing tests (eg., LateDataDroppingDoFnRunnerTest).
They also encapsulated it in a TestRule:
#Rule TestMetrics metrics = new TestMetrics();
#Test
public void invalids() {
final DoFnTester<InputT, OutputT> doFnTester =
DoFnTester.of(fixture);
doFnTester.processElement(input);
assertThat(metrics,counterValue(fixture.ctr), is(1L));
}

How to mock riak java client?

I'm trying to unit test code that uses com.basho.riak:riak-client:2.0.0. I mocked all riak client classes and was hoping to get a useless but working test. However, this fails with a null pointer:
java.lang.NullPointerException
at com.basho.riak.client.api.commands.kv.KvResponseBase.convertValues(KvResponseBase.java:243)
at com.basho.riak.client.api.commands.kv.KvResponseBase.getValue(KvResponseBase.java:150)
at com.basho.riak.client.api.commands.kv.FetchValue$Response.getValue(FetchValue.java:171)
My test looks like this:
#Test public void test() {
RiakClient riakClient = mock(RiakClient.class);
#SuppressWarnings("unchecked")
RiakCommand<FetchValue.Response, Location> riakCommand = (RiakCommand<FetchValue.Response, Location>) mock(RiakCommand.class);
Response response = mock(Response.class);
when(riakClient.execute(riakCommand)).thenReturn(response);
Response returnedResponse = riakClient.execute(riakCommand);
when(response.getValue(Object.class)).thenReturn(new Object());
MyPojo myData = returnedResponse.getValue(MyPojo.class);
// Make assertions
}
How do you unit test code that uses the riak client? Eventually I would like to ensure that the expected type/bucket/key combination is used and that the expected RiakCommand is run.
EDIT: I dug more into the FetchValue class and found this structure:
FetchValue
- is public final
FetchValue.Response
- is public static,
- has a package-private constructor Response(Init<?> builder)
FetchValue.Response.Init<T> is:
- protected static abstract class Init<T extends Init<T>> extends KvResponseBase.Init<T>
And there is FetchValue.Response.Builder:
static class Builder extends Init<Builder>
- with build() that: return new Response(this);
I assume that Mockito gets lost somewhere among the inner classes and my call ends up in KvResponseBase.convertValues, where the NP is thrown. KvResponseBase.convertValues assumes a List<RiakObject> of values and I see no sane way of assigning it.

I have investigate a bit your case. I have reduce your example to this simple SSCCE:
import static org.mockito.BDDMockito.given;
import static org.mockito.Mockito.mock;
import org.junit.Test;
import com.basho.riak.client.api.commands.kv.FetchValue.Response;
public class RiakTest {
#Test
public void test() throws Exception {
Response response = mock(Response.class);
given(response.getValue(Object.class)).willReturn(new Object());
}
}
which throws this error:
java.lang.NullPointerException
at com.basho.riak.client.api.commands.kv.KvResponseBase.convertValues(KvResponseBase.java:243)
at com.basho.riak.client.api.commands.kv.KvResponseBase.getValue(KvResponseBase.java:150)
at com.basho.riak.client.api.commands.kv.FetchValue$Response.getValue(FetchValue.java:171)
at RiakTest.test(RiakTest.java:12)
After some digging, i think i have identified the problem. It is that you are trying to stub a public method which is inherited from a package (visibility) class:
abstract class KvResponseBase {
public <T> T getValue(Class<T> clazz) {
}
}
It seems that Mockito fails to stub this method so the real one is invoked and a NullPointerException is thrown (due to an access of a null member: values).
One important thing to note is that if this function invocation not fails, Mockito would show a proper error:
org.mockito.exceptions.misusing.MissingMethodInvocationException:
when() requires an argument which has to be 'a method call on a mock'.
For example:
when(mock.getArticles()).thenReturn(articles);
Also, this error might show up because:
1. you stub either of: final/private/equals()/hashCode() methods.
Those methods *cannot* be stubbed/verified.
Mocking methods declared on non-public parent classes is not supported.
2. inside when() you don't call method on mock but on some other object.
I guess it is a Mockito bug or limitation so i have open an issue in the Mockito tracker where i have reproduce your case with simple classes.
UPDATE
The issue i opened is in fact a duplicate of an existing one. This issue will not be fixed but a workaround exists. You may use the Bytebuddy mockmaker instead of the cglib one. Explanations could be found here.

You can not mock final classes and final and/or static methods with mockito. Note that static nested classes are fine. This is because mockito subclasses (I'm not 100% sure this is the exact operation, it uses CGLIB to generate classes) objects, but isn't allowed to override final methods or extend the final classes. For static methods no overriding is ever possible.
In your code you probably are trying to call a final class or method. It is hard to tell which class causes the problem, from your NullPointer stackstrace you should suspect the first object on it that you have mocked (going up starting from the testcase method). The method on the mock should not be calling any other methods (expect internal to mockito), so probably that is final, because you is does not seem to call a 'mocked' method.
In your case the stacktrace is not complete (as your testcase is not on it). In a quick look on the riak framework I couldn't find the method take a look at FetchValue$Response.getValue.
Also note the following. From the snippet you posted, I can not tell what you are testing in your testcase. All objects you create are mocks. Normally you have 1 (or a few) real classes that you are testing. The other classes (that interact with your classes under test) you mock, to be able to simulate complex behavior.

Follow up:
Thanks #gontard I was able to find this:
<dependency>
<!-- We need this fix: https://github.com/mockito/mockito/pull/171 to use mockito with Riak -->
<!--http://stackoverflow.com/questions/28442495/how-to-mock-riak-java-client#28474106-->
<groupId>org.mockito</groupId>
<artifactId>mockito-core</artifactId>
<version>2.0.52-beta</version>
<scope>test</scope>
</dependency>
That has the fixes included.
Unfortunately, if you use both Fetch and MultiFetch (likely), you're up a creek.
MultiFetch.Response is a final class (so you can use mockito, you need to use PowerMock)
FetchValue.Response has the issues you outlined, and can only be fixed with the beta mockito, not available with powermock yet...
Update, I figured out how to use both mockito & powermock together (until powermock upgrades):
<!-- We need this to mock Multi-Fetch responses from Riak, which are final -->
<!-- However, we need the beta version of mockito due to bugs (see below),
so we _cannot_ use the mockito api provided by powermock, do _not_ include _powermock-api-mockito, it'll mess stuff up -->
<dependency>
<groupId>org.powermock</groupId>
<artifactId>powermock-module-junit4</artifactId>
<version>1.6.4</version>
<scope>test</scope>
</dependency>
<!--If we don't include this, we get: -->
<!--java.lang.IllegalStateException:
Extension API internal error: org.powermock.api.extension.proxyframework.ProxyFrameworkImpl could not be located in classpath.-->
<!-- it looks like this is due to some discrepancy in packaging with mockito 2, this may be fixed in Fall 2016:
https://groups.google.com/forum/#!topic/powermock/cE4T40Xa_wc -->
<dependency>
<groupId>org.powermock</groupId>
<artifactId>powermock-api-easymock</artifactId>
<version>1.6.4</version>
</dependency>
<!-- We need this fix: https://github.com/mockito/mockito/pull/171 to use mockito with Riak -->
<!--http://stackoverflow.com/questions/28442495/how-to-mock-riak-java-client#28474106-->
<dependency>
<groupId>org.mockito</groupId>
<artifactId>mockito-core</artifactId>
<version>2.0.52-beta</version>
<scope>test</scope>
</dependency>

Mocking a dependency on a property file

I'm trying to write a unit test that involves using legacy code. Problem is, from what I can tell, the legacy code uses a key/value pair from a property file to initialize one of its final static private members, and I haven't the slightest as to where that property file might be (the entire application is quite huge).
So, in my test, I want to do something like this (using Mockito):
LegacyClass legacyClass = mock(LegacyClass.class);
I end up getting a ExceptionInInitializationError which indicates it can't find a certain property key.
In LegacyClass.java, there's:
private static final int LEGACY_PROPERTY =
Integer.parseInt(LegacyPropertyManager.getProp("legacy.property.key"));
Is there a way to write a test that uses this legacy class, even if the property key it's looking for doesn't exist? Can it be mocked somehow?

You might not get very far without a clever library like PowerMock. Note that your LegacyClass.java initializes this property in a static final field, which means that the initializer will run as soon as it's loaded. PowerMock uses deeper magic (read: bytecode manipulation) to allow you to mock the static getProp method you cited above.
You'll need to do the following to get started with PowerMockito:
#RunWith(PowerMockRunner.class)
#PrepareForTest(LegacyPropertyManager.class)
public class YourClass {
#Before public void stubLegacyPropertyManager() {
Mockito.when(LegacyPropertyManager.getProp("legacy.property.key"))
.thenReturn("42");
}
#Test public void yourTest() {
// ...
}
}
Note the class-level annotations, which respectively allow for PowerMock initialization and register the correct class for static-level mocking.

An example of mocking a properties file, completely ignoring where it is would be the following. You can create your own set of key/value pairs right in the mocked properties object.
Properties mockProperties = mock(Properties.class);
when((mockProperties.getProperty("keyName"))).thenReturn("value");
Getting this object to be used by your legacy class may involve code changes to the legacy class such as changing the private properties object to protected or creating a set method.
When unable to enhance the legacy code, I know you can venture into the area of partial mocks/spies (Mockito 1.8) such as described in the following post .. Mockito bypass static method for testing and its link to Effective Mockito. I haven't used them so I cannot offer more help. Good luck.

gwt-log and gwt-test-utils not playing nice together

I've got a project that has gwt-log logging lines scattered throughout. Now I'm trying to write some unit tests and nothing seems to be working.
Any class I test that uses the gwt-log facility causes the following exception to be raised:
Caused by: com.googlecode.gwt.test.exceptions.GwtTestConfigurationException:
A custom Generator should be used to instanciate
'com.allen_sauer.gwt.log.client.LogMessageFormatter',
but gwt-test-utils does not support GWT compiler API,
so you have to add our own GwtCreateHandler with
'GwtTest.addGwtCreateHandler(..)' method or to declare your
tested object with #Mock
I have no need for the logger to function during unit tests, I'd prefer to mock it away.
I've attempted to use Mockito to mock the logger, in a few different ways... obviously I have no idea what I'm doing here, none of the following code snippets helped the situation:
public class ClockTest extends GwtTest {
#Mock private LogMessageFormatter lmf;
...
or
...
#Before
public void init() throws Exception {
LogMessageFormatter lmf = mock(LogMessageFormatter.class);
...
Any clues on how to work this out would be most appreciated!

Colin is right, you have 2 ways to deal with your error :
1) Mock the LogMessageFormatter, or at a higher level, mock your Logger instance. gwt-test-utils provides a simple API for mocking with both Mockito or EasyMock : http://code.google.com/p/gwt-test-utils/wiki/MockingClasses
2) provide your own GwtCreateHandler to instanciate the LogMessageFormatter, or at a higher your own Logger instance.
Internally, gwt-log relies on GWT's deferred binding to instanciate a LogMessageFormatter object based on your configuration, which is parsed at compile time. It use GWT's generator API to create the LogMessageFormatter class, but gwt-test-utils is not able to use those kind of Generators.
You'll have to do it "by hand", with gwt-test-utils deferred binding support : GwtCreateHandlers.
Your "LoggerGwtCreateHandler" could use JDK's InvocationHandler and Proxy classes to write a proxy for the Logger interface which would simply silent each method call, since I guess you won't care about any log call in your tests.
Here is a discussion on how to write a GwtCreateHandler : https://groups.google.com/forum/?fromgroups#!topic/gwt-test-utils-users/r_cbPsw9nIE

From the error message you posted:
you have to add our own GwtCreateHandler with
'GwtTest.addGwtCreateHandler(..)' method or to declare your
tested object with #Mock
These are the two options you have to proceed. I've only just begun to work with gwt-test-utils, but the main premise is that it doesn't run the GWT compiler or Dev Mode, so it needs other ways to handle implementing 'magic' features like GWT.create. Its method is to either require you to mock the instance (this should be a fairly common idea in most of your tests for other objects involved in testing) or to provide something like a generator, and hook it up using GwtTest.addGwtCreateHandler.
Building a mock logger shouldn't be too bad, nor should implementing GwtCreateHandler - you just need to make something that has all the log methods. If you want the logging to work, then those methods need to actually invoke some other logger, like java.util.Logger, log4j, slf4j, etc but that is not required for just getting the tests to run (but may be handy for making sure that you logging works, or finding out why your test is failing.

for those still in pain with this damn problem here is what I managed to get (With a lot of pain too ...). It'll solve the conflict between Gwt-test-utils and Gwt-log.
You're of course welcome to modify the format method ;) :
#Before
public void correctLog() {
this.addGwtCreateHandler(new GwtCreateHandler() {
#Override
public Object create(Class<?> classLiteral) throws Exception {
if (classLiteral.isAssignableFrom(LogMessageFormatter.class)) {
return new LogMessageFormatter() {
#Override
public String format(String logLevelText, String category,
String message, Throwable throwable) {
return message + " : " + throwable.getLocalizedMessage();
}
};
}
return null;
}
});
}