[ros-kernel] Designing modules

[Jarmo M. <jm...@ho...>]

Hello,

I've been reading this list for a while. I thought that I could
comment about designing.

I learned UML two weeks ago and now I am trying different UML CASE
tools. Enterprise Architect is the best one I've tried so far. I
drew two versions of ROS modules. One version is based on the idea
that every supported subsystem (Win32, OS/2, Posix, etc) has its
own module. Other version which doesn't have Win32 subsystem.

If Win32 is in its own module, this means that gdi32.dll,
kernel32.dll, user32.dll and others would be very thin. They just
call functions in ROS kernel module. Usually just calling
equivalent function in kernel. This would add small overhead to
every Win32 API call. For example

int MessageBox( ... )
{
 return RosMessageBox( ... );
}

RosMessageBox would do the hard work (my imaginary name for the
function). ROS functions should be compatible with Win32 API
because of the speed i.e. the parameters and return values should
be the same as in the same Win32 function. Some APIs could be
changed for simplicity and for other issues such as better
supporting other OSes.

The OS/2 subsystem would call the same function(s) in kernel.

int WinMessageBox( ... )
{
 // translate parameters before the call

 int result = RosMessageBox( ... );

 // translate result

 return result;
}

Remember to map error codes!

I understand that the above idea is dropped. Files like user32.dll
are moved to ReactOS kernel module. Win32 applications calls the
function as usual, but now the user32.dll does the hard work. OS/2
application links with their usual libs, but the libs calls
user32.dll and such. This approach might be a problem for some
APIs.

int WinMessageBox( ... )
{
 // translate parameters before the call

 // call user32.dll
 int result = MessageBox( ... );

 // translate result and map error code

 return result;
}

Both designs has their advantages and disadvatages. Anyway, I
think that the Win32 module includes is better one. It has more
capacity to get better. NT has layers in which API calls lower
level and which may call lower level. Does it actually hurt so
much if there is an extra level which does so little? But on the
other hand, you get more speed to Win32 apps if user32.dll etc
does the job without extra call.

Attached is both of the designs.

JMu