Do not assume that there is no memory aliasing between output of mutating functions and their input arguments.

[mateuszbaran]

It can lead to hard-to-find bugs, see #392 .

[kellertuer]

I agree, the should not assume that.
But this is phrased quite vague – should we add the test I did generally to test_manifold to have this in mind?

[mateuszbaran]

It's a bit vague because I'm not sure what exactly should be done. Checking that all functions on all manifolds correctly handle aliasing solves this issue but it would add a relatively long time to each CI run.

[kellertuer]

It would not be all functions, but sure, epx/log/retract/inverse_retracte/vector_transport_* at least.

[kellertuer]

So lets make the list precise

• embed!
• exp!
• inverse_retract
• log!
• mid_pont!
• project!
• retract!
• vector_transport_to!
• vector_transport_direction!
• vector_transport_along! (maybe not since we do not have good generic tests until now I think)

I think we do not have to check zero_vector!. I am not sure about pole_ladder! and shields_ladder! since they are generic in ManifoldsBase. We could introduce a parameter in test_manifold to activate them?

[mateuszbaran]

Thanks for the list, checking them makes sense. A parameter in test_manifold that activates them makes sense.