Hacker News Reader

Comment by digi_owl | Hacker News Reader

Seems like if you leave it up to the cable manufacturers to implement parts of a standard (beyond X wires at Y gauge minimum) they are bound to fuck something up.

Just look at the amount of screw-ups over the resistors used in USB C cable variants, in particular the one used to indicate that the cable is an A to C converter.

View on HN

Applies to software too. Standards need to be strongly defined with no leeway. Parsing should be tight and leave no room for creativity. (Text protocols like HTTP I'm looking at you!) Anything that deviates should be rejected by reference implementations instead of trying to be "robust" by accepting junk.

"Standards need to be strongly defined with no leeway."

Actually, I think there should be leeway - but only in one direction.

Specifically, I am thinking of the suggestion: "be conservative in what you send and liberal in what you receive".

I think that this is what the parent was arguing about: Being liberal about what you receive often leads to every different implementation handing the undefined cases in their own particular way, which leads to incompatibilities or one of the implementation becoming the de-facto standard.

Liberal in what you receive should only apply for human user interfaces.

If you apply the rule on machine to machine interfaces it eventually leads to all applications having to be bug-for-bug compatible with the most popular implementation. (Internet explorer 6 usually being the prime example of this)

Reply on news.ycombinator.com