James Lucas

> I'm a little worried that this is a dangerous pattern to start using. I understand and agree with the concern. However - I'm not sure that there is a...

> Do you think something like this could maybe work? It looks quite appealing in the way you've sketched it out. However, I'm not convinced we can allow the user...

> My idea for `UninitializedBuffer` is to not expose its data in any way whatsoever besides the accessors in the `MatrixBuffer` trait. I think that this could work actually. At...

@bluss @Andlon -I have to admit that I don't know too much about BLIS. I also looked into it around a year ago when you first published your `matrixmultiply` crate...

I think the BLIS/BLAS composition you described sounds like a reasonable approach. I'm not sure exactly how it will work (due to a lack of familiarity with BLIS) but it...

I think this is a good change from the user perspective! @Andlon am I missing anything here?

Thank you so much for opening this! I think it will take me a little while to digest this, but from what I do understand it will be really valuable.

I was just reading through this again @bluss . How do you differentiate between the contiguous and non-contiguous cases in ndarray? A contributor suggested we add specialist types for contiguous...

Yes thanks for pointing this out! We are aware of quite a few SVD problems. Hopefully in the near future we will have a more robust implementation.

Ah, you're right. I'll try to port the branch this weekend - though I'm not sure how close it is to being usable.