Rounding error on 32-bit makes Order.size float instead of integer

[IMYin]

Discussed in #953

^{Originally posted by IMYin March 31, 2023}
My trading target is futures, with a minimum trading size of 1 lot and a minimum unit of 300 per lot. When I trigger a certain signal, I only want to reduce my position instead of closing it completely. However, the parameter in the close function is a proportion value. So, if I hold a position of 7 lots and my size is 2100, the parameter set for self.position.close(portion=6/7), which is not wrong. But the result is that my position is 301, which is incorrect and causes a chain reaction of position errors later on.

The sample shown in the picture is incorrect.

The correct one should look like the following.

So I am wondering if it is possible to adjust the setting of the portion parameter?

[kernc]

backtesting.py/backtesting/backtesting.py

Lines 555 to 560 in 0ce24d8

	def close(self, portion: float = 1.):
	"""Place new `Order` to close `portion` of the trade at next market price."""
	assert 0 < portion <= 1, "portion must be a fraction between 0 and 1"
	size = copysign(max(1, round(abs(self.__size) * portion)), -self.__size)
	order = Order(self.__broker, size, parent_trade=self, tag=self.__tag)
	self.__broker.orders.insert(0, order)

>>> (1 - 6/7) * 2100
300.0000000000001

>>> from math import copysign
>>> copysign(max(1, round(abs(2100) * (6/7))), -2100)
-1800.0

Looks like a rounding error on your platform. 32-bit OS by chance?