Re: [xml] "atoms" for name/attribute strings

[Daniel Veillard <veillard redhat com>]

On Sun, May 18, 2003 at 07:53:14PM -0400, Chris Ryland wrote:
> Forgive me for throwing out wild ideas that might have large 
> consequences for libxml2 (whose distilled wisdom I greatly respect from 
> my brief acquaintance so far), but I had a wild idea that might greatly 
> speed up tree processing after parsing.
>
> If all name (and namespace) and attribute (xmlChar *) strings were 
> "atomized" (like Lisp atoms or Python strings--hashed to unique 
> strings) by the parser, then all input tree processing (after atomizing 
> all "interesting" name/attribute strings) would be reduced to comparing 
> string pointers, not string values.

  The main effect would be memory consumption and reduction of the number
of malloc()/free() needed for parsing if done correctly.
  The main drawback is that libxml2 trees are mutables so there is some
thinking/extra work required when processing to add nodes to a tree and
when freeing, it's not trivial but doable, I have been thinking about
this for a while.

> This could be a parser configuration setting (like entity expansion) 
> that would only affect clients that requested it, and wouldn't 
> necessarily affect a whole lot of the code. (In fact, it might be 
> completely localizable in the form of an allocation function called for 
> such names.)

  Hum, there is far more than that needed.

> Is this is a good idea, looked at from any of you old grizzled 
> veterans' points of view?

  It's a good idea but the devil is in the details, in order to not
break the large amount of existing code.

Daniel

-- 
Daniel Veillard      | Red Hat Network https://rhn.redhat.com/
veillard redhat com  | libxml GNOME XML XSLT toolkit  http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine http://rpmfind.net/