I am writing a C++/cli wrapper around a C++ project so that I can call it from C#. The C# side calls a cli method [say Put(String^ key, String^ value)] millions of times to pass a large set of key/value managed data, followed by another call [say Commit()] to ask the C++ object to do something with the data. I need to be able to keep all that data in memory until the last call -- so I am using a std::vector and trying to convert all that managed Strings to unmanaged std::string and hold them in the vector in my cli code.
My question is -- What's the best way to convert the managed data in this scenario? Pinning the managed data inorder to keep unmanaged pointers valid seems really bad idea as the data is not used immediately (also this won't work if pinning gets out of scope immediately after the function call - that results in access voilations). Another option is to allocate a "big enough" buffer through malloc and keep adding the data to it -- if that's better, then: (1) is pin-ptr followed by a memcpy (the pinning would go out of scope once Put method returns) the best way to convert/copy a single String^ data? (2) since I won't know the total size upfront for maloc, what's the best way to expand the memory from initial malloc? I would appreciate any pointer to a good design solution for this scenario.
If you need to queue the data up before you use it, I'd probably use a managed container to do that. You called the two strings 'key' and 'value', are the keys unique? If so, use a managed
Dictionary<String^, String^>. If not, use aList<KeyValuePair<String^, String^>>, and do all the conversions to unmanaged strings when Commit() is called.When Commit() is called, you could do one big malloc to hold all the individual strings, or you could convert each string individually, with many individual mallocs (or other allocation method). As long as you properly release the memory in all cases, either should work fine.