Division algorithm not specified or consistent

[austinabell]

It is not mentioned anywhere in the spec, but there is no distinction between which division algorithm is used. By default, golang uses truncating division for integers and the big.Int library that is used does Euclidean division.

An example of where the difference can be seen:

func TestDivRounding(t *testing.T) {
	dd = big.NewInt(200)
	dv = big.NewInt(3)
	r = big.Div(dd, dv)
	assert.Equal(t, r.Int64(), int64(66))

	ddi = int64(200)
	dvi = int64(3)
	ri = ddi / dvi
	assert.Equal(t, ri, int64(66))

	dd = big.NewInt(-100)
	dv = big.NewInt(3)
	r = big.Div(dd, dv)
	assert.Equal(t, r.Int64(), int64(-34))

	ddi := int64(-100)
	dvi := int64(3)
	ri := ddi / dvi
	assert.Equal(t, ri, int64(-33))
}

There was a discrepancy when calculating the AlphaBetaFilter estimate in the reward actor with this test vector during the first update network KPI call which updates the filter estimate. The division happened at this line if it helps.

This behaviour, especially since both division algorithms are used in the protocol, should at least be written in the spec