Fix Optim version for NLsolve

[pkofod]

ref JuliaNLSolvers/Optim.jl#217

It only uses linesearch, so it's not affected. I've run the tests in the package as an extra check.

The KernelEstimators will run fine as well.

[tkelman]

Awesome, thanks for checking these things out. Hopefully having a look at how Optim gets used by other packages is a bit useful as you work on making major improvements to it?

PkgEval picked up a possible issue with Mads.jl: https://gist.github.com/4609c24a5c7b8bb8222012c9564958c6

[pkofod]

Awesome, thanks for checking these things out. Hopefully having a look at how Optim gets used by other packages is a bit useful as you work on making major improvements to it?

Yes, quite helpful actually :)

PkgEval picked up a possible issue with Mads.jl: https://gist.github.com/4609c24a5c7b8bb8222012c9564958c6

Seems like it could be a bug in Optim. Thanks for letting me know.

[pkofod]

PkgEval picked up a possible issue with Mads.jl: https://gist.github.com/4609c24a5c7b8bb8222012c9564958c6

Seems like it could be a bug in Optim. Thanks for letting me know.

Okay, I hadn't noticed, but Mads is calling unexported helper functions from Optim. I'll make a PR to fix their usage. Should I reinstate the limit?

[tkelman]

Should I reinstate the limit?

Yes, I think so, since they had been passing on June 6th - http://pkg.julialang.org/detail/Mads.html

[pkofod]

I will, and then they can change it once madsjulia/Mads.jl#2 is merged, and they tag a new version.

[tkelman]

It should just do the right thing automatically when they make a new tag, since new tags populate the metadata require file from the current package version, not the previous tag's metadata version.