Cost zero division

[accurrently]

Fix for bug #19.

Updated cost() to include a kwarg called zvalue which defaults to 0.5. Changes should have little effect on existing calls.

It effectively prevents zero values for nbytes from creating a ZeroDivisionError by providing a nonzero value less than 1. The function still raises ZeroDivisionError if zvalue==0.

New return value:

time / (nbytes if nbytes else zvalue) / 1e9

[jrbourbeau]

Thanks for the PR @accurrently, I appreciate you fixing this! I wonder if instead of adding an additional zvalue keyword, could we just use 1 if nbytes is 0. I.e. something along the lines of your original comment (#19 (comment)):

def cost(nbytes, time):
    return float(time) / max(nbytes, 1) / 1e9

What do you think about this approach?

[accurrently]

We can. Though bytes should never be negative, part of used the if/else statement to protect against a zero value instead of having to worry about negative values.

The reason I chose .5 is to differentiate between an object of 1 byte and a value that clearly indicates a zero/null. If the difference in outcome isn't significant, 1 should be fine.

If nbytes can be guaranteed to be >=0, then using max() should work just fine, in my opinion.

And frankly, max() is much nicer to read than the if/else. If the value of one works, that should be fine as well.

Agree on elimination of the kwarg.

[mrocklin]

I think that we don't need this import.

[mrocklin]

Let's not assert that they have to be these values exactly. I think that it would be fine if things just worked, and that the cost was relatively low, like maybe

Suggested change

	assert cost(0, 500) == 5e-07
	assert 0 <= cost(0, 500) <= 100

[mrocklin]

Suggested change

	def test_cost():
	def test_cost_with_zero_bytes():

[martindurant]

@accurrently , were you planning to finish this off?

[martindurant]

Thanks for the original work here, @accurrently