Tom de Geus

I did not know that that could be done. Sound completely reasonable to support this. What would it involve? I glanced at pybind11. How does it work with CMake configuring...

Thanks @spectre-ns . I have not looked yet at the PR, but here is already a broad comment. I think it'd be great to be able to choose multiple algorithms...

That makes sense. The only thing I was wondering though: should we think about an API in which a user can change the backend without changing much of its code...

I would probably do that indeed. Maybe the others could comment @JohanMabille @wolfv @SylvainCorlay ?

Yes! I was just hoping that the other could share their experience from other parts of xtensor to keep a uniform way of having the drop-in backend

(Also, I'm working on a few things at the same time, so it will be not before later this week that I can look at this)

I think that the preprocessor suggestion you made excellent!

That seems reasonable because I think that there is no cast from `std::string` to `int`. You have to use `std::stoi` (see https://stackoverflow.com/a/7664227/2646505 ). I don't know if we have an...

As method `reshape` seems to support having one entry `-1`, see [docs](https://xtensor.readthedocs.io/en/latest/quickref/basic.html?highlight=reshape#reshape). In my opinion `reshape_view` should also support this. However, in your case `-1` seems to be casted to...

One thing you can do is ```cpp xt::xarray b = xt::empty({a.shape(0), a.shape(1) + 1}); xt::view(b, xt::all(), xt::range(3)) = a; a = b; ``` but there are probably more efficient solutions