Feature Request: Provide non-throwing methods to open/create shared_memory_object

まず shared_memory_object API と、参照されている managed_open_or_create_impl.hpp のコード、特に 285-338 行を読んでください。例外をスローしない create/open 操作で、POSIX と Windows において要求されたサイズを含め、セグメントが作成されたかどうかを報告する方法を確認してください。API がコンストラクター例外なしで同期のユースケースをサポートし、作成とオープンを区別できれば完了です。

My use case is as follows. I need to synchronize multiple processes or threads initializing a shared memory region. No select process can be considered the "master", so it is not known which of the processes will actually create and initialize the shared memory object. The natural solution would be to construct shared_memory_object(open_or_create), but this introduces a problem that the caller doesn't know whether the shared memory was created or an existing region was opened. This is important to later decide whether the caller should resize the region to the desired size and initialize the contents or try and adopt the existing one. (Testing the region size for 0 is not safe because it introduces a race condition.)

The current API does not provide a way to implement this efficiently. It is possible to implement a loop trying to construct shared_memory_object(open_only) or shared_memory_object(create_only), but the constructors throw on failure, which is unnecessary overhead and awkward to use in the caller. Indeed, you have implemented exactly that here. It would be better to have a set of methods in shared_memory_object that allow to create or open the shared memory and able to report the result without throwing - for example, through an error_code argument.

Also, please consider adding a way to create a shared memory segment of a given size. On POSIX, this would allow the aforementioned loop to be implemented in Boost.Interprocess instead of user's code. On Windows, CreateFileMappingW already accepts the desired size. This new API would also have to provide a result so that the caller knows whether a new segment was created or not.