When supplying numpy arrays for ghi and dhi that have zeros (e.g. for nighttime), a warning is emitted:

import pvlib
import pandas as pd

times = pd.date_range("2019-06-01", freq="1min", periods=1440, tz="Etc/GMT+5")
location = pvlib.location.Location(40, -80)

sp = location.get_solarposition(times)
cs = location.get_clearsky(times)

ret = pvlib.bifacial.infinite_sheds.get_irradiance(
    surface_tilt=20,
    surface_azimuth=180,
    solar_zenith=sp['zenith'],
    solar_azimuth=sp['azimuth'],
    gcr=0.5,
    height=2.0,
    pitch=0.5,
    ghi=cs['ghi'].values,  # arrays
    dhi=cs['dhi'].values,
    dni=cs['dni'].values,
    albedo=0.2,
)

\lib\site-packages\pvlib\bifacial\infinite_sheds.py:345: RuntimeWarning: invalid value encountered in divide
  diffuse_fraction = np.clip(dhi / ghi, 0., 1.)

We could simply suppress the warning, but I think a better solution is to do away with the calculation entirely. This diffuse fraction is used only in the _poa_ground_shadows function, which computes the average irradiance reflected from the ground, accounting for the shadows cast by rows and the restricted ground->sky view factor:

where df is the diffuse fraction and poa_ground is ghi * albedo. The math is equivalent to the following:

albedo * (f_gnd_beam * (ghi - dhi) + vf_gnd_sky * dhi)

This alternative means we don't need to calculate the diffuse fraction at all, and (to me) is an improvement in clarity as well. I propose dropping diffuse_fraction and using the above calculation instead. Note that the line in question is in a private function, so it is no issue to change the relevant function signature.