Array docs: Clarify DenseArray vs. "strided" #26085
Conversation
Dense Arrays absolutely must be contiguously arranged in memory. This attempts to make documentation match the current reality.
mbauman
added
docs
|
Cc: @dlfivefifty — since you wrote the strided array interface section I think it'd be nice to get your eyes on this. |
doc/src/manual/arrays.md
Outdated
| to call a wider range of BLAS and LAPACK functions by passing them either [`Array`](@ref) or | ||
| `SubArray` objects, and thus saving inefficiencies from memory allocation and copying. | ||
| A "strided" array is stored in memory and has its elements are laid out in regular offsets such that | ||
| it can be passed to external C and Fortran functions that expect this memory layout. Strided arrays |
There was a problem hiding this comment.
... such that an instance with an isbitstype eltype can be passed ...?
|
I think it should be renamed |
|
Yeah, we can tackle a renaming separately. This is just aiming to get the documentation correct. |
We can tackle more general definitions later.
doc/src/manual/arrays.md
Outdated
| `StridedVector` and `StridedMatrix` are convenient aliases defined to make it possible for Julia | ||
| to call a wider range of BLAS and LAPACK functions by passing them either [`Array`](@ref) or | ||
| `SubArray` objects, and thus saving inefficiencies from memory allocation and copying. | ||
| A "strided" array is stored in memory and has its elements are laid out in regular offsets such that |
There was a problem hiding this comment.
This should be either "and has its elements laid out" or "and its elements are laid out".
|
The reason I think it should be renamed is that it actually changes the meaning of |
| must define a [`strides(A)`](@ref) method that returns a tuple of "strides" for each dimension; a | ||
| provided [`stride(A,k)`](@ref) method accesses the `k`th element within this tuple. Increasing the | ||
| index of dimension `k` by `1` should increase the index `i` of [`getindex(A,i)`](@ref) by | ||
| [`stride(A,k)`](@ref). If a pointer conversion method [`Base.unsafe_convert(Ptr{T}, A)`](@ref) is |
There was a problem hiding this comment.
Is this true? Is getindex(A,i) a way to directly access elements based on offsets from stride calculations, or is it really still linear indexing that assumes getindex(A,i_1,i_2,...) == getindex(A, i_1 + size(A,1)*(i_2-1) + ...)?
There was a problem hiding this comment.
Ah, yes, you're right, thank you. getindex is totally and completely decoupled from strides.
|
I just came across this "clarification" PR. I agree with @dlfivefifty that it is a change in meaning rather than a clarification. The previous meaning (which was initially slated to be called The new meaning still works for all the |
|
I’m not sure which change started changing the meaning here, but it wasn’t this one. I believe the problem was that all the Base DenseArrays happened to be contiguous, and so it just gradually became the expectation. |
|
Oh, yeah, this actually started with me trying to reclaim the meaning of There are lots of things you can drag me for from 6 years ago — I have plenty of other regrets that my own brain won't let me forget. I don't think this one needs to be added to that pile. |
Dense Arrays absolutely must be contiguously arranged in memory. This attempts to make documentation match the current reality.