[security-dev] Dependency of idm/impl on config module

[Shane Bryzak]

On 08/02/13 13:00, Marek Posolda wrote:
> On 08/02/13 14:31, Shane Bryzak wrote:
>> On 08/02/13 02:00, Marek Posolda wrote:
>>> On 07/02/13 17:01, Shane Bryzak wrote:
>>>> Oops, I didn't mean to commit that before discussing with you, it must
>>>> have snuck in with some other changes that I had to make to help out
>>>> Bruno with one of the Aerogear examples.  In any case, we do need to
>>>> remove the dependency on the config module from idm - I'm ok with a
>>>> dependency going the other way though.
>>> Ok, but why we need to remove it? Is it some xml related conflict? I 
>>> have same problem today again. I fixed it locally in my env, but i 
>>> won't commit it until we sort this out.
>> The most immediate problem is that it's breaking deployments - the 
>> config module is pulling in the federation module, which causes a 
>> deployment error in AS just from having the jar file present. 
> Hmm... actually the "config" module shouldn't be dependent on 
> "federation" module. Few weeks ago it was the case (and maybe you are 
> using older version of picketlink in aerogear), but in latest 
> picketlink master it's not the case anymore. Currently it should be 
> opposite. In other words, "config" module is currently dependent only 
> on "common" module and
> both "federation" and "idm" modules are dependent on config module. Or 
> am I missing something?

You're right, rebuilding the module fixed the dependency issue.

>> Besides that though, the only dependency we should have in idm is on 
>> the common module.  Also, configuration should be able to be written 
>> as a totally separate concern from IDM - is there a reason that 
>> XMLBasedIdentityManagerProvider and all the resolver/* classes can't 
>> go in the config module?
> I think that if we want to use XML configuration in IDM unit tests, we 
> need to have those XML configuration classes and classes like 
> XMLBasedIdentityManagerProvider accessible from idm module. Those XML 
> type classes also needs to be accessible from "federation" module.
>
> I think that deployment structure with:
> - "common" module as the base module
> - "config" module dependent only on "common" module
> -  "federation" module dependent on "config" module (and "common" module)
> - "idm" module dependent on "config" module (and "common" module)
>
> seems to me like most natural.
>
> wdyt?

-1, if we want to use the config module in the tests then just make it a 
test-scoped dependency.  In many if not most deployments PicketLink will 
be used within an EE environment and the config module will be unnecessary.

> Marek
>