Add backwards-compat for `jac_lin*` functions defined with `x`

[klamike]

@amontoison

Since #501, the jac_lin* functions no longer take an x argument. This breaks JSO packages (such as QuadraticModels.jl) that define/use the functions still including the x argument.

[amontoison]

@klamike
The culprit PR is this one: https://github.com/JuliaSmoothOptimizers/NLPModels.jl/pull/501/files
I asked for backward compatibility but it seems that @arnavk23 was not careful with that.

@arnavk23 Can you fix it? It is a big issue for the ecosystem.

[arnavk23]

@amontoison I'll see to it immediately.

[klamike]

I ended up patching enough of ADNLPModels, NLPModelsTest, NLPModelsModifiers, and QuadraticModels to get their tests passing with the latest NLPModels (see the mk/update_jaclin branch on my forks of those, and for NLPModelsTest, also merge in JuliaSmoothOptimizers/NLPModelsTest.jl#129).

@arnavk23 I'll wait for your fix here before PRing those, likely a lot of those changes can be eliminated once that is done, though there are some more changes there that still wouldn't be fixed by just back-compat for both signatures, for example #500 broke the show tests in NLPModelsModifiers, and SparseConnectivityTracer compat needed a bump from 0.6 to 1.0.

[tmigot]

Was this really a big problem ? It just means that the next release is a breaking one, but we didn't make any release since this change.

[tmigot]

We deprecated the function that changed in this commit , 3091fb0, is there an error there ?

@deprecate jac_lin(nlp::AbstractNLPModel, x::AbstractVector) jac_lin(nlp)

[amontoison]

Yes the deprecation doesn't allow to use the old API.

[amontoison]

@tmigot We want to call jac_lin_coord! and not jac_lin.

[tmigot]

Let's try to fix the deprecate in @arnavk23 's PR, and then we will make a new release so that @klamike can make PRs to adapt.

[tmigot]

@klamike NLPModels 0.22 is out. Can you open pull requests with your fixes?

[klamike]

Should I just open them all? the NLPModelsTest and NLPModelsModifiers depend on each other..

[arnavk23]

Should I just open them all? the NLPModelsTest and NLPModelsModifiers depend on each other..

I think you should. It will also help Tangi to better understand the scale of the change you are planning.

[klamike]

Okay, here we go:

JuliaSmoothOptimizers/NLPModelsTest.jl#130 (needs new NLPModelsModifiers)
JuliaSmoothOptimizers/NLPModelsModifiers.jl#133 (needs new NLPModelsTest)
JuliaSmoothOptimizers/ADNLPModels.jl#363 (needs new NLPModelsTest, NLPModelsModifiers)
JuliaSmoothOptimizers/QuadraticModels.jl#162 (needs all of the above)

[klamike]

I make no claims about completeness... this is what I needed to change for QuadraticModels tests to pass. It is possible there are more things to change that the tests just didn't hit (not to mention JSO packages other than these)

I still think it would be best to figure out how to do backwards compat, though I don't know how to do that in this case. 0.22 will likely break a lot of user codes, though it is properly marked as breaking...