At its worst, formalized verification is an elaborate and overcomplicated process for converting coding bugs into specification bugs.

This is a huge copout and isn't at all correct. It dismisses a bunch of insanely intelligent peoples' life works.

Can't you use this same argument when being critical about any technology, or really anything complex at all?

It is some kind of logical fallacy - just because really smart people worked on something and dedicated their lives to it doesn't mean it is good or correct or that it works.

The point was more the parent comment was dismissing the work clearly without understanding any of it. They employed a sound byte to disclaim something. You're welcome to go read the prose on the subject. Parent commenter clearly hasn't.

It's a wry observation that I made when I was forced to learn it by those very same researchers who hyped it up as the future of software development 20 years ago.

It was sold as the "rolls royce of testing" which I think overstated its capabilities and effectiveness.

Okay but things are entirely different than 20 years ago.

The problem with formal verification is that in practice, many problems can't even be formally specified. Try formally specifying the requirement "the GUI should work as expected" – it's hard enough even to test this, and impossible to formally specify.

Of the remaining problems, the complete, formal specification would be larger, more complex, and more bug-ridden than the implementation. "Hey adwn, can you add a button to the GUI so the user can export the current dataset to a CSV file?" Maybe 20-50 lines of code, but a full specification would be a multi-year project – and it would be almost certainly wrong.

> The problem with formal verification is that in practice, many problems can't even be formally specified.

And even if they can, where does that formal spec come from? Sooner or later, it comes from some informal understanding of what's supposed to happen. Well, how do you check that the formal spec is correct? Because if you can't do that, all formal verification can do is tell you that your program correctly implemented the wrong thing.

And I don't think you can do that. First, I don't think you can go from informal to formal by formal means. Second, even if you can, you still can't verify the informal spec. So formal verification can only go so far.

I can recall several things that were bugs in the spec. Log4j is the most recent example. Heartbleed I think was a specification bug. Back in the day, a lot of computer viruses were helped on their way by bad ideas in the specification of Microsoft Outlook.

Not everyone has a spare 1000 programmer-years to write their script in Coq though.

Model checking can be pretty productive relatively speaking, however.

Yeah this is unfortunately true. The developer experience aspect needs a lot of work here. I don't disagree. But the assertion that you can't write bug free code is still false, regardless of the amount of effort it takes.

Lots of my bugs are because of unexpected data. I imagine these systems cannot account for that?

That can be improved if you practice to "switch hats" and try to break your own code in all ways you can think of. That is something worth doing when you review other people's PRs as well. Expect all possible inputs before they hit you. Kind of a human fuzzer.

Well, yes this is good advice no matter what, but it doesn't get rid of the bugs!

Of course they can.

How?