Kervin Pierre schrieb:

In the solution, there's a C# project called "OpenConnector". This buildings the executable used to access the network amongst other things. This exe communicates to the MAPI plugin via a shared memory protocol. Almost all ( if not all ) CalDAV communication is done in this exe.


Maybe my old abandoned Pascal project is still of use for the structure:
I implemented only stubs, these just serialized the request and sent it to a socket. On the other end there was a worker daemon unserializing the call, doing the work, serializing the reply and sending it back.

The idea behind that (and this part worked OK) was, that the worker daemon could run on the same machine (use either socket to localhost or named pipe) or on the server (use real network socket). The worker daemon itself can do different things: Go to the Database directly (e.g. Zarafa server or just a DB instance - MSSQL or MySQL or whatever) or proxy to something else (Zarafa via API, Scalix, CalDAV, etc.)

My (finally unsolvable) problem was, that there was no good way for memory management - using the functions from the MAPI helper object was not really working from Free Pascal 1.x, all workarounds and hacks ended in either extreme memory leaks (Read: Never free any memory) or weird segfaults.

The problem currently is that I am using a course-grain locking/polling algorithm for synchronization between the C# executable and the MAPI plugin. It's really too slow for any meaningful volume of data.


Just to make that clear: In my model there can be (quite many) concurrent requests from one instance of the MAPI plugin to the worker daemon, so the locking is moved to the datastore level, as I figured a Database is quite good in locking.

The MAPI API and the .Net CLR cannot be loaded in the same process space. So using regular C#/COM is out. But one solution I haven't tried as yet is COM+ and dllhost.exe. See...

http://www.codeproject.com/KB/dotnet/complusnetpracticalapp01.aspx

This _/should/_ provide the isolation between MAPI and .Net CLR that we need without the course-grain locking.


Sorry again - I am an old fart and have never (really) learned C[++], I come from the dark Pascal ages and skipped directly to Mono (.NET) and PHP.

Basically: If somebody (the two of you?) manage to get the C++ stubs running, so that every call is just serialized in some documented way and passed on to a socket, I am happy to provide the following layer (in C#).

I guess it's a smaller step from there to connectors for different groupware servers.

HTH,

Eugen

------------------------------------------------------------------------------
_______________________________________________
otlkcon-devel mailing list
otlkcon-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/otlkcon-devel

Reply via email to