Deprecate `hyp2f0`, `hyp1f2`, and `hyp3f0`?

[person142]

The functions special.hyp2f0, special.hyp1f2, and special.hyp3f0 come from Cephes. As far as I can tell they were originally intended to be internal helper routines for computing the confluent hypergeometric function of the second kind and Struve functions. As such, their implementations are incomplete; for example hyp2f0 only contains an asymptotic expansion. It looks like they were included in special essentially by accident.

These functions come up occasionally (wiki page), but probably not enough that it's actually important to have them in special. I doubt anyone will ever feel inspired to complete their implementations. Seeing as there are already enough problem functions in special, I think we should just cut them loose. Thoughts?

Note: I am not proposing that we deprecate the important and frequently used hyp0f1, hyperu, hyp1f1, and hyp2f1.

[rgommers]

Sounds reasonable to me. Propose it on the mailing list as well please, that's where we should put all deprecations up for discussion.

[pv]

Ok with me. We should ideally not ship routines that are not useful. At minimum known deficiencies would need to be stated in the documentation, but if it's only asymptotic expansion in some regimes, then it may as well be best to remove them.

[rainwoodman]

My collaborator just figured the result of hyp1f2 and diverges. We do want to use hyp1f2 -- any recommendation on where to find a useable implementation?

[vhaasteren]

And then scientists come here to find out why the function they rely on has been removed from scipy, only to find that there is not really an alternative.

I understand the reasoning stated above, but without another way to calculate this function I feel like this was a wrong decision

[vhaasteren]

If anyone is ever interested, I packaged two of the deprecated functions that I needed here:

https://github.com/vhaasteren/scipydepr