`taylorinteg` not working properly with BigFloats about taylorintegration.jl HOT 7 CLOSED

Can you isolate the problem without TaylorIntegration? I mean, create such a vector, fill it partially, and transpose it. Perhaps this should be posted directly to Julia.

from taylorintegration.jl.

I think the problem can be solved replacing the lines which look like:

    return view(tv,1:nsteps), view(transpose(xv),1:nsteps,:)

by

    return view(tv,1:nsteps), transpose(view(xv,:,1:nsteps))

Maybe it impacts in performance, but my tests make it difficult to decide...

from taylorintegration.jl.

Can you isolate the problem without TaylorIntegration? I mean, create such a vector, fill it partially, and transpose it. Perhaps this should be posted directly to Julia.

Following your recommendation, I think an example of this would be:

julia> A = Array{Float64}(2,3)
2×3 Array{Float64,2}:
 2.34463e-314  2.34463e-314  2.34463e-314
 2.34463e-314  2.34463e-314  2.34463e-314

julia> A[1,:] .= rand(3)
3-element SubArray{Float64,1,Array{Float64,2},Tuple{Int64,Base.Slice{Base.OneTo{Int64}}},true}:
 0.171995
 0.35663 
 0.263829

julia> transpose(A)
3×2 Array{Float64,2}:
 0.171995  2.34463e-314
 0.35663   2.34463e-314
 0.263829  2.34463e-314

julia> B = Array{BigFloat}(2,3)
2×3 Array{BigFloat,2}:
 #undef  #undef  #undef
 #undef  #undef  #undef

julia> B[1,:] .= BigFloat.(rand(3))
3-element SubArray{BigFloat,1,Array{BigFloat,2},Tuple{Int64,Base.Slice{Base.OneTo{Int64}}},true}:
 5.221675004832033728519036230864003300666809082031250000000000000000000000000000e-01
 9.018393963488640086012537722126580774784088134765625000000000000000000000000000e-01
 3.842859182000140094714879523962736129760742187500000000000000000000000000000000e-01

julia> transpose(B)
ERROR: UndefRefError: access to undefined reference
Stacktrace:
 [1] transpose_f!(::Base.#transpose, ::Array{BigFloat,2}, ::Array{BigFloat,2}) at ./linalg/transpose.jl:54
 [2] transpose(::Array{BigFloat,2}) at ./linalg/transpose.jl:121

Do you think it would be worthwhile to report this issue to the corresponding Julia org?

from taylorintegration.jl.

I agree that #29 could be a workaround!

from taylorintegration.jl.

Maybe it impacts in performance, but my tests make it difficult to decide...

Do you want me to check some performance benchmarks?

from taylorintegration.jl.

Yes, I think it is a good idea to perform some benchmarks.

Regarding the issue with the transpose, I do think this is an issue for Julia directly. Yet, as you can see in your example, when you allocate the matrix of Float64, it is filled with whatever Float64s, which is not the case for BigFloats. That is the reason that the error does not occur with the Float64 matrix.

from taylorintegration.jl.

Closed by #33

from taylorintegration.jl.