neither ignore_nan=True nor allow_nan=False handle Decimals correctly

[garborg]

import math
from decimal import Decimal
import numpy as np
import simplejson

fmath = [1/3, math.nan, 0.2]
fnp = np.array([1/3, np.NaN, 0.2])
dmath = [Decimal(v).quantize(Decimal('0.01')) for v in fmath]
dnp= [Decimal(v).quantize(Decimal('0.01')) for v in fnp]

print(simplejson.dumps(fmath, ignore_nan=True))
print(simplejson.dumps(fnp.tolist(), ignore_nan=True))
print(simplejson.dumps(dmath, ignore_nan=True))
print(simplejson.dumps(dnp, ignore_nan=True))

[0.3333333333333333, null, 0.2]
[0.3333333333333333, null, 0.2]
[0.33, NaN, 0.20]
[0.33, NaN, 0.20]

[etrepum]

If this issue is important to you, the best way to expedite it is to submit a PR with tests

[etrepum]

The text here is definitely a great start for writing some tests, btw

[garborg]

Do you think that'd mean digging in to the C code? If not, I'll try to circle back to it when I'm not under the gun.

A second inconsistency (change the issue title to reflect it):

simplejson.dumps(decimal.Decimal('NaN'), allow_nan=False) # NaN
simplejson.dumps(np.NaN, allow_nan=False) # exception

[etrepum]

I don't recall much C code that deals with decimal, but without writing the tests and Python implementation or auditing the C implementation I couldn't say for sure whether or not any digging there is required. A PR with failing tests (or half-failing when C extensions are enabled) is going to speed up the process if this is something you need.

[garborg]

This isn't in our workflow anymore, so while I think it'd be good to fix, it's off my radar for now.

[etrepum]

To be honest when this feature originally went in, I had no idea the Decimal even had a representation for special values since I had never seen one in the wild, so I didn't think to look into it.

Maybe the easiest thing to do would be to add something like d if d.is_normal() else float(d) in that code path, were anyone to want to look into this.