Individual task: generalization of aliases testing

[fnhirwa]

Hello,
Generally, alias functions have the same behavior as the original functions.
To generalize their tests and to make sure that we are testing them also. I modified the test_frontend_function and when we are testing the original function we will test all its aliases in that frontend by providing the namespace to that alias.

The tests are done the same way we test for frontend functions. I included some code lines to inner functions to avoid repetition of statements and added two main parameters to the helper function.
has_aliases: When the function has aliases we will set it to True
all_aliases: a list that holds the namespaces of all aliases to that function.

Whenever we make an alias we will go and update the unit test of that original function to include the namespace of the newly introduce alias.

Thanks🙂

[djl11]

looks good to me, just a small request to remove a redundant extra argument. Will request another review from @CatB1t and @sherry30 given that they have worked on this part of the code quite a lot.

[djl11]

Once @sherry30 and @CatB1t have given the green light, it's good to merge!

[CatB1t]

Just one question, why do we need to test the function again? when assigning an alias basically both names will bind to the same memory location, so for example mean_absolute_error and mae will bind to function 0x01 in memory, it doesn't do any copying or anything. I'd suggest just testing if alias is correctly binding to the original function and it's not misspelled or assigned to another function by mistake.

[fnhirwa]

Well from what I see we want to make sure that the aliases from ivy frontends are providing the same results as the corresponding one from the target framework, and you're right the memory binding for both mean_absolute_error and mae will be the same as they're one function technically. The reason, why I think we can do the testing again, is for example in the TensorFlow frontend we can have tf.eig and tf.linalg.eig which I think when we implement one as alias of the other we will need to test our function against both tf.eig and tf.linalg.eig
Thanks:)

[CatB1t]

@hirwa-nshuti I get your point on making sure the alias actually are aliases, but in that case we can compute 1 more and do 3 arrays comparison, instead of running 1 more redundant test. Regarding the second example, I'm not sure how tf.linalg.eig can be an alias for tf.eig, but even though, we shouldn't run our function twice.

[CatB1t]

Anyways, I think the optimization can be done later as only few functions are affected by that double-run instances, should keep in a TODO list though, once @sherry30 gives the greenlight feel free to merge 👍🏻

[sherry30]

Hi,

Just a small change requested, the rest looks good to me.
Feel free to merge after this change.
Thanks