type instability fix

[daviehh]

Tentative PR for the type instability/runtime dispatch and the resulting long loading time, may fix #128
Maybe it may be handled in a more clean/elegant fashion..

Also, since eltype(<:Number) just returns the type of the number, the line

u_elType = uType <: Number ? uType : eltype(u)

maybe just changed to u_elType = eltype(u). Not sure how robust it can be

[Deltadahl]

Nice one!
Just before I had to leave yesterday I also found the problem and solved it with

max_trust_radius = convert(typeof(max_trust_radius), max(norm(fu), maximum(u) - minimum(u)))

in #131. Sorry for not telling you, but I thought this comment #131 (comment) would trigger a note in #128.
Anyhow, maybe your way to solve it is better, nice work :)

[ChrisRackauckas]

                     max_trust_radius::Real = 0.0,
                     initial_trust_radius::Real = 0.0,
                     step_threshold::Real = 0.1,
                     shrink_threshold::Real = 0.25,
                     expand_threshold::Real = 0.75,
                     shrink_factor::Real = 0.25,
                     expand_factor::Real = 2.0,

what we really should be doing is change those values to Rational numbers. If that is done, they will convert on-demand when hitting u types, for example 1//2 * 0.33 == 0.165 and 1//2 * 3.3f-2 == 0.0165f0. That makes the default values "type neutral". I wouldn't require that the numbers are rational, I would expect users to give Float64 value (or correctly type-matching values), but since the overwhelming default will be for the user to just use the default values of the algorithm parameters, this would be a safer option. The correct conversions should also then be done when the values are put into the cache object as well, doing a convert(eltype(u),x) on all of them.

[codecov]

Codecov Report

Merging #132 (b902aa3) into master (e1c35f5) will increase coverage by 0.28%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #132      +/-   ##
==========================================
+ Coverage   91.00%   91.28%   +0.28%     
==========================================
  Files           7        7              
  Lines         500      505       +5     
==========================================
+ Hits          455      461       +6     
+ Misses         45       44       -1     

Impacted Files	Coverage Δ
src/trustRegion.jl	98.46% <100.00%> (+0.06%)	⬆️
src/utils.jl	66.15% <0.00%> (+1.53%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[Deltadahl]

hmm, I made the changes suggested by Chris, but I wasn't allowed to push to @daviehh
(Think its because: "The author of the PR also may need to explicitly allow you to push to their branch." in https://gist.github.com/wtbarnes/56b942641d314522094d312bbaf33a81)

[ChrisRackauckas]

You can PR to his branch.

[ChrisRackauckas]

Hmm something happened with the merge ambiguity fix. It's probably just a simple thing with the type parameters.

[Deltadahl]

The problem is that step_threshold and step_size got different types (trustType, suType and uType) see the "arrows" (<-----------------)

    step_threshold::trustType <-----------------
    shrink_threshold::trustType
    expand_threshold::trustType
    shrink_factor::trustType
    expand_factor::trustType
    loss::floatType
    loss_new::floatType
    H::jType
    g::resType
    shrink_counter::Int
    step_size::suType  <-----------------
    u_tmp::tmpType
    fu_new::resType
    make_new_J::Bool
    r::floatType

    function TrustRegionCache{iip}(f::fType, alg::algType, u::uType, fu::resType, p::pType,
                                   uf::ufType, linsolve::L, J::jType,
                                   jac_config::JC, iter::Int,
                                   force_stop::Bool, maxiters::Int, internalnorm::INType,
                                   retcode::SciMLBase.ReturnCode.T, abstol::tolType,
                                   prob::probType, trust_r::trustType,
                                   max_trust_r::trustType, step_threshold::suType, <-----------------
                                   shrink_threshold::trustType, expand_threshold::trustType,
                                   shrink_factor::trustType, expand_factor::trustType,
                                   loss::floatType, loss_new::floatType, H::jType,
                                   g::resType, shrink_counter::Int, step_size::uType, <-----------------
                                   u_tmp::tmpType, fu_new::resType, make_new_J::Bool,

Unfortunately I got no time to fix it right now, but if its not fixed in ~8h I'll have time to do it