New tests are required

[vkostyukov]

In order to test new operations new tests should be introduced. We have to test all the possible combinations of pair: Sparse-Dense. New tests should also test inPlace operations.

[ilya-murzinov]

@vkostyukov, hello!
I am looking for some jump-in issue to start contributing to la4j.
I think this issue is sutable for my purpose, am I right?

[vkostyukov]

Hi @ilya-murzinov, that's correct! New tests are something that is needed right now. I'm in the middle to introduce new operations, so we have to be able to test them.

I would suggest you to start with infrastructure. There are already tests in src/test package that tests paired operations such as testing the linear-system-solver. We somehow have to understand how to write a single test that take a couple of factories/vectors and launch it with all the possible combinations of factories.

This is not an easy issue but I believe it's possible. Please feel free to ask any questions in this issue and submit your PR either in operations or master branches.

[vkostyukov]

Hi Ilya @ilya-murzinov,

Plsee see the issue #174 for more details. I think it would be a good idea handle these issues together in order to not do the same work twice.

[ilya-murzinov]

@vkostyukov, ok
Right now I'm in the middle of "What the hell is that?!" stage, but I'm trying to figure out how is everything working here 😄

[vkostyukov]

Sure! I'm not a big expert in writing tests that's for sure. So, we can turn it in any direction we want. That is a good time for changes in a whole test package. As a contributor you a have a great chance to design everything from scratch.That is not exactly the the jump-in issue anymore. But I think it's more challengeable now.

[ilya-murzinov]

@vkostyukov, I have a good experience in software testing (I'm actually QA automation specialist right now), but I can't say that it's gonna be easy for me. I can only promise to do my best :)

If we want to start from scratch, it's a good idea to create a branch for it and just remove test folder there completely. Then we can start.

[vkostyukov]

@ilya-murzinov,

Right. I'm ok with rewriting everyting. The only thing I care is all the tests that are already in la4j. Some of them are good regression-tests. We somehow have to migrate tests to the new infrastructure rather then inventing new test-cases.

[ilya-murzinov]

@vkostyukov, yes, of course, I am going to base new tests on old ones.
First thing I'm donna do is to figure out how tests are working right now and then try to rewrite them one-by-one.

[vkostyukov]

@ilya-murzinov that's a nice plan.

Tests are primitive. Everything is bult around factoires (that can produce vectors/matrices). There are 4 factories: CRS (Compressed Row Storage), CCS (Compressed Column Storage), Basic1D, Basic2D that might generate 4 corresponding matrices and 2 vectors Compressed and Basic.

The current implementation is based on a simple idea having an abstract test classes (i.e., AbstractVectorTest) with 4 child classes that overrides a factory() method withit a corresponding factory instance. This works fine but it doesn't test the combinations of factories (like Sparse-Dense), which is becoming important, since new operations are cover these cases.

[ilya-murzinov]

@vkostyukov I'm very sorry, but I don't think I'm capable of solving this issue now. I'm experiencing lack of motivation for such a big thing. May be I should try to do something smaller for the start.
Again, I'm very sorry for letting you down :(

[vkostyukov]

Hi @ilya-murzinov,

No worries! How about take a look at the issue #172. Just refactor the methods to use a proper agument times. Also, don't forget to mark the old methods depricated.

[SamoylovMD]

According to #216, new equals() methods in AbstractVector and AbstractMatrix should be tested.