You are viewing a plain text version of this content. The canonical link for it is here.
Posted to java-dev@axis.apache.org by Aleksander Slominski <as...@cs.indiana.edu> on 2005/05/25 22:57:48 UTC
[Axis2] memory vs performance of AXIOM [Re: [Axis2] Proposal for
adding a method in OMElement API.
Venkat Reddy wrote:
>We can remove the namespace only if it is found in its own hashmap. So
>no problems with parent's stuff. For its children, the namespace is
>gone forever, if they are only storing incremental additions of
>namespaces, but then thats what the user asked for, right?
>
>
my recommendation (memory wise!) - do not use hashmap and optimize for
common case which is *few* namespace declarations ... in future if it
ever comes up as problem there can be some logic to create hashmap if
element has more than N namespaces or use some optimized storage/lookup
structures but i would do it only when profiler shows it to be hot spot ...
alek
>-Venkat
>
>
>On 5/25/05, Eran Chinthaka <ch...@gmail.com> wrote:
>
>
>>Wait wait, we are missing something here.
>>
>>Remember we have to think about the memory foot print as well. AXIOM is all
>>about memory and performance.
>>
>>If we copy parent's namespace stack to children, we are wasting memory.
>>Isn't it. I agree that for us to better implement this remove namespace that
>>is a good solution. But if you think about the common case, this is utter
>>waste of memory.
>>
>><a>
>> <b>
>> <c/>
>> </b>
>></a>
>>
>>Think: each element defines 10 namespaces, in your suggested approach, we
>>will ended up with 60 hashmap elements. (10 from a, 20 from b and 30 from
>>c). But the current implementation will have just 30 entries + just object
>>references to the others.
>>
>>Plus, even by doing this, will not solve the problem. What if you have some
>>attributes with the namespace you gonna remove. We have to change them as
>>well.
>>
>>I agree that removeNamespace should be supported, but I have still no GOOD
>>idea about, how to do it.
>>
>>Comments/Suggestions ?
>>
>>Chinthaka
>>
>>----- Original Message -----
>>From: "jayachandra" <ja...@gmail.com>
>>To: <ax...@ws.apache.org>; "Eran Chinthaka" <ch...@gmail.com>
>>Cc: "Venkat Reddy" <vr...@gmail.com>; "Ajith Ranabahu"
>><aj...@gmail.com>
>>Sent: Wednesday, May 25, 2005 2:51 PM
>>Subject: Re: [Axis2] Proposal for adding a method in OMElement API.
>>
>>
>>Hi Eran et al,
>>Removing a namespace that child is using from a parent is certainly a
>>user fallacy and not able to handle that, I guess, is not any
>>compromise. If a user is following a API, he is expected to follow it
>>in its spirit, we can't help it. If that were a blocking compromise,
>>standardized SAAJ API will never have included the method, isn't it?
>>In my honest opinion, it's good to have this innocuous method.
>>
>>Regards
>>Jaya
>>
>>On 5/25/05, Eran Chinthaka <ch...@gmail.com> wrote:
>>
>>
>>>Hi Venkat and all,
>>>
>>>
>>>
>>>>Okay - the usecase we have is the following method in SAAJ API on
>>>>javax.xml.soap.SOAPElement:
>>>>
>>>>
>>>>boolean removeNamespaceDeclaration(java.lang.String prefix)
>>>>
>>>>
>>>>We can't implement this without the undelying OM design catering to such
>>>>needs.
>>>>
>>>>
>>>>So here goes my +1 for implmenting this method on OMElement.
>>>>
>>>>
>>>>Regarding the uniqueness of prefix, i dont think it is an issue,
>>>>because every element has its own hashmap object for namespaces, and
>>>>for one prefix, there can be only one namespace in the scope, so
>>>>removal such namespace will not affect other elements or other
>>>>namespaces, i guess.
>>>>
>>>>
>>>Every element has a namespace map. But if a child is using a namespace
>>>declared in the parent, that child is pointing to the parent for that. So
>>>if
>>>we remove the namespace from the parent, we have some trouble.
>>>
>>>I agree that u need a method like this for SAAJ impl. But how can we
>>>compromise ?
>>>
>>>
>>>
>>>>When in doubt, lets vote :-)
>>>>
>>>>
>>>Is it a joke ? ;) ;).
>>>
>>>Regards,
>>>- Chinthaka
>>>
>>>
>>>
>>>>- Venkat
>>>>
>>>>
>>>
>>>On 5/25/05, Ajith Ranabahu <aj...@gmail.com> wrote:
>>>
>>>
>>>>Hi all,
>>>>Well since our OMElementImpl keeps its own list of namespaces adding /
>>>>declaring a namespace would affect that Element only. if the reference
>>>>is
>>>>removed from the parent I don't see a problem with the children since
>>>>they
>>>>keep their own reference to the particular namespace. We ask for a
>>>>mandatory
>>>>namespace at creation, rememebr ?
>>>>The only thing that will not be elegant is the serialization.(I don't
>>>>see
>>>>a
>>>>problem though. The serializer handles it right and generates a
>>>>sematically
>>>>valid XML. That generated XML is not elegant)
>>>>So even though a non-used namspace does no harm (except serializing it
>>>>on
>>>>the declared element making the XML ugly) we can techinically remove it
>>>>without a problem.
>>>>
>>>>On 5/25/05, Eran Chinthaka <ch...@gmail.com> wrote:
>>>>
>>>>
>>>>>Hi Jaya and all,
>>>>>
>>>>>I still find difficult to give you a method to remove namespaces.
>>>>>
>>>>>Namespaces, once defined in an element, that will be used by all of
>>>>>its
>>>>>children. So if we try to remove a namespace, that will be a total
>>>>>mess.
>>>>>Isn't it ?
>>>>>
>>>>>So I'm in doubt to give *any* of the methods to remove namespaces
>>>>>.....
>>>>>
>>>>>Regards,
>>>>>Chinthaka
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>--
>>>>Ajith Ranabahu
>>>>
>>>>
>>>
>>>
>>--
>>-- Jaya
>>
>>
>>
>>
>
>
>
--
The best way to predict the future is to invent it - Alan Kay