[omniORB] Is resolve_initial_references() implemented right?

[Renzo Tomaselli]

>Does anyone have more experience with DLLs on Windows?

We use dynamic loading on Windows, Linux and Solaris without any problem:
same architecture, same sources beside a few very small sysdependent issues
like dynloading syntax.
A loading service which is a dll itself is implicitely linked to an
application (which can be a service on NT). Then it can load explicitely
other dll services on-demand (and unload them as well), since the loader is
itself a published CORBA service.
We even keep stubs/skeletons into separate dll for space optimization
reasons and all of these dlls in turn link OmniORB dlls. For Windows we keep
a double copy of all because of the well known issue of release/debug
forking.
All of this works from python (both in/out of process), from an Active-x
wrapper (Windows only), from an NT shell-extension.
Any loaded service dll call orb->resolve_initial_reference() to get the root
POA.
One thing we do not use at all is the name service, since we preferred to
build up our own registry-like repository (we need to store generic named
values for config/admin purposes, not just objrefs; access control is
another issue).
Absolutely no problems for current OmniORB version (3.04), nor for previous
versions since 2.6.
Cheers,
                                             Renzo Tomaselli
---------------------------------------------------------------------------
TecnoTP s.n.c. Special Information System Design
Maso Pelauchi I38050 Ronchi Valsugana,  Trento TN  ITALY
Tel. +39 0461 773164      Fax. +39 0461 771514
e-mail: renzo.tomaselli@tecnotp.it
---------------------------------------------------------------------------