2015-09-05 20:33 GMT+02:00 Thomas Heller <thom.heller@gmail.com>:

I don't think that is what Hartmut or I wanted to suggest. What should be possible though is to implement something like HPX on top of Boost.Fiber.

OK, then it was a misunderstanding

I currently don't see why this design wouldn't allow to have threads migrated

I'm confused - I assume you mean '... have fibers migrated...'?!

, even if the TLS value is cached, it should still be valid, right? In that case you would have something like a benign "race".

- fiber F1 runs in thread T1

- T1 runs fiber_manager FM1

- F1 gets migrated to another thread T2 which runs fiber_manager FM2

- as I figured out that F1, running in T2, still accesses FM1, but its should be FM2

so I'm wondering how HPX gets the TLS issue right