set_with_dataframe: TypeError: Object of type 'int64' is not JSON serializable about gspread-dataframe

[robin900]

@sam-s I discovered why this bug has been elusive in the past. When a dataframe has multiple columns and one of those columns is of dtype float32, float64, string, or object, calling dataframe.values causes pandas to find a common dtype for a numpy array to hold values from all the columns. And since float32, float64, string, and object dtypes all result in values that are JSON serializable, there's no bug. Only for a dataframe whose columns are all non-JSON-seralizable dtypes (like int64) does the bug surface.

The solution in PR #36 decides on 'object' as the common dtype in all cases, and uses the recommended dataframe.to_numpy method instead of the deprecated dataframe.values.

[robin900]

@sam-s: Reading through all the pandas documentation, I'm torn between two possible approaches to solve this problem:

1. Use a custom JSON serializer that supports all the pandas/numpy data types.
2. Make pandas do the work of converting the dtypes to Python basic data types.

I'm leaning toward option 2, and I'm wondering if you think it's a good idea to use dataframe.to_numpy('object') and dataframe.index.to_numpy('object') to let pandas/numpy do the necessary type conversions. I'm getting correct results with this technique, you can see it on pull request #36 for branch int64-json-35.