Dynamic Libraries

[Chris Johns]

Angelo Fraietta wrote:

> What I mean is that a user can develop compiled modules and download
> them into the flash filesystem. When RTEMS loads, looks in the flash
> disk and loads any modules. This allows me to add modules at run-time
> instead of compile time. The modules would be quite specific. It
> ultimately means that the firmware can grow with the user -- i.e. they
> can design and add modules to their hardware product without having to
> rebuild the whole target.

Why not supply the libs they need to build a program ?

I am not saying loadable modules do not have there uses, they do, 
how-ever it seems in this case as something better suited to a build 
machine to do once rather than having the target do each time it is 
started plus the dependency problems separate parts generate.


Other issues come into play with loadable modules which can become 
important. The first is a fixed configuration table. The "unlimited" 
support for objects (not POSIX threads as discussed else where) is a 
start. The other is the workspace and heap are separate. A single memory 
region for the workspace and heap to use may be a better design. For 
example I may have a program with many many threads and large stacks, 
while someone else has a few tasks but malloc is used all the time for 
large amounts. These require different workspace/heap splits.

-- 
  Chris Johns, cjohns at cybertec.com.au