Nick

Could this be done via `set_target_properties(boost_math PROPERTIES CXX_VISIBILITY_PRESET hidden)`? You'll have to excuse my naivety if this is totally underestimates the effort; I could never figure out what symbol visibility...

@wthrowe : Does the Halley iterate succeed? In general I find the Halley iterate has a more sane basin of convergence and moreover it can degrade gracefully to a newton...

> I guess the question is to what lengths we're prepared to go to protect against bad initial guesses? Yeah, I think the worry is that if you add too...

Not an unreasonable request. Looks like all the building blocks are already in Boost.Math. Got a number of other things to do at the moment, but I'll try to get...

@pcjmuenster : If you want to submit a PR I could help with the details. I'd probably just `cp normal.hpp von_mises.hpp` and just make the necessary adjustments to make it...

Looking at the [Von Mises distribution wiki](https://en.wikipedia.org/wiki/Von_Mises_distribution), the mean, variance, mode, pdf and entropy are straightforward. The only one that would require significant testing is the CDF. Also a decision...

@pcjmuenster : Could direct numerical quadrature work in the case where the series converges slowly? We have quite a few quadrature routines available, though some require expensive node-weight precomputation. Note...

@pcjmuenster : My opinion, such as it is, is that as long as it is documented, it's fine to have the support be `[mu-pi, mu+pi]`. @jzmaddock, @pabristow : What's your...

Yeah, for periodic functions, `support()` is a bit ambiguous. Maybe it should be the whole real line?

@pcjmuenster : I think this will require writing very careful error messages that self-document the requirements of the function. I might start with the definition `support() := [µ-π, µ+π)`. Then...