Test naming changed from 1.8 to 1.9 for floats

[TWiStErRob]

I had this test:

@RunWith(TestParameterInjector.class)
...

@Test
public void testLightStateOnDevice(
	@TestParameter({"45", "0", "-3", "-6", "-9", "-12", "-15", "-18", "-21"}) float angle,
	@TestParameter Preset preset
) {

I'm using it for screenshot test recording, so the test names between previous and next versions has to match (otherwise hard to compare screenshots :). When updating from TPI 1.8 to 1.9 my PR failed:

java.lang.AssertionError: File src/test/snapshots/images/net.twisterrob.sun.android.logic_SunAngleWidgetViewScreenshotTest_testLightStateOnDevice[45.0,Nice_Preview].png does not exist

when inspecting the golden values on master (1.8) I found:

net.twisterrob.sun.android.logic_SunAngleWidgetViewScreenshotTest_testLightStateOnDevice[45,Nice_Preview].png

Since the only thing changed in the PR was the TPI version number, it's very likely that this had an unintended behavior change:
e390bab#diff-f0f3c84fc90c4da3bd5e1267e7b74d5774029389e86dfe1bba895740b8b886d0R142

[nymanjens]

Hi!

Thanks for letting me know.

The behavior you are seeing surprises me, because we have a unit test since 2020-08-05 that tests that

    @Test
    public void withFloat(@TestParameter({"1.2", "2"}) float param) {
      testNameToParameterMap.put(testName.getMethodName(), param);
    }

gets the test name

"withFloat[param=2.0]"

which seems to match your use case.

(I can't link to the test because it is removed when exporting to Github due to a dependency on textproto parsing)

I suspect this may be a combination of a change in TestParameterInjector and your JVM being different (e.g. Anroid vs openJDK).

[nymanjens]

Hmm, scrap that. Even before this library was open sourced, we had a unit test that ran on an Android emulator that checks that @TestParameter({"1.2", "2"}) float param) gives "withFloat[param=2.0]".

Since May 2022, we extended this test with multiple Android versions, starting at SDK version 15.

So that's probably not it either.

[nymanjens]

it's very likely that this had an unintended behavior change:
e390bab#diff-f0f3c84fc90c4da3bd5e1267e7b74d5774029389e86dfe1bba895740b8b886d0R142

(just reading this now)

It feels like it can't be a coincidence, I agree.

In your case, since parameter names are not present and the value is not an array, it should change value() to valueAsString(), which in turn becomes String.valueOf(value());.

So then String.valueOf(value()) should give a different result as MessageFormat.format("{0}", value()).

From the documentation of MessageFormat, it seems like it uses something like NumberFormat.getInstance(getLocale()).format(argument) to format the float. That could explain why you are seeing a different result than the Google test infrastructure.

If that's the case, I'd argue this is actually an improvement because it makes the result independent of the locale.

[TWiStErRob]

Ok, I'd say fair evaluation, nice! :)
Would you mind adding a hint in the changelog as this is a "breaking" behavior change (depending on usage of course).

[TWiStErRob]

Oh, btw my execution is on JVM, not Android devices. Paparazzi is amazing tech!

[nymanjens]

Would you mind adding a hint in the changelog as this is a "breaking" behavior change (depending on usage of course).

Done in cd15758