Re: [equinox-dev] Equinox Aspects and OSGi Weaving Hook Service Specific

[Thomas Watson <tjwatson@xxxxxxxxxx>]

Hi, Martin and Alexander.

Martin is correct. The caching aspect of weaving.hook has to be implemented as an equinox framework extension. I think it would be great to get the o.e.e.weaving.aspectj to be implemented on top of the standard OSGi weaving hooks and then provide the performance improvements (caching etc.) as special extensions to Equinox.

Thanks.

Tom



Martin Lippert ---05/12/2011 01:41:59 AM---Hey Alexander!

From:	Martin Lippert <lippert@xxxxxxx>
To:	Equinox development mailing list <equinox-dev@xxxxxxxxxxx>
Date:	05/12/2011 01:41 AM
Subject:	Re: [equinox-dev] Equinox Aspects and OSGi Weaving Hook Service Specification (56)

---

Hey Alexander!

> Ok, I now have a much more clearer view on all of the existing logic.
> I agree that going the short way is just a workaround and not a
> upgrade so I will take the long road and will reimplement all bundles
> to use OSGi spec interfaces. Do I correctly understand that
> o.e.e.weaving.hook bundle will therefore be obsoleted completely or
> does it have any extra functions?

I don't think that you can implement the caching part on top of existing 
pure OSGi specifications, but I am not sure.

What I think would be great is:

- having o.e.e.weaving.aspectj to work purely on the OSGi Weaving hook API.
So people who wanna use aspect weaving can use that on any OSGi runtime 
that implements that spec.

- having o.e.e.weaving.hook and o.e.e.weaving.caching to work on Equinox 
only and provide caching for woven bundles, so that people can use that 
together with o.e.e.weaving.aspectj, if they wanna have caching for 
woven stuff.

I have no idea what that would mean for the implementation of the hook 
and the caching, but I would start with adapting o.e.e.weaving.aspectj 
to the new OSGi Weaving spec and then would try to reduce the hook to do 
caching only.

Just my thoughts...
Tom, what do you think?

Cheers,
-Martin





>
> On Wed, May 11, 2011 at 11:36 PM, Martin Lippert<lippert@xxxxxxx>  wrote:
>> Hey Alexander!
>>
>> It wasn't clear to me if you would like to "just" change the existing
>> implementation to use the new hook or if you would like to implement a
>> completely new aspect weaving mechanism.
>>
>> So the existing structure of Equinox Weaving is:
>>
>> - org.eclipse.equinox.weaving.hook
>> This uses the basic Equinox hooks to inject a general bytecode weaving and
>> caching mechanism. There is nothing special here for aspect weaving. This
>> hook provides a service interface that other bundles can implement to
>> provide concrete bytecode modifications and a service hook for caching
>> services.
>>
>> - org.eclipse.equinox.weaving.aspectj
>> This implements the service interface from the hook mentioned above and uses
>> the AspectJ weaver to weave aspects into bytecode.
>>
>> - org.eclipse.equinox.weaving.caching
>> This implements the caching service from the hook mentioned above and
>> implements an asynchronous cache storing on the hard drive.
>>
>> My understanding of the OSGi Weaving specification is that it partly
>> implements what the weaving hook mentioned above implemented (so that
>> another bundle can implement a service interface in order to inject the real
>> bytecode modification). So in an ideal world, the
>> org.eclipse.equinox.weaving.aspectj bundle would use that interface in the
>> future to realize bytecode weaving instead of the old one from
>> o.e.e.weaving.hook. But doing this would also mean that the caching would no
>> longer work for that weaving (because the old o.e.e.weaving.hook is the glue
>> between weaving and caching).
>>
>> The other way would be to re-implement the o.e.e.weaving.hook bundle to use
>> the new OSGi Weaving spec hook instead of the old Equinox-specific ones
>> while providing the same interface to the outside world as before (the one
>> that the aspectj component uses). That would allow the caching to work with
>> this. But from an design point of view, this wouldn't provide a big step
>> forward, I think.
>>
>> Just some thoughts...
>>
>> Cheers,
>> -Martin
>>
>>
>>
>>
>>> I thought that the only thing this reimplementation would affect would
>>> be how (using osgi weaving hook vs equinox internal mechanisms) the
>>> aspects are woven, and would not affect other things such as which
>>> aspects to weave, caching etc. But given your question it seems I
>>> thought wrong.. Could you please clarify? Also I'm not quite sure what
>>> is opt-in and opt-out.. Is this about which side initiates the weaving
>>> (Eclipse-SupplementBundle vs Require-Bundle)?
>>>
>>> On Wed, May 11, 2011 at 10:49 AM, Martin Lippert<lippert@xxxxxxx>    wrote:
>>>>
>>>> Hey Alexander!
>>>>
>>>>> Thanks for your answer, Martin.
>>>>>
>>>>> I will indeed try to re-implement aspect weaving to comply to the
>>>>> specification. If I succeed I'll send you my result so you could maybe
>>>>> just review it and give it a go saving some of your development time..
>>>>
>>>> Any contributions are of course highly welcome. Please let me know when
>>>> you
>>>> have something working here. Would be glad to take a look.
>>>>
>>>> The CVS resources are at the RT repository:
>>>>
>>>>
>>>> http://dev.eclipse.org/viewcvs/viewvc.cgi/org.eclipse.equinox/weaving/bundles/?root=RT_Project
>>>>
>>>>
>>>> http://dev.eclipse.org/viewcvs/viewvc.cgi/org.eclipse.equinox/weaving/features/?root=RT_Project
>>>>
>>>> Hope that helps!
>>>>
>>>> Are you planning on implementing just pure aspect weaving for AspectJ, or
>>>> are you also looking at the different aspect resolving mechanisms (opt-in
>>>> vs. opt-out) and the caching for woven classes?
>>>>
>>>> Cheers,
>>>> -Martin
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>> Therefore could you confirm that these are correct links for latest
>>>>> source of Equinox Aspects (I got them from the same web site which as
>>>>> you said is a little outdated):
>>>>>
>>>>>
>>>>>
>>>>> http://dev.eclipse.org/viewcvs/viewvc.cgi/equinox-incubator/aspects/org.eclipse.equinox.weaving.hook/
>>>>>
>>>>>
>>>>> http://dev.eclipse.org/viewcvs/viewvc.cgi/equinox-incubator/aspects/org.eclipse.equinox.weaving.aspectj/
>>>>>
>>>>>
>>>>> http://dev.eclipse.org/viewcvs/viewvc.cgi/equinox-incubator/aspects/org.eclipse.equinox.weaving.caching/
>>>>>
>>>>>
>>>>> http://dev.eclipse.org/viewcvs/viewvc.cgi/equinox-incubator/aspects/org.eclipse.equinox.weaving.caching.j9/
>>>>>
>>>>> I was confused by 'equinox-incubator' part. You told me that the
>>>>> project has graduated to the main SDK so maybe it now resides
>>>>> somewhere else?
>>>>>
>>>>> On Tue, May 10, 2011 at 8:28 PM, Martin Lippert<lippert@xxxxxxx>
>>>>>   wrote:
>>>>>>
>>>>>> Hi Alexander!
>>>>>>
>>>>>> Equinox Aspects is not the implementation of the OSGi Weaving Hook
>>>>>> Service
>>>>>> Specification. That specification is implemented as part of Equinox
>>>>>> 3.7,
>>>>>> I
>>>>>> think. So if you use Equinox 3.7, you already have that specification
>>>>>> implementation at hand.
>>>>>>
>>>>>> Equinox Aspects (now called Equinox Weaving) has graduated and ins now
>>>>>> part
>>>>>> of the Equinox SDK, so regularily updated by nightly builds, etc.
>>>>>> Nevertheless there isn't much activity in this area over the past
>>>>>> month,
>>>>>> but
>>>>>> its pretty stable and usable.
>>>>>>
>>>>>> The relationship between these two things are:
>>>>>> - Equinox Weaving was implemented before the weaving spec came out, to
>>>>>> it
>>>>>> is
>>>>>> still based on some Equinox-specific hooks deep inside of Equinox and
>>>>>> is
>>>>>> currently build mostly to weave aspects into OSGi bundles.
>>>>>> - The OSGi Weaving spec was done to allow all kinds of bytecode weaving
>>>>>> on a
>>>>>> general level, so you should be able to re-implement the aspect-weaving
>>>>>> part
>>>>>> of Equinox Weaving on top of the new OSGi spec right now. I haven't
>>>>>> done
>>>>>> that (haven't found the time), but its on my list.
>>>>>>
>>>>>> The web will be updated soon to reflect all this...
>>>>>>
>>>>>> Hope this answers your questions!
>>>>>> -Martin
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>> Could you please clarify for me whether the Equinox Aspects is the
>>>>>>> implementation of OSGi Weaving Hook Service Specification? The thing
>>>>>>> that confuses me is that Equinox Aspects was last update May 1 2009
>>>>>>> (according to it's home page
>>>>>>> http://www.eclipse.org/equinox/incubator/aspects/index.php) and is
>>>>>>> still in incubator state with Milestone 7 as the latest version while
>>>>>>> OSGi Weaving Hook Service Specification (56) was added to the
>>>>>>> specification starting with version 4.3 which was released about a
>>>>>>> month ago.
>>>>>>>
>>>>>>> If the Equinox Aspects really is the implementation of this
>>>>>>> specification then what is the current status of this project, what is
>>>>>>> it's latest version and is it stable?
>>>>>>>
>>>>>>> If the Equinox Aspects is NOT the implementation of the specification
>>>>>>> mentioned then what are the differences between the two and what can
>>>>>>> help me in choosing between them.
>>>>>>>
>>>>>>> Thanks in advance...
>>>>>>> _______________________________________________
>>>>>>> equinox-dev mailing list
>>>>>>> equinox-dev@xxxxxxxxxxx
>>>>>>> https://dev.eclipse.org/mailman/listinfo/equinox-dev
>>>>>>
>>>>>> _______________________________________________
>>>>>> equinox-dev mailing list
>>>>>> equinox-dev@xxxxxxxxxxx
>>>>>> https://dev.eclipse.org/mailman/listinfo/equinox-dev
>>>>>>
>>>>> _______________________________________________
>>>>> equinox-dev mailing list
>>>>> equinox-dev@xxxxxxxxxxx
>>>>> https://dev.eclipse.org/mailman/listinfo/equinox-dev
>>>>
>>>> _______________________________________________
>>>> equinox-dev mailing list
>>>> equinox-dev@xxxxxxxxxxx
>>>> https://dev.eclipse.org/mailman/listinfo/equinox-dev
>>>>
>>> _______________________________________________
>>> equinox-dev mailing list
>>> equinox-dev@xxxxxxxxxxx
>>> https://dev.eclipse.org/mailman/listinfo/equinox-dev
>>
>> _______________________________________________
>> equinox-dev mailing list
>> equinox-dev@xxxxxxxxxxx
>> https://dev.eclipse.org/mailman/listinfo/equinox-dev
>>
> _______________________________________________
> equinox-dev mailing list
> equinox-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/equinox-dev
_______________________________________________
equinox-dev mailing list
equinox-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/equinox-dev