On Thu, Apr 20, 2017 at 06:15:38PM +0200, Joerg Sonnenberger wrote: > (1) >> (2) >> (3) > > As I said, I consider (3) a functional regression. I don't like the > semantic overload of (2), especially since it makes the interface even > less orthogonal.
The current state is clearly broken and makes use of this interface error prone (which is not a big issue, as it is clearly internal "^_"). Doing (1) + force SS_DISABLED for the new thread is the minimal change to get a sane state. > All ucontext using interface must die and be replaced > with something sane, so making them more complicated doesn't help. Maybe, but this does not help the issue at hand, unless you have a proposal for a non-ucontext based interface to create a new thread and want to go there right now. Martin
