Daniel Karrasch
Daniel Karrasch
This is now new-feature-only, nothing breaks (except for the lack of complete type inference for the convenience constructor). Thanks @fredrikekre for retriggering some thought, thanks @JeffFessler for your comments and...
BTW, I changed the default values of `dest` and `source` in `_compositemul!` to `nothing`, and moved the allocation to the function body (if necessary). We don't use these temporary arrays...
Out of curiosity, why not make the overloaded functions available in `StatsBase.jl`? Do you need special fields of structs that are owned by this package here?
> Out of curiosity, why not make the overloaded functions available in `StatsBase.jl`? Do you need special fields of structs that are owned by this package here? I still wonder......
I find this strange. Can't you fix the issue in `InfiniteLinearAlgebra`? Is it used there in some generic context, or in a context where you know it's always an `AbstractQ`?...
Those (not exactly those though) indexing/slicing tests were passing back then, I was struggling with it quite a lot. The issue (on v1.10) was created in https://github.com/JuliaLinearAlgebra/InfiniteLinearAlgebra.jl/pull/148 rather recently.
Well, I guess this is ready to go then.
This is internal, but breaking. It affects a function that we encouraged data storage packages such as SparseArrays.jl, GPUArrays.jl etc. to overload. If we want to do that, then it...
@nanosoldier `runbenchmarks("linalg", vs = ":master")`
Awesome, this seems to speed up multiplication of small matrices significantly.