[jboss-dev] building the documentation for JBoss public and private API...

[Anil Saldhana]

Think of a situation:
There is a system integrator who wants to integrate his project/product 
to run on JBAS5+.  Now he has a need for a deployment descriptor - 
some-vendor-xxx.xml sitting in the META-INF or WEB-INF or on the 
classpath. The recommended approach in JBAS5+ is to write a parsing 
deployer to generate some metadata and then write some component 
deployer to make use of the metadata to create the MC beans etc. 

Given this,
a) first someone has to tell this SI that he needs multiple deployers.  
Let us assume that he will figure this out from tutorials/articles that 
we will write. :)
b) next, he starts coding the deployers.  Either he can subclass one of 
the umpteen AbstractXXXDeployers in the MC/MC-integration or he can just 
subclass one of the many deployers already available in AS5+.  But 
certainly once he figures out what is needed, he needs all the 
information for the classes in scope (and definitely needs all the 
javadocs).

This would be a use case where there is deep integration into the JBoss 
infrastructure. We have to assume that most of the api in usage is going 
to be public.

There there is the other side,  pure Java EE developers for whom the API 
in question is just JavaEE usage.  But all the annotations that are from 
JBoss need to be documented.

All documented API is public.  Everything else is private.

Scott Marlow wrote:
> Anil Saldhana wrote:
>> If you look at the Java EE javadoc bundle for example - there is no 
>> clear separation between what is applicable to SPI implementers, 
>> container vendors and JavaEE users.
>>
>> We really need documentation on API that is for public usage ( both 
>> users as well as JBoss system integrators) and of course the private 
>> api.
>>
>
> When we mark an API as public, what are we trying to say to 
> application developers?  I have the same question about why we mark an 
> API as private.  One possible answer for public API, is that customers 
> are expected to call public API.  Another comment that we might make 
> about public API, is that it is not expected to change often (but 
> may).  We might want to say that an API is marked private if it can 
> change more often (perhaps in a minor release cycle).  We need to 
> establish some rules that we can live with (for a number of years) 
> about the implied contract behind the public/private API annotations.  
> We could even do something similar to the Open Directory Server link 
> that I referenced earlier 
> (http://www.opends.org/promoted-builds/2.0.0-RC1/javadoc/org/opends/server/types/PublicAPI.html).  
> The OpenDS javadoc is pretty clear on the contractual meaning of their 
> PublicAPI annotation.
>
> Anil, great suggestion about adding more documentation (we have seen 
> that point raised repeatedly in the last few days).  Not everyone in 
> the community is going to be as genius as the original implementors of 
> feature X.  The same is true for internal engineers 
> (development/qa/support).  I would like to see these 
> (internal/external) engineers succeed in their tasks.  Improving the 
> Javadoc, will lower the barrier for these folks to achieve success in 
> their projects.
>
> Hopefully, in a few days, we will have an understanding of how to 
> start approaching the way that we mark public/private API in our code 
> base (so we can make some progress in that area).  Thanks for your 
> support!  :)
>
>> Scott Marlow wrote:
>>> One illegal usage might be detecting that a private (JBoss internal) 
>>> API was used by a customer application.  This would be useful for 
>>> our customers to know as private API can change unexpectedly from 
>>> release to release.
>>>
>>> Applications that use private APIs will need more refactoring during 
>>> migration to a new AS release.
>>>
>>> Perhaps we could also have a usage check for unsupported APIs (some 
>>> projects have contributions that may be considered experimental).
>>> David M. Lloyd wrote:
>>>> Illegal usages such as what?
>>>>
>>>> - DML
>>>>
>>>> On 06/01/2009 04:04 PM, Emmanuel Bernard wrote:
>>>>> +1 for annotations as we could also flag illegal usages at compile 
>>>>> time via an annotations compiler.
>>>>>
>>>>> On  Jun 1, 2009, at 17:00, David M. Lloyd wrote:
>>>>>
>>>>>> On 06/01/2009 03:19 PM, Scott Marlow wrote:
>>>>>>> We were just talking about building a list of all JBoss AS 
>>>>>>> public API calls.  This will be used to document the public API 
>>>>>>> versus what is considered private (we will include only the 
>>>>>>> public API in this "public api" documentation).
>>>>>>> One idea mentioned already is that we could use an annotation 
>>>>>>> (e.g. something like @publicAPI mentioned elsewhere 
>>>>>>> http://www.opends.org/promoted-builds/2.0.0-RC1/javadoc/org/opends/server/types/PublicAPI.html).  
>>>>>>> We would then build the "public API" documentation based on the 
>>>>>>> @publicAPI annotations (the how to be determined).
>>>>>>> We might want to also include a @privateAPI tag, for source 
>>>>>>> files that contain a mix of public and private API (or maybe we 
>>>>>>> should move anything private into separate modules).
>>>>>>
>>>>>> The way I traditionally tackle this is by using package-private 
>>>>>> access for non-API stuff in the API package.  Otherwise, a taglet 
>>>>>> would be a great way to do this (annotations are probably 
>>>>>> overkill).  Javadoc may be an old technology but it's still decent.
>>>>>>
>>>>>> One thing I also have for Remoting 3 is a set of javadoc tags 
>>>>>> that I can stick on a class or interface which causes a generic 
>>>>>> explanation to be appended to the javadoc (like "This interface 
>>>>>> is intended to be implemented by the end user" or "This interface 
>>>>>> is intended for service providers, so new methods may be added 
>>>>>> without notice", etc).  I can provide a link as soon as I get 
>>>>>> around to getting the javadoc published for 3.1.
>>>>>>
>>>>>> - DML