Is it possible under Solaris 2.x to build a shared "entity"
( using C) having the following features:
1) global variables in this shared object would be readable
AND writable by executables linked to this shared "entity"
2) after all executables that use this shared object had
completed, the current values (in memory) of the globals
would be written back to disk.
3) it would be nice to have a "main()" in this shared "entity"
( making it an executable itself ) for initialization
4) I am assuming that any functions in this shared entity
would still be sharable amoung the executable pgms
linked to this shared "entity"
These features would allow you to have the following
1) running the main() shared "entity" would initialize the
2) global variables would retain their last values across
3) A second pgm, pgm2, would be able to "see" changes that
another pgm, pgm1 had made.
I am NOT looking for an ipc (Solaris shared memory) option.
A shared file option would give the same effect..., except
this is also not what I am looking for. This option imposes
pointers/handles for programming purposes.
I want the linker to do ALL the work for me, in the sense
that if I had a global variable A in the shared entity, I
could simply reference A in pgm.c, and NOT have to use a
pointer/handle to A in the mmapped file case.
This really boils down to a question of how to manipulate
the linker to produce the desired results. It appears that
shared objects linked under Solaris, when you write to them,
the pgm gets its own copy of the segment. The linker needs
to be told not to do this, and flush any changes back to disk.
As an aside, under VMS and Fortran, you could create an
executable shared data/code object, and link it into other
pgms, and have the features that are mentioned in items
1) thru 4).
Any help or suggestions would be appreciated.