Your work is appreciated, and I really want to try it, unfortunately it doesn't support MinGW yet.
maybe MinGW is supported later (requires only to adapt the Windows asm to GAS style + entry in Jamfile).
1) The doc says suspend() throws nothing, but it does throw ex_unwind_stack for unwinding. Maybe it'd be better to make ex_unwind_stack part of API, and let the user catch and rethrow.
ex_unwind_stack is swallowed by boost::context - it is required in order to implement expilict stack unwinding. It is important that the user dosn't swallow this exception! I think the best would be that the user dosn't know the type of this exception (unnamed exception?) - not sure if I should supress this info or note it explicitly in the docu.
2) Many functions seem unnecessarily virtual, IMO, only dtor and exec need to be virtual. icontext could be replaced with platform-specific context_base without Allocator stored, and you could store Allocator in context_object instead, passing it to templated methods of context_base when needed (e.g. on constructon & cleanup).
boost::context should be independed from the used Allocator. That requires that the pointer to the impl held inside boost::context doesn't contain the allocator type in its class declaration. Thatswhy icontext was introduced. context_base uses the allocator in order to allocate/deallocate the stack as well aplly it to fcontext_t. But maybe you can send me your solution so I can take alook at it? Oliver -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de