speclite.filters.get_ab_maggies alters input wavelength array? about speclite HOT 4 CLOSED

These lines will modify the input wave array under certain conditions:

• validate_wavelength_array() does not need to make a copy, e.g., to strip units or convert from python to numpy array.
• the filter response is zero for some of the wavelengths in the input array.

However, I am also unable to reproduce this with a MWE, so there's something I still don't understand. For moving forwards with simqso, I suggest that you use the copy() for now, until I can spend more time getting to the bottom of this.

from speclite.

Hi,

This code will consistently corrupt the wavelength array

import speclite.filters
import numpy as np
filters = speclite.filters.load_filters('gaiadr2-BP', 'gaiadr2-RP',
                                        'gaiadr2-G')
waves = np.arange(3000, 11000, 0.11)
waves0 = waves.copy()
flux = waves * 1.
filters[0].get_ab_magnitude(flux, waves)
print(np.nonzero(waves != waves0), waves[2249], waves0[2249])

Output:
(array([ 2249, 40932]),) 3247.4999999999995 3247.3900000002864

IMO validate_wavelength_array() must make a copy even at a performance cost, because
the current behaviour is dangerous as it leads to silent corruption (i've encountered this bug while working on stdstars DESI script)

S

from speclite.

Thanks for this @segasai. I will take care of this but presumably you can copy your wavelength array in the stdstars script so this is not blocking you?

from speclite.

Yes, that's what I did for the time being.
I just was suggesting that given the danger of the bug, it may be worth putting a stopgap copy() inside validate_wavelength_array() now before coming with a proper fix, but that's just IMO.

from speclite.