Bernd schrieb:
I ended up starting a separate thread in my dll, told the RTL that this new thread is now the main thread (and was surprised that this actually worked) and then wrote my own message loop (because application.run did not seem to work properly).
Up to here I'd think that a new TApplication type for DLLs would be a nice solution.
I also had to synchronize all other function calls from the host with this thread because they came from even more different threads,
This sounds a bit weird. Or is the DLL *assumed* to synchronize all these calls with the GUI thread (be application or DLL main thread)?
At the end it worked but it all looked so ridiculously complicated and fragile and I had the impression it was sheer coincidence that it actually worked so that I finally decided to abandon these experiments and instead make the gui part of my dll a separate .exe file that communicates via some sort of IPC (I ended up starting the GUI exe with a TProcess and send/receive simple commands via stdin/stdout). This still seemed somehow complicated but at least it did no dirty undocumented tricks with the LCL and seemed quite robust (and enough for my needs).
I'd thought that a SendMessage would serve the same purpose (thread switch). DoDi -- _______________________________________________ Lazarus mailing list Lazarus@lists.lazarus.freepascal.org http://lists.lazarus.freepascal.org/mailman/listinfo/lazarus
