Update an NLPModelMeta

[MaxenceGollier]

Hello, I am trying to work on Quasi-Newton approximations of NLPModels.
Given some nlp::NLPModel, I would like to construct a SpectralGradientModel(nlp) i.e an NLPModel where I approximate the hessian $\mathcal{H}_f \approx \sigma I$. For this, I want to update the NLPModelMeta so that nlp.meta.nnzh = nlp.meta.nvar, so I want to create a copy of nlp.meta but with just one field modified. Since NLPModelMeta is an immutable struct, I can't just set nlp.meta.nnzh = nlp.meta.nvar. Since there are quite a lot of keyword arguments and since only one field is modified, I don't want to write each time

new_meta = NLPModelMeta( nnzh = nlp.meta.nvar, field1 = nlp.meta.field1, field2 = nlp.meta.field2, field3 = nlp.meta.field3, ...)

Also, I tried using Setfields.jl but this does not seem to work.

I am quite new to Julia so I was wondering if you had an idea on how to do this.

[tmigot]

Unfortunately, I don't think there is a very clean solution here. We could add a new constructor for NLPModelMeta that simplifies this task though. Something like

NLPModelMeta(meta::AbstractNLPModelMeta; kwargs...) = NLPModelMeta(meta.nvar; x0 = meta.x0, etc...., name = meta.name, kwargs...)

[MaxenceGollier]

That's what I was thinking but I wondered, if meta is AbstractNLPModelMeta but not NLPModelMeta for some reason, then your solution could be wrong ? I don't know if something else than NLPModelMeta has ever been implemented ?

[tmigot]

What do you mean by it would be "wrong"?
There could be potential errors if the meta is missing some fields that are in the given keyword arguments, but anyway this would break several functions in the API.
In JSO, I am not aware of any other specialization of AbstractNLPModelMeta.