Re: [SciPy-dev] RFR 503, 849: more robust implementation of real Bessel I_v