You already answered it, the memory addresses are virtualized, that's both for security and to allow 32bit programs to use the full possible memory range and not mess each other. That's already been done with Windows NT if not before. The only problem with the relocation table would be, when accidentally it collides with a loaded dll that can't be relocated either - which however is very unlikely since it is a key aspect of dlls to work with each other.Spocks-cuddly-tribble wrote: ↑Mon Apr 18, 2022 11:53 am If I'm not mistaken only the first instance of BotF can use the preferred RAM address? (albeit modern OS's might do some virtual magic to solve the issue?)
Regarding mpr97, I found the old engine crashes trying to determine the device name. But to my believe the root to the issue is that dxwnd didn't hook the second window, which it can't since it doesn't support to map hooks for a second instance with same executable path, cause the path is used to map the hooks. The documentation even states, that only the first instance will be hooked, and that the multi-instance feature is unfinished.
TC you already noticed it is not hooked when it tried to run in full screen mode.
It is pretty clear that Dxwnd needs to be fixed to get this work. If it was mpr97, then you shouldn't be able to run two single player instances with copied folders either, nor be able to run a second mod. The issue is bound to the path name. Why it is implemented so I can't tell, but I guess it's just not been considered when it's been written, and now there'd be quite some code to refactor.