Debugging type level bugs

[sassa_nf]

I don't know, do people really use debuggers for real code?

https://github.com/oracle/helidon/blob/355f646492e6529734a5b1a2aa50422bf8c75a5a/common/reactive/src/main/java/io/helidon/common/reactive/MultiFlatMapPublisher.java#L198-L235 - the bug here, to my mind, is best described as a type-level bug.

Not a variable of the wrong type, more like part of the computation has the wrong type.

I am not sure how a debugger can help, even if one can find a way to coax the code go through the problematic code path. (But first one would need to realize it is problematic, right?..)

The only assurance you can have, is a formal proof. The next best thing is code review.

Comments

[juan_gandhi]

Using Scala might have helped them, but well... Seems like since people just don't know much about functors, they know even less about monads.

And the code is barbaric. Debug or not, how would one test it. Hopeless.

[sassa_nf]

How does one test linearizable code is linearizable? You can only prove formally. (And that's what I find fascinating about these types of problems)

The code is horrible. But even the (reactive) spec (by this guy and a few Akka guys) has a few bugs. I attribute it to the lack of education.

[juan_gandhi]

Well, lack of education is a sad fact of life. Another problem with it is that you can't explain to them anything.

[sassa_nf]

In this case it went like this:

- I have a proof there are the following issues ...
- Do you have a reproducer?

:)

[juan_gandhi]

It was a non-constructive proof.

[sassa_nf]

There is a gap between a constructive proof (show a valid interleaving of events can lead to invalid outcomes) and a reproducer (actually cause the events line up at run time).

Actually, I did line them up, through tedious modification of the code - literally suspend threads reaching certain lines, and resume in a particular order. But I found it funny that the demand is to get "a reproducer". It's not like "if you run with these inputs, you always hit the issue". It is more like "if you yell at your hard disks hard enough, you will get this sometimes".

[juan_gandhi]

I see. Maybe it's hard to convince the management that it's important. A couple of weeks ago I created a case named "race condition in..." - and then they wanted me to prove it is one. I did not care, I just wanted to walk around and provide a solution (retry).

Then they wanted me to "investigate and reproduce". Refactoring not allowed, so I can't seriously even test the huge method. So I said, ok. Then another guy decided to tweak that method, to "make it better". For some reason, he was allowed to do it. Did not help. Now it's called "race condition".

If it were like 10 years ago, I would have been totally pissed off. Now I just smile at those guys (like dr. Freud, kind of). Who cares, actually.

[sassa_nf]

I am probably you 10 years ago :) a lot to catch up

[juan_gandhi]

:) Yes, it's all psychology and hormones. (And some "negative" experience.)