Add a keyword argument iostream for all Krylov methods

[amontoison]

https://juliasmoothoptimizers.github.io/Krylov.jl/previews/PR677/

We will be able to store the outputs of verbose mode with this modification.

[github-actions]

Package name	latest	stable
CaNNOLeS.jl
DCISolver.jl
FletcherPenaltySolver.jl
JSOSolvers.jl
LLSModels.jl
Percival.jl
RipQP.jl

[dpo]

Good idea! Next best thing to a logger!

[codecov]

Codecov Report

Base: 98.19% // Head: 98.24% // Increases project coverage by +0.04% 🎉

Coverage data is based on head (e8a45cb) compared to base (630faac).
Patch coverage: 91.22% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #677      +/-   ##
==========================================
+ Coverage   98.19%   98.24%   +0.04%     
==========================================
  Files          37       37              
  Lines        6601     6599       -2     
==========================================
+ Hits         6482     6483       +1     
+ Misses        119      116       -3     

Impacted Files	Coverage Δ
src/krylov_utils.jl	92.55% <ø> (ø)
src/cr.jl	66.35% <31.03%> (ø)
src/bicgstab.jl	100.00% <100.00%> (ø)
src/bilq.jl	100.00% <100.00%> (ø)
src/bilqr.jl	100.00% <100.00%> (ø)
src/cg.jl	100.00% <100.00%> (ø)
src/cg_lanczos.jl	100.00% <100.00%> (ø)
src/cg_lanczos_shift.jl	100.00% <100.00%> (+2.54%)	⬆️
src/cgls.jl	100.00% <100.00%> (ø)
src/cgne.jl	100.00% <100.00%> (ø)
... and 24 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[amontoison]

I don't understand why I have allocations when I use a sparse matrix and not when I use a dense one. 🤔

using Printf

function krylov(A, b::AbstractVector{T}; verbose::Int=0, iostream::IO=stdout) where {T}
  verbose > 0 && @printf(iostream, "%3d\n", 42)
  return nothing
end

A = rand(10, 10)
b = rand(10)
krylov(A, b)  # warm up
@allocated krylov(A, b)                   # 0 byte
krylov(A, b, iostream=stdout)  # warm up
@allocated krylov(A, b, iostream=stdout)  # 32 bytes

A2 = sprand(10, 10, 0.5)
b2 = rand(10)
krylov(A2, b2)  # warm up
@allocated krylov(A2, b2)                   # 48 bytes
krylov(A2, b2, iostream=stdout)  # warm up
@allocated krylov(A2, b2, iostream=stdout)  # 32 bytes

const kstdout = stdout

function krylov2(A, b::AbstractVector{T}; verbose::Int=0, iostream::IO=kstdout) where {T}
  verbose > 0 && @printf(iostream, "%3d\n", 42)
  return nothing
end

A = rand(10, 10)
b = rand(10)
krylov2(A, b)  # warm up
@allocated krylov2(A, b)                    # 0 byte
krylov2(A, b, iostream=kstdout)  # warm up
@allocated krylov2(A, b, iostream=kstdout)  # 16 bytes

A2 = sprand(10, 10, 0.5)
b2 = rand(10)
krylov2(A2, b2)  # warm up
@allocated krylov2(A2, b2)                    # 0 byte
krylov2(A2, b2, iostream=kstdout)  # warm up
@allocated krylov2(A2, b2, iostream=kstdout)  # 16 bytes

[dpo]

What does the @code_warntype say? If nothing useful, you may have to look at the low-level translation.

[amontoison]

What does the @code_warntype say? If nothing useful, you may have to look at the low-level translation.

I checked with @code_warntype and it says nothing special.
We have some small differences with @code_llvmbut I don't understand them.

I will open an issue in the GitHub repository of Julia.

[amontoison]

@dpo
I found the problem, stdout from Base is a global variable and its type is not fixed.
It means that the compiler can only determine the type at runtime.
The type of stdout is different if we use redirect_stdout(mystream) for instance.

I don't have allocations with Core.stdout because its type can't be modified.
I propose to use this one by default and use kstdout as alias.

[dpo]

great that it was not a bug!