We're currently using message_queue on Windows, but want to avoid files being created on the user's filesystem (even in the application's config/cache directory). Moreover, we'd like to have the semantics of windows_shared_memory (mostly the destruction once all handles to the memory are closed).
Currently, message_queue uses shared_memory_object. Would it be possible for users to specify the underlying abstraction (e.g. message_queue_t<offset_ptr<void>, windows_shared_memory>)?
I experimented a bit by using managed_open_or_create_impl<windows_shared_memory, 0, false, false>, which worked fine for me (I might have missed some bugs as I don't know IPC too well). One downside is that message_queue_t::remove can't be used with windows_shared_memory, as it doesn't provide that functionality (the documentation for ~message_queue_t should be updated, too).
I'd be fine if this was done via a #define, but it would be nicer if an approach similar to the previously mentioned could be taken.
If this is something that could get included in the library, I'd be fine with implementing it.
We're currently using
message_queueon Windows, but want to avoid files being created on the user's filesystem (even in the application's config/cache directory). Moreover, we'd like to have the semantics ofwindows_shared_memory(mostly the destruction once all handles to the memory are closed).Currently,
message_queueusesshared_memory_object. Would it be possible for users to specify the underlying abstraction (e.g.message_queue_t<offset_ptr<void>, windows_shared_memory>)?I experimented a bit by using
managed_open_or_create_impl<windows_shared_memory, 0, false, false>, which worked fine for me (I might have missed some bugs as I don't know IPC too well). One downside is thatmessage_queue_t::removecan't be used withwindows_shared_memory, as it doesn't provide that functionality (the documentation for~message_queue_tshould be updated, too).I'd be fine if this was done via a
#define, but it would be nicer if an approach similar to the previously mentioned could be taken. If this is something that could get included in the library, I'd be fine with implementing it.