>>>Do you think caching could create autoinc duplicates?
>>
>>The only way you can get duplicates if one PC updates autoinc (in the table header) and other still see old version of the header. Does it sound as caching to you? :)
>
>Id does sound as caching,

Sounds more like bad caching to me.

>and I believe it is caching, but is a logic based argument, and the counter argument is that if cache is not safe then we would see a lot more data integrity problems across the system. Unless I can come with a way to prove it, it remains more like an opinion.

But then maybe not - the other cases where you may expect the caching to screw up the table are old cases, which pretty much existed as early as 2.x, and were probably met, recognized, diagnosed and fixed already. Autoinc fields are a bit of a novelty, haven't exactly gained widespread use (because the majority of us already had our own ways of generating keys, or had commercial or our own frameworks which did that), so I'd expect that only a handful of people ever ran into a problem with that. You may actually know the other two, if we only knew who they were ;).