nan from scipy.special.struve

When looking for the origin of a NaN in my program, I ended up discovering that scipy.special.struve can output NaN if the expected value is close to zero. Is there anything that can be done to avoid this ?

Python 3.4.2 (default, Oct  8 2014, 10:45:20) 
[GCC 4.9.1] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import scipy.special
>>> scipy.special.struve(0,25.76534)
-3.9070380020700526e-06
>>> scipy.special.struve(0,25.76538)
2.2841077550668455e-06
>>> scipy.special.struve(0,25.76536)
nan

[Perhaps related to https://github.com/scipy/scipy/issues/4446]

[pv]

gh-4446 is unrelated, the implementations are separate.

The nan comes around the zero of the function — it is due to the code noticing it cannot achieve the desired relative precision of the function value, nor the desired absolute precision which is fairly strict.

So the issue likely is just in the precision check and not a problem in the evaluation algorithm. In principle, the precision check should also use information based on the derivative of the function, but it does not.

[pv]

Ok. Maybe the following suggestion would be OK: add at the end of struve_hl, to be done if none of the standard accuracy estimates is accepted:

envelope_value = some non-oscillating order of magnitude estimate, eg. 
from DLMF 11.6.5, maybe combined with something else for small v
;
if (fabs(value) < ZERO_EPS * envelope_value && fabs(err) < 
ACCEPTABLE_EPS * envelope_value) {
    return value;
}

where value and err correspond to the best of the estimates found by the three methods.