At 15:34 4/11/2008, faconti [..] ... wrote:
>Hi everybody.
>
>I have made several improvements (well, I guess they are) to
>DeploymentComponent. I am looking forward for the feedback of the community.
>I removed a couple of bugs and made changes aiming to improved the
>Deployment Component.
>
>First I tell you the philosophy of the changes:
>
>1) You want to load and unload components easily and several times.
>For example, you have a producer and 2 consumers. You may want to
>stop both the consumers but NOT the producer.
>2) Very ofter you do testing and you notice a bug; you correct it
>and you upload the new file on the robot to test again. In our
>example, you want to:
>- kickOut one of the two consumers (consumer_A),
>- overwrite the dynamic library (consumer_A.so),
>- reload the library and the consumer
>- start the comsumer_A again.
>- of course you want to do this WITHOUT shutting down the producer
>and/or consumer_B.
>
>So, summarizing, these are the main changes I have made:
>
>A) some more functions to kickOut a single component or the
>components that were launched by an XML file.
>B) bug correction: you have to check that there aren't duplicated
>items in the vector ConnectionData, otherwise you will get a false
>in configureComponents and you will not be able to startComponents.
>C) it is much better to remove the element of the map "comps" in the
>function unloadComponentImpl(cit) (written by Peter), otherwise it is a mess!

So do your patches require my patch ?

>D) BIG CHANGE: You can reload dynamic libraries in run-time. If you
>are asked to import a dynamic library that was already loaded you
>can close and reopen such library (in this way you can apply a
>change in case you have overwritten the file.so), but ONLY if there
>isn't any instance allocated whom Component Type was loaded from
>this file (actually, I don't know what would happen otherwise, but
>it sounds safer). For this reason I need to have changed the vector
>LoadedLibs and I store the type of a component in ComponentData.

Cool!

>E) in configureComponents() if you try to configure a peer that is
>already configures and works correctly, just give a warning, but
>don't force to return a false.

As I mailed before, this is ok for me, the current implementation
is/was way to strict.

>let me know what you think about the changes.

Thanks for contributing, I believe your patches made a big usability
improvement for rapid testing.

Peter