Why compute value for nmax+1 in bessel_Jn_array()?

[Ruslan Kabatsayev]

Hello,

I'm looking into implementation of gsl_sf_bessel_Jn_array(). There
stat_np1 is computed for nmax+1 and stat_n for nmax, so that later the
backward recurrence relation is used to obtain values for nmin through
nmax-1. It seems strange to me though that nmax+1 is used at all —
can't the function just go down from nmax and nmax-1 instead of nmax+1
and nmax? It seems currently it does one extra computation only to
effectively throw away its result. This also becomes a problem e.g.
when x==0.7 and nmax==141: the underflow error makes it fail, although
if the function started from nmax instead of nmax+1, it'd succeed as
it currently does for nmax==140.

Could someone clarify this to me?

Regards,
Ruslan