invalid int equals itself in comparison, while np.nan doesn't

[zhangyingmath]

np.nan == np.nan False

a = np.array([1,2,3,np.nan]) ds = rt.Dataset({'f': [0,0,0, 0], 'g': a})

a == a array([ True, True, True, False])

ds['g'] == ds['g'] FastArray([ True, True, True, False])

ds[0, 'f'] = np.nan ds['f'] == ds['f']

FastArray([ True, True, True, True])

[zhangyingmath]

Here is another limitation in current implementation of inv integers: if I write my own function taking a FastArray as input, it is easy to make a mistake:

ds = rt.Dataset({'f': [1,2,3,4]})
ds[0, 'f'] = np.nan
print(ds)
#     f
-   ---
0   Inv
1     2
2     3
3     4

print(ds['f'])
[-9223372036854775808                    2                    3
                    4]
tmp = 0
for i in range(4):
    tmp += ds[i, 'f']
    
print(tmp)
-9223372036854775799