DarkWyrm wrote: >> 50-50. >> Yes, because it's elegant and easy to work with. >> No, because, if a window crashes while it holds that area's lock, >>it would >> affect other windows. >> Yes, because that area won't be used that often. > > Not really. I wasn't very clear in the fact that these shared areas > would be created for each app. If you look at R5's Application.h, you > can see setup_server_heaps, ro_offs_to_ptr, etc. which are all > functions to deal with the server heap, which is nothing more than an > area shared between the app_server and the app. Just because R5 does it this way, it doesn't mean you should do it the same way. The risk stays the same. >> We do that? I though we were using a only-once allocated buffer. >> [ having a look at BPortLink... ] >> Ahhhh, I'm gonna kick some ass, and that's gonna be Phatz's! > > I just got up a bit ago after not sleeping well last night, so I'm not > thinking completely clearly, but I just looked at Attach, and it > doesn't look like it does. That is, unless it does it in FlushCompleted > or something. I'll have to look at it more later. And AdjustReplyBuffer(). >> I thought we agreed on using a static buffer!!! If I send a BIG >> picture/region/polygon to server the buffer will get bigger and >>bigger and >> then drawing instructions instead of being sent to server, are >>cached >> slowing down server response. By NO means the buffer should be >>greater >> then 2K!! >> DW, do you agree with Phatz? Why? > > Relax. If that is what BPortLink does, we can add another reallocation > check to downsize the buffer if it's above 2K. That way, we could have > our cake and eat it, too. I really don't think so. There are enough memory allocations/deallocations. Better use an area. > My only question is how big a message buffer > we can send over a port. 2MB. Adi.