Re: [riena-dev] Interface not visible from class loader!

[Scott Lewis <slewis@xxxxxxxxxxxxxxxxx>]

Hi Christian,

Christian Campo wrote:
> Hi Olaf, Scott,
>
> @Olaf,
>
> I think your solution will work, but I am not a big fan of looking at 
> the call stack. Also your solution confuses me because some variables 
> like CLASSCONTEXT are static. Are you sure that this solution works 
> even if multiple threads try to create a proxy at the same time with 
> different bundles trying to do that ?
>
> Not sure why that is any better than the buddy policy. Also the buddy 
> policy thing not only deals with the interface itself but also with 
> all return values or paramters. If they are not in the same package as 
> the interface they need to be buddy to. Your solution would not help 
> anymore.
>
> @Scott: While that solution should work I guess, I am afraid that it 
> makes the system way slower. I have already read somewhere in the 
> internet that buddy policies are a timing problem. Because they mean 
> that for every class the classloaded search a large list of bundles. 
> If it would search all bundles, I believe the result could be even 
> worse. Since they are not only consulted for the interfaces but 
> possibly also for return values and parameters, I am not sure that way 
> is a good idea.

You are right that Dynamic-ImportPackage can be relatively slow, but 
compared to the deserialization and i/o in most environments it's not 
going to be a problem.

> but thanks for the input. I know BuddyPolicies are not ideal but they 
> are currently the best way to handle this problem I know.

Buddy policies can be used, but they have their own disadvantages...for 
example with some VMs it has been possible to get deadlocks.  Also, they 
only work with Equinox (i.e. they are not OSGi but rather Equinox 
specific)...and except for the global policy there can/will be 
dependency issues (for example, having interface plugins (used to create 
proxies) depend upon part/parts of Riena).

Scott