Hacker News Reader

That may be, but it ignores the fact that software was written for the express purpose of not just identifying if a test was being ran, but then taking action to change the results of the test has no real world use. Any case where you would want to change engine performance has no reason to be triggered only when testing. To get both included in production as-is and turned on requires quite a series of consecutive mistakes to be considered accidental.

> In a field more of us here understand, it would be similar to coding a feature in a server that dumps user passwords to the log file in plain text.

No, it's like coding that feature but explicitly making it turn off whenever it detects a security review. How is that justifiable?

Yeah, I like your continuation of my analogy. However, as I understand it, from someone I know who worked in the automotive sector a while back, all vehicles have a special emissions test mode. It exists for legitimate reasons such as you don't want to have ABS turned on when on a dyno for safety reasons. It probably also exists for less legitimate reasons arising out of the history of testing practice -- essentially that the test never has been truly representative of real road conditions, and everyone involved in the testing process knew this. That being the case, all you need then is to add one more flag to the list of "things to enable/disable under test conditions". I agree that somewhere there must be some kind of record of when and who made that configuration change though.

These details of emissions modes and engine control software configuration could easily explain the defeat.

What remains is explaining how engineering managers higher up (all the way to the top) could remain willfully ignorant of the fact that VW's diesel engines "passed" emissions without resorting to exotic tech like Mercedes' bluetec exhaust fluids. They knew there was some sort of magic in their engines that set them apart, but they either understood it was cheating, or they never asked any questions.

I suspect this will turn out to be due to some corporate mass psychosis similar to when an airliner suffers controlled flight into terrain because nobody wanted to disagree with the captain.

> What remains is explaining how engineering managers higher up (all the way to the top) could remain willfully ignorant of the fact

> What remains is explaining how engineering managers higher up (all the way to the top) could remain willfully ignorant of the fact that VW's diesel engines "passed" emissions without resorting to exotic tech like Mercedes' bluetec exhaust fluids.

How often do high level managers at a big company have a good handle on technical implementation details of specific systems and what the implications of those details actually are?

It's possible that these people either had non-technical backgrounds, technical backgrounds in unrelated areas, or out of date technical backgrounds (like they last worked with this stuff back when cars still commonly had carburetors).

Martin Winterkorn (the VW CEO who resigned) was an engineer. Specifically, he was an powerplant engineer who helped spearhead VW's diesel technology.

I wouldn't be surprised to see him go to jail.

Okay, I can accept there may be a legitimate reason to want to determine if a test is being run for a case like that, but that's a component that isn't related to engine performance and emissions. I'm not sure there's any case where you could legitimately want the engine performance and emissions to be different during emissions testing. In any case you want it different (for example, testing different parameters), you would also want to test it on the road.

> That may be, but it ignores the fact that software was written for the express purpose of not just identifying if a test was being ran, but then taking action to change the results of the test has no real world use.

I think it depends on how general the implementation was. I could imagine real world uses for a feature to generically detect certain conditions and change parameters in response (e.g. to support different settings for highway vs city driving). If you have something general enough, then the "defeat device" could have just been a ruleset that only matched a very narrow set of conditions that usually only occurs during the tests.

My point is I'm not sure of any scenario which would need that criteria to me met, and the chance that it was accidentally implemented such that it just happened to match test conditions and not other conditions, and just happens to enable running conditions that are sub-optimal for operation but allow it to pass emissions seems so vanishingly small that I can't consider it a possibility without further information coming to light to explain why it would make sense.

Incompetence happens, but requiring chained events of successive incompetence to explain an outcome to explain an event doesn't seem rational to me.

Oh, I think that however it was done, it was done intentionally. My point was only that the "code" could have been kosher and the tampering done later via configuration.

Interesting. If this is the story, then VW didn't really "need" to cheat to make the cars reasonable, so they shouldn't get too much worse when patched. That would be a nice outcome.

I'm still cynical.

I don't understand how that follows. From all I've heard, people suspect that the difference between cheating and test mode is large.

> Such a feature would be useful for certain debugging situations, but it would be disastrous to have it enabled on a production box.

I've seen that exact thing (and worse...) on a production server in some job recently. But that big all-caps warning wasn't there, the docs were absent and there was no config bit to switch on/off, I found the file more or less by accident (wondering what all the io was that ate up the machine).