Sheehan Olver

You cut off the full error message and I can't see the types of `X` and `Y`

OK the issue is this line: https://github.com/JuliaArrays/BlockArrays.jl/blob/32990fcdf401a18921bbf4b15dc14c599e466c47/src/blocklinalg.jl#L47 I’m not sure the best way to solve it. Perhaps if we have a function `blocktype` that returns the the type of the...

The easy option is to make a new package BlockCuArrays.jl that overloads the right operations. I'm not sure if there's a way of getting it to work otherwise...

No I don't think so. In the case of `N = 1` this could be done by comprehension: ```julia [arr[globalrange(block_sizes, K)] for K=1:nblocks(block_sizes,1)] ``` should automatically infer the correct type...

This is unfortunately a non-trivial question as we need to think about breaking a `QRCompactWY` into blocks. In fact, `geqrt` allows specifying block size during factorization which might help: http://www.netlib.org/lapack/explore-html/dd/d9a/group__double_g_ecomputational_gaddcf152e87deec6123a1899f6f51101e.html#gaddcf152e87deec6123a1899f6f51101e...

Can you use a `PseudoBlockVector`? This should work

are you aware you made a matrix of matrices, not a block matrix?

It's not clear how things are stored in memory in your use case, is it like `BlockArray` where they are not contiguous?

OK as I said it's a hard problem to fix properly. The easiest would be to make a `copy` into a `PseudoBlockVector` though not sure whether that extra allocation is...

It’s breaking at the ` lmul!(Q’, b) ` call (the stuff about ` materialize! ` is only relevant if you wanted to support “traits”, that is, multiple array types that...