Re: [xml] libxml memory consumption

[Gavin Sherry <swm linuxworld com au>]

On Mon, 6 Aug 2001, Daniel Veillard wrote:

> On Mon, Aug 06, 2001 at 11:42:20PM +1000, Gavin Sherry wrote:
> > I have not noticed any reference in the documentation nor (from the skim I
> > gave it) anything in the source which suggests that libxml can deal with
> > just localised parts of the XML document. That is, to take a basic
> > example, if I have a document such as:
> >
> > <doc>
> >    <info>
> >     <data>123</data>
> >     <data>456></data>
> >    </info>
> >    <otherinfo>
> >     <data>abc</data>
> >     ...
> >    </otherinfo>
> > </doc>
> >
> > and the size of the tree under 'otherinfo' is 200 MB but I want only to
> > look at the tree under 'info' then I am using a very large amount of
> > memory to do it. Has anyone considered building a 'contextually' (for lack
> > of another word) sensitive read ahead system into libxml which could read
> > only the section of the tree you are interested in into memory or is this
> > also featured in some part of the library I am yet to explore?
>
>   Not provided directly, but you can probably build it using mostly existing
> code. The tree is built by the default SAX handlers in SAX.c.
>   If you override them and call the default ones only for content within
> the parts you are interested you can probably build what you need 
> relatively quickly. The ctxt->nameXXX provides the current stack
> context of the parser, you should be able to build your selection on this.
>   This will require a bit of custom glue code, though.

Would there be any value in integrating this into the main source tree and
building it into the user API?

Gavin