I run some simulations whose result is a dict like this:

results = {
    'this': 5,
    'that': 6,
    'those': [2.34, 5.67]
}

I save it in a human-readable format using code like this:

s = ''
for key, value in results.items():
    s += f'{key}: {value}n'

One of my values happens to be a list of numpy.float32 values.

All was well until I upgraded to numpy 2.0.0. In this version, the code above produces strings like [np.float32(2.34), np.float32(5.67)] instead of [2.34, 5.67] (which I got in numpy 1.26.1).

Should I complicate my saving code and use different code for saving values of different types? Or is there some workaround which could produce human-readable output for all types, including those composed of float32?

Additional examples (in an interactive python environment):

import numpy
x = numpy.float32(2.3)

x
    numpy 1 — 2.3
    numpy 2 — np.float32(2.3)

f'{x}'
    numpy 1 — '2.299999952316284'
    numpy 2 — '2.299999952316284'

f'{[x, x]}'
    numpy 1 — '[2.3, 2.3]'
    numpy 2 — '[np.float32(2.3), np.float32(2.3)]'

y={x:x*x}
y
    numpy 1 — {2.3: 5.29}
    numpy 2 — {np.float32(2.3): np.float32(5.29)}

I can’t make sense of it. Is it a bug that I could expect to be fixed, or a feature which is unlikely to change?