Yes, Samisa, I wrote a shallow copy that used the ref counted attributes, while creating new copies of element and text nodes as these are not ref-counted today. As I wrote it in the context of a C++ application, I need to do some cleanup to make it suitable to be included in the axiom environment. I'll try to work something up that people can comment on.
When I thought about it some more, I realized that nothing prevents it from being a complete copy routine, avoiding the ref-counts on the attributes. That just involves a little more execution overhead. Of course, such a copy would still share namespace structures with the original tree. Another interesting thought I had is that even a serialize/deserialize clone implementation has the same difficulty that Kasun raised with detach. If the namespaces were declared at a higher level than the subtree being serialized, then deserialization would fail as the xml structure would refer to namespace prefixes that were not declared. I think that serialization would need essentially the same logic I wrote recently to handle the namespace references in the subtree for detach. Bill -----Original Message----- From: Samisa Abeysinghe [mailto:[EMAIL PROTECTED] Sent: Sunday, February 17, 2008 8:07 PM To: Apache AXIS C Developers List Subject: Re: Issue in using 'detach' for cloning Hi Bill, Kasun and others, Thank you all for taking time to explain this problem in detail. Looking at the ADB use case that Bill has portrayed, I am sure other users too would be interested in such a case where they can 'get' form one ADB type and 'set' on another ADB type. Lets create a branch for this and try to solve the problem. Bill, have you already done some work with respect to ref-count? If so we can start form that. I can think of the following cases that we can try to fix/implement: 1. Implement the ref-cout shallow copy mechanism proposed by Bill 2. Try and implement a deep copy version of detach (so that we can detach and attach) 3. Implement a serailize/deserialize based deep copy I can look into 2 and 3. Thanks, Samisa... [EMAIL PROTECTED] wrote: > Samisa, I can't speak directly to Kasun's use case, but I can describe what > prompted me to need a tree copy and/or clone method. The generated adb stubs > return an object of wsdl type any to the app as a node tree, detached from > the response message. They also accept an object of wsdl type any as a node > tree, to be attached to the request message. In each case, the interface has > ownership of the object, and frees the object when the interface object is > freed. So the get accessor method to the object returns a pointer to the > object tree but does not transfer ownership. The set accessor method accepts > a pointer to the object tree and takes ownership of it. > > You are correct that cloning alone would be used in a routing app. But > imagine that the app receives an object from the service, modifies it, and > sends the app back to the service. To be able to free the response interface > object after processing the response, the app needs the ability to copy or > clone the object to be able to late use it. After modifying the object, the > app would then pass the object back to the service in a later request. If > the app needs the ability to pass the object to the service while retaining a > copy of it for later processing, it needs a copy or clone operation again as > the service will discard the version that was sent when the request interface > object was freed. With a copy or a clone one can achieve the same result in > different ways; a copy that allows one to configure what is copied lets the > code add what it needs to a reduced tree, whereas a clone allows one to later > remove or modify what is unneeded from the cloned tree. > > Were the generated adb interface different, the requirement for a copy would > be reduced. Say the adb interface provided an accessor method, get, that > transferred ownership of the tree to the client app, and that the set method > accepted a pointer to the object to be passed without freeing the tree when > the request interface object was freed. Then these copies would be > eliminated, but the free of the tree after sending a request would now be the > responsibility of the client app. One would need a copy method only if the > original object was a template for generating multiple objects. > > This may be very application specific, and I don't know how general the need > is. I do know that I encountered it, and it sounds like Kasun has a very > similar requirement. When I encountered it and found no suitable method > present, I wrote one that met my needs. But this requirement may be common > enough to warrant inclusion. > > I'm pretty sure you are correct, there is little need for this inside the > SOAP engine. The need is driven more by the interface to axiom, at least > with the adb classes. Kasun's issue with detach losing namespace references > and the tree becoming invalid arises, of course, in the context of the adb > classes detaching the object from the SOAP response message, and so can arise > for anyone using the service/client interface even without the adb classes. > > Whether addressing this involves a significant architecture change depends on > the implementation. One can implement a copy method with options for what is > copied using the current interfaces, it just requires a bit of research into > the intricacies of axiom_node and its friends. A clone by way of serializing > and deserializing the structure I think uses builder interfaces that are not > public and would need to be included, but does not involve an architecture > change. Certainly a copy-on-write clone algorithm does involve some > pervasive changes, but such an algorithm has other problems as well, already > discussed. > > Bill > > -----Original Message----- > From: Samisa Abeysinghe [mailto:[EMAIL PROTECTED] > Sent: Saturday, February 16, 2008 10:33 PM > To: Apache AXIS C Developers List > Subject: Re: Issue in using 'detach' for cloning > > Kasun Indrasiri wrote: > >> Hi all, >> >> I think, through out this thread there are various possible >> implementations were discussed with their pros and cons. I agreed with >> Senaka that the clone method should not be the 'best' or 'fast' one. >> And also Bill's ref count based implementation would suit where we may >> have to explicitly worry about the performance (and lazy coping also >> got its own pros and cons.) >> So, as Bill suggested earlier we can have several clone methods where >> user can select the method according to his preference. We can >> basically have one method for Shallow copy with ref counts and another >> with the normal serialize/desterilize approach. (and the lazy coping >> approach is also possible). Anyway it's a good move to implement these >> stuff with out affecting the existing axiom. >> > > IMHO, cloning is not a use case that we really need for a SOAP engine. > If you are trying to use the same XML over and over again, passing it > here and there, without modification, you are really routing stuff and > not doing that many business logic processing. > I am still trying to figure out why would one want to detach and attach > a node in a module or a service. If the use case is less than 10%, I do > not think it is a good idea to major changes to the current AXIOM > implementation that works. > > Samisa... > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
