OverflowError when raise_on_invalid=True about fastnumbers

[SethMMorton]

This is a good find!

If this were written as an int (i.e. 3 * 10**106196) then Python would be able to return it as an int just fine without overflow. In this sense, string_contains_intlike_float isn't wrong to return true.

On the other hand, the algorithm being used to return an int from fast_real if parsing the input as an integer fails is to parse as a float, and then convert float output to int. In this case, float("03e106196") is inf, which creates an OverflowError when passed to int.

I think the bug is not in string_contains_intlike_float, but rather in str_to_PyInt_or_PyFloat. Consider the following:

In [2]: import fastnumbers

In [3]: fastnumbers.fast_float("03e106196")
Out[3]: inf

In [4]: fastnumbers.fast_real("03e106196")
Out[4]: '03e106196'

In [5]: fastnumbers.fast_real("03e106196", coerce=False)
Out[5]: inf

In [6]: fastnumbers.fast_real("03e106196", raise_on_invalid=True, coerce=False)
Out[6]: inf

In [7]: fastnumbers.fast_real("03e106196", raise_on_invalid=True)
---------------------------------------------------------------------------
OverflowError                             Traceback (most recent call last)
<ipython-input-7-64eb4b6453ff> in <module>()
----> 1 fastnumbers.fast_real("03e106196", raise_on_invalid=True)

OverflowError: cannot convert float infinity to integer

In [8]: fastnumbers.__version__
Out[8]: '2.0.4'

If fast_float can return a number, fast_real should always be able to return a number, but this is not the case with coerce=True (the default). This is because the string looks like it should be an int but is too big to be stored as a float.

The correct thing to do would be to detect if inf was returned, and in that case do some special handling on the string to ensure an int is returned. I'll give this some thought.

[SethMMorton]

In your opinion, what should be the correct behavior?

>>> fastnumbers.fast_real("03e106196") == inf  # coerce=True

or

>>> fastnumbers.fast_real("03e106196") == 3 * 10**106196  # coerce=True

I would expect the same result whether raise_on_invalid is True or False.

[davidsmf]

oh, good question. Python says "inf" for the literal 3e106196, so I'd go with that.