klaus triendl

@fnc12 @PandarinDev is right in that anything can happen because even the increment/decrement operation itself is not atomic (it consists of read/operation/write). On top of it, the `retain()` and `release()`...

@Akster009 @PandarinDev Are you able to try the PR branch whether it solves your problem? Just note that this is on latest dev.

@robin-arcsite You are absolutely right about this kind of race condition. The way connections are currently handled requires that you use `open_forever()` when using multiple threads. At this stage I...

@fnc12 Because anything can happen between increasing/decreasing the reference count and calling the sqlite APIs in a multi-threaded scenario.

Sure, it's called "asterisk" for a reason :) The bad thing is that the problem hits you unexpectedly and is a silent type hole. It's not even obvious to experts...

BTW, this is working in the CTEs feature branch, example [select_from_subselect()](https://github.com/FireDaemon/sqlite_orm/blob/f69701e56028766a550952a50f3992ce6954b4c8/examples/common_table_expressions.cpp#L517).

@ldionne Yes, this would actually solve the need. Variable templates OTH are quite handy to actually work with objects, so I would choose a different nomenclature to get both. @ricejasonf...

> I have a better idea how to implement it Apart from even more room for improvement, I just want to mention that sqlite_orm has a very nice appveyor test...

@nmgwddj Can we close this ticket?

This is fixed by PR #1143, and is very closely related to [Is it possible to write 2 or more asterisks in a select? #1106](1106). So it is possible to...