[roedog88]

What about certifying the software that Joe the developer creates, rather than Joe the developer? This is the approach the FAA takes with DO178B for the flight software on the jet planes that we ride.

These are very large projects, using hundreds of developers. The quality of the team members varies. Half are below average for the population at large.

Adding certification on top of this doesn't change the fact that half the developers will be below average. The projects are too large.

[hernan7]

Do you mean half are below the median?

(Actually, if the rockstar programmer stories are true, then it's way more than half of the programmers that will be below average. Which strengthens your point in a way.)

[gdp]

> What about certifying the software that Joe the developer creates, rather than Joe the developer?

There's a simple reason why - it costs x to certify 1 developer, and y to certify a piece of software.

For most software, "y" is still too expensive to offset the potential damage caused by software failure (compared to a jetliner, where the potential cost is huge).

A programmer may create hundreds or even thousands of pieces of software during a career. Average x over that number of projects, and it's negligible. Multiply that number by y, and you have an enormous figure.

Also, the cost of certifying a piece of software to provide any kind of assurances of quality is time consuming and expensive. Compare this to a certification that shows that anyone who wants to claim to be a "software engineer" actually knows the first thing about software engineering, and is at least capable of producing software in a systematic way.

It's just like we license drivers once and assume that that basic level of competency will ensure "quality driving" on a variety of road types. We don't certify each driver for each road individually.

[amalcon]

The problem is that we need to guess at what x and y are for particular levels of confidence. We can't use real-world numbers, because none of the existing developer certification programs grant sufficient confidence.

You can guess one way -- as you clearly do -- and it looks like the up-front developer certification cost dwarfs the software certification cost when amortized over all projects. This might remain true even when multiple developers are considered.

You can also guess another way -- as I do -- and it looks like the up-front and recurring costs to properly certify a developer and maintain that certification would be very large. It could be so large that, even when amortized across many projects, it's still comparable to (or even more costly than) certifying the individual projects.

[roedog88]

I think the idea that certifying developers and not certifying safety critical software is a non starter. On safety critical projects, certification will add to existing and onerous validation and verification costs.

The process requirements on these projects (DO-178B) call for an organization to institute a systematic way of developing software across the project that everyone on the project follows. An additional certification for each individual is not necessary. Either they can follow the process or they are reassigned.

[gdp]

You appear to have inferred incorrectly that I am talking about safety critical work.

I am not. I'm talking about anything that people are expected to spend money on. The certification passes to the software, and then users are capable of making slightly more informed decisions.

I'm a formal methods geek, so I generally advocate theorem provers for stuff which is actually safety critical. To me, certification of developers would just be a way of lifting the general quality of the non-safety-critical stuff that still has the opportunity to cost people lots of time and money.

Think about viruses, for example. People complain about viruses causing millions of dollars of damage.

The virus didn't actually do any damage. The shonky programming that allowed the virus to propagate in the first place did.

And that's why certification is important. Hopefully it would at least ensure that people writing software are familiar with the issues that can affect software quality.