Add more tests

[intrigus]

We should add more tests.
I'm currently trying to setup the tests from https://github.com/robovm/robovm/tree/master/tests/dalvik, but I'll also have a look at the (original) tests from android.
I'll also have a look at https://github.com/robovm/drlvm-vts-bundle :)

(Comments for me)

030 Can't be fixed, due to finalizer
031 Can be fixed, fails, due to difference between javac/dx/jtransc
044 Failed, don't know why
046 Can be fixed, error, because they use other API
067 API diff between Android and Java SE
070 Not important

[soywiz]

Cool :)

030:
Finalizers could be supported, but just on some targets directly (C++, C#, Java, Flash, PHP). Javascript won't allow to do this, even with WeakMaps I tried to do a hack similar to one I did on flash but it would require being able to iterate the map or getting its length. But it doesn't allow it. So there is no way (at least standard) to determine if an object has been collected on JS.

044:
Not tried. But right now jtransc dynamic proxies just allow one interface. Which is the most common. It could support more interfaces if required, but that probably will have a cost con runtime performance on casts.

Threads:
Threads are currently not supported. But it is possible to support it on some targets: C++, Java, C#. In the future, when javascripts supports concurrent shared heap, we could try too.
We could support already threads that do not share global state.

[intrigus]

Hey, it's a little bit unclear, how exactly tests are run.

It would be great, if we can run the tests on all platforms haxe supports.
Could you maybe add a little example on how to do that?

[soywiz]

Yep I wanted do to do it. But also it would be cool to run them in parallel, since for example C++ could take much longer due to compilation times.

Right now there is a test that run in JS y PHP:

https://github.com/jtransc/jtransc/blob/master/jtransc-main/test/HaxeGenSuiteMiscTest.kt#L36

Though starting with 3.3.0-rc ( http://gamehaxe.com/wp-content/uploads/2016/07/Wwx2016-Hxcpp.swf with HXCPP_COMPILE_CACHE) following compilations even in other projects reuse already compiled classes ( https://www.youtube.com/watch?v=e64T96OOzKQ&index=6&list=PLyIetEt7wxr6yo_ARVaQv9UpeGbAYp_Z2 )
So maybe we could run more tests just with a first great compilation.

Haxe itself uses the travis matrix + environment variables to run cpp in parallel: https://github.com/HaxeFoundation/haxe/blob/development/.travis.yml#L88
https://github.com/HaxeFoundation/haxe/blob/development/.travis.yml#L117

The way testing a specific target is like this:

https://github.com/jtransc/jtransc/blob/master/jtransc-main/test/HaxeGenSuiteMiscTest.kt#L38

There is a lang argument which specifies the target.

The actual code instructing JTransc to compile and run it at the test level (without entering other code) can be found here:
https://github.com/jtransc/jtransc/blob/master/jtransc-main/test/HaxeTestBase.kt#L108

[intrigus]

I'm currently porting the tests from AndroidRT to OpenJDK, but there is one test, that is problematic, because the toString() method of an annotation isn't deterministic on OpenJDK, the test always fails...
I'm currently trying to fix it, but my approach is really hacky and heavily relies on reflection.
At least reflection is also heavily tested :D

[soywiz]

Great. I think the reflection is fine, if it works :)