Re: Languages don't matter. A mathematical refutation

[jm.tarrasa@gmail.com]

El jueves, 2 de abril de 2015, 18:41:18 (UTC+2), Dmitry A. Kazakov  escribió:
> On Thu, 2 Apr 2015 07:48:45 -0700 (PDT), j_...@gmail.com wrote:
> 

> You must start with showing that "severity", "frequency" etc are random
> variables. There is a clear definition of what a random variable is. It is
> a measurable function from the set of elementary outcomes. The first step
> is presenting the set.

What's the problem with severity and frequency?

They are measurable. 

Program "A" in a two-years period
had bug1 which crashed program one time per day, and took a week to fix it
had bug2 which destroyed data, one per month, and took two days to fix it
had bug3 which was annoying, three per day, and took a month to fix it

Program "B" in a two-years period
had bug1 which crashed program one time per hour, and took a month to fix it
had bug2 which destroyed data, one a month, and took a week to fix it

Apply some formula (discrete, continuous or a complex algorithm) to each bug, sum the values of each bug for each program. Call it "bug score".

The formula just joins the data to come out with a single and comparable value to overcome summing "apples and oranges". It is a very common practice to compare several cases with many variables. Chess programs uses such formulas to score a position, and they can be very complex and they are constantly evolving. Google scores pages with a formula that involves as lot of variables and comes out with a single value.

Of course the formula is highly debatable. You can play with the formula to change the results up and down. Favour one factor reduce the other. But the data measured and is there waiting for a better formula.

Probably studying bugzilla history of several programs would be a good starting. Although I think bugzilla doesn't have "frequency", people classifies as more critical an annoying bug that happens a lot of times to many people than a destroying data one that happens once every blue moon.

Nevertheless, I insist that there are many variables in a program more important than language. So you would need to find a sample of programs equal in everything but the programming language, so you can isolate the variable language. And that wouldn't be easy.