Hi, Bhakti.
Yes, I am aware of the current arrangement. That's why I asked whether
it would make sense to split the classes up along client/server lines.
I discussed this briefly with Harold some time ago and his feeling then
was that there would be little savings because so many of the classes
would be needed in the client partition anyway.
I just wanted to revisit this in case there had been any changes that
would make such a division more worthwhile.
So app clients that use web services and therefore need the Metro
classes will still need to rely on the already-resident version in the
JRE or put up with the download of some more recent build that's part of
GlassFish, right? In the second case, we can encounter the situation
again that we had early in Java SE 6 in which the Metro that was part of
the JRE (2.0 at that time) was incompatible with what GlassFish checked
for (2.1). Java Web Start launches of app clients that used web
services failed due to the way Java Web Start manages the class loading
- the JRE classes were loaded before the GlassFish logic could intercept
the requests and provide the ones bundled with GlassFish.
Thanks.
- Tim
Bhakti Mehta wrote:
> Tim,
> FYI We do not separate client runtime from server runtime. If there is
> an appclient built with metro it will need webservices-rt.jar in
> classpath at runtime.
>
> Regards,
> Bhakti
>
> Timothy Quinn wrote:
>> My (predictable) contribution to this topic is to ask whether it
>> makes sense to have a JAR (or set of JARs) that supports what clients
>> need separate from what servers need. This comes from the continuing
>> effort we all need to share to keep the footprint for app clients as
>> small as possible. In the past I've heard that there is not much
>> possibility for improvement in this, but I will continue to raise the
>> issue at every chance! Especially as long as GlassFish needs a
>> different version of such JARs vs. what is in the JRE the size of
>> these JARs will remain an issue for app clients.
>>
>> - Tim
>>
>> Kohsuke Kawaguchi wrote:
>>>
>>> Putting my Metro hat on here,
>>>
>>> Really *the* motivation for Metro to deliver things as one big blob
>>> is so that we can release one set of binaries as stand-alone and
>>> then deliver the same bits to v2 and v3.
>>>
>>> Metro needs to keep delivering into v2 and v3 for some time to come,
>>> and we do stand-alone releases for foreseeable future. If we can
>>> deliver our bits in the exact same fashion into two versions of GF,
>>> that would considerably simplify our build qualification and testing
>>> effort. And this is one area where we are really stretched thin.
>>>
>>> This also has an added benefits to users, as they can take the
>>> stand-alone bits and upgrade GF by simply copying the jars.
>>> (Although this is a secondary issue for v3, assuming it has its own
>>> update channel to deliver new bits.)
>>>
>>> You are right that there's a possibility that EE6 defines profiles
>>> that only require a certain portion of Metro but not the others, but
>>> so far that hasn't happened yet, and I thought it won't be too late
>>> to revisit the packaging when/if that happens.
>>>
>>> There's also a small issue that nucleus already depends on a StAX
>>> implementation, which is today delivered as a part of Metro, but we
>>> felt that that was small enough duplication and the benefit of
>>> single bits delivery outweighs the problem.
>>>
>>> So all in all, what I'm suggesting is to start the Metro integration
>>> into v3 according to Bhakti's suggestion, with the understanding
>>> that we should be open to the possibility of revisiting this.
>>>
>>>
>>> Bhakti Mehta wrote:
>>>> Jerome,
>>>> Please read inline
>>>> Jerome Dochez wrote:
>>>>> my initial feeling is that it is too coarse grained.
>>>> I tend to agree but if different profiles start plugging in their
>>>> own versions of the subparts wouldnt that be a nightmare to debug?
>>>> Hence I brought up this question here should metro be a module or
>>>> should there be dependencies for individual technologies in metro?
>>>>
>>>> Regards,
>>>> Bhakti
>>>>>
>>>>> For instance, for people who don't care about java ee
>>>>> compatibility and would want to only support say JAXB+JAXWS (which
>>>>> I assume would be doable), how would that work ?
>>>>>
>>>>> also in Java EE 6, different profiles may require different
>>>>> subpart of the metro stack, wouldn't that make things harder to
>>>>> decouple...
>>>>>
>>>>> Jerome
>>>>>
>>>>> On Feb 28, 2008, at 3:18 PM, Bhakti Mehta wrote:
>>>>>
>>>>>> Hi,
>>>>>> I wanted to check if there are individual modules in v3
>>>>>> depending on various technologies from the metro stack like jaxb,
>>>>>> jaxws, sjsxp, stax-ex,saaj?
>>>>>>
>>>>>> Since all these technologies are bundled in
>>>>>> webservices-rt/webservices-tools.jar and the 1.1.2-SNAPSHOT
>>>>>> version of metro is already on maven repository
>>>>>> Imaybe we should just refer to webservices-rt,webservices-api and
>>>>>> webservices-tools in our dependencies
>>>>>>
>>>>>> The only case maybe where Kohsuke suggested sjsxp (people just
>>>>>> wanting to use the parser may not want to get the whole metro stack)
>>>>>> In that case we can let the dependency for sjsxp remain in the
>>>>>> poms but for other technologies I can just update the poms to
>>>>>> use metro.
>>>>>>
>>>>>> Do other modules see issues with that?
>>>>>>
>>>>>> Thanks,
>>>>>> Bhakti
>>>
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe_at_glassfish.dev.java.net
>> For additional commands, e-mail: dev-help_at_glassfish.dev.java.net
>>
>>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe_at_glassfish.dev.java.net
> For additional commands, e-mail: dev-help_at_glassfish.dev.java.net
>