naturalsize output changed between 4.11.0 and 4.12.0

[davidboweninrupt]

Possibly a result of #218, we are seeing that the output has changed:

1. podman run -it python bash
2. python -m pip install humanize==4.11.0
3. python -c "import humanize; print(humanize.naturalsize(1.123456789))"

1 Bytes

4. python -m pip install humanize==4.12.0
5. python -c "import humanize; print(humanize.naturalsize(1.123456789))"

1.123456789 Bytes

Expected behaviour is that output would not change in a minor version or that it would be documented in the release notes.

[hugovk]

Thanks for the report.

This was caused by some bad formatting replacement in fb61d5a / #226.

Let's fix it properly and make a patch release.

---

As an aside, we get:

>>> import humanize
>>> humanize.naturalsize(1.123456789 * 1000)
'1.1 kB'
>>> humanize.naturalsize(1.123456789 * 1000000)
'1.1 MB'

So I think it would be better to also get:

>>> humanize.naturalsize(1.123456789 * 1000000)
'1.1 Bytes'

Instead of:

>>> humanize.naturalsize(1.123456789 * 1000000)
'1 Bytes'

But that's another issue. And we're definitely missing tests for non-integer input:

humanize/tests/test_filesize.py

Lines 14 to 75 in a0602c7

	([300], "300 Bytes"),
	(["1000"], "1.0 kB"),
	([10**3], "1.0 kB"),
	([10**6], "1.0 MB"),
	([10**9], "1.0 GB"),
	([10**12], "1.0 TB"),
	([10**15], "1.0 PB"),
	([10**18], "1.0 EB"),
	([10**21], "1.0 ZB"),
	([10**24], "1.0 YB"),
	([10**27], "1.0 RB"),
	([10**30], "1.0 QB"),
	([1000**1 * 31], "31.0 kB"),
	([1000**2 * 32], "32.0 MB"),
	([1000**3 * 33], "33.0 GB"),
	([1000**4 * 34], "34.0 TB"),
	([1000**5 * 35], "35.0 PB"),
	([1000**6 * 36], "36.0 EB"),
	([1000**7 * 37], "37.0 ZB"),
	([1000**8 * 38], "38.0 YB"),
	([1000**9 * 39], "39.0 RB"),
	([1000**10 * 40], "40.0 QB"),
	([300, True], "300 Bytes"),
	([1024**1 * 31, True], "31.0 KiB"),
	([1024**2 * 32, True], "32.0 MiB"),
	([1024**3 * 33, True], "33.0 GiB"),
	([1024**4 * 34, True], "34.0 TiB"),
	([1024**5 * 35, True], "35.0 PiB"),
	([1024**6 * 36, True], "36.0 EiB"),
	([1024**7 * 37, True], "37.0 ZiB"),
	([1024**8 * 38, True], "38.0 YiB"),
	([1024**9 * 39, True], "39.0 RiB"),
	([1024**10 * 40, True], "40.0 QiB"),
	([1000**1 * 31, True], "30.3 KiB"),
	([1000**2 * 32, True], "30.5 MiB"),
	([1000**3 * 33, True], "30.7 GiB"),
	([1000**4 * 34, True], "30.9 TiB"),
	([1000**5 * 35, True], "31.1 PiB"),
	([1000**6 * 36, True], "31.2 EiB"),
	([1000**7 * 37, True], "31.3 ZiB"),
	([1000**8 * 38, True], "31.4 YiB"),
	([1000**9 * 39, True], "31.5 RiB"),
	([1000**10 * 40, True], "31.6 QiB"),
	([1000**1 * 31, False, True], "30.3K"),
	([1000**2 * 32, False, True], "30.5M"),
	([1000**3 * 33, False, True], "30.7G"),
	([1000**4 * 34, False, True], "30.9T"),
	([1000**5 * 35, False, True], "31.1P"),
	([1000**6 * 36, False, True], "31.2E"),
	([1000**7 * 37, False, True], "31.3Z"),
	([1000**8 * 38, False, True], "31.4Y"),
	([1000**9 * 39, False, True], "31.5R"),
	([1000**10 * 40, False, True], "31.6Q"),
	([300, False, True], "300B"),
	([3000, False, True], "2.9K"),
	([3000000, False, True], "2.9M"),
	([1024, False, True], "1.0K"),
	([1, False, False], "1 Byte"),
	([3141592, False, False, "%.2f"], "3.14 MB"),
	([3000, False, True, "%.3f"], "2.930K"),
	([3000000000, False, True, "%.0f"], "3G"),
	([10**26 * 30, True, False, "%.3f"], "2.423 RiB"),

[hugovk]

#240 adds tests and should revert to the previous behaviour. How does it look?