Skip to content


Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP


"Moving" a subtree to a new document... #32

psikore opened this Issue · 6 comments

3 participants


I am trying to convert a large codebase from using tinyxml to tinyxml2.

In the original code it moves a subtree from one document to another via this way:


but with tinyxml2 it 'controls' the memory model so LinkRemoveChild() no longer exists.

Do you have any suggestions on how to 'move/transfer' a subtree of nodes from one document to another efficiently? ie. Least amount of code futz and smallest amount of memory allocation?

Do I need to use the Visitor pattern now somehow?


This is the drawback of the memory managed approach, definitely.

Check out "ShallowClone". It copies a node, but not the nodes children, and is the main tool for the job. (There is example code in xmltest.cpp). For copying a full document, you'll need a recursive walk, and visit() seems like a reasonable choice although i haven't tried it.

Implementing "DeepClone" in the API is the right way, but I haven't implemented it yet.


From my investigation it seems like the major memory problem in Tiny1 arises from the constant string reserve calls as xml is parsed into TiXmlString. It seems like pooling the strings in a string memory allocator would cleanup the allocations and still let the nodes be dynamically allocated? As it is I'm finding it tough to work out how to use Tiny2 when the code I am porting "picks" sub-trees out of an existing doc to create a new doc.


Yeah, it's involved. The strings should be put in at least a per-document string pool. That's not required for a recursive copy, but certainly is in the spirit of things. But we could start with a recursive copy first and then go from there.


For what its worth I modified TinyXML1 locally to support allocator 'hooks' such as:
TIXML_EXTERNFUNC( void , AllocHook )( size_t in_size );
then modified the tinystr.h function to call these hooks instead of directly allocating int arrays - ie...
size_t bytesToAllocate = intsNeeded * sizeof(int);
rep_ = reinterpret_cast<Rep
instead of:
rep_ = reinterpret_cast( new int[ intsNeeded ] );
then I fed these allocations into a very fast fixed storage allocator.

In my tests this speeds up XML parsing by 3.5 times (I loaded 40 odd big xml files previously in 50seconds and now in 18 seconds).

If the lib can support the user settting up an external allocator as per above that will be helpful to others I bet... although its easy to modify since all the code is there. ;).


Yeah, the main performance gain with TinyXML-2 is the memory management. Putting a better system into TinyXML-1 I'm sure helps there as well. But TinyXML-2 doesn't use statics or globals, which also helps with the threading model. I wanted to be sure that multiple XMLDocuments could be used simultaneously on multiple threads without locks. But to be clear: a given TinyXML-2 XMLDocument can only be used on a single thread.


The deepCloning function would be greatly appreciated !
Does anyone have already a working visitor sample to do the same job ?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.