Re: [Qexo-general] Re: TreeList implementation

[Per Bothner]

Joseph Coffland wrote:

> The more I've looked around at Qexo the more I've realized that providing
> an alternative to TreeList is not practical.  Much of the code depends on
> AbstractSequence.

Yes.  Perhaps in a year I'll want to consider improving the data
structures, but for now a complete implementation of XQuery is
the main priority, and TreeList/NodeTree are at least acceptable.

> BTW using a string table could improve TreeLists storage
> requirements even further.

You mean for element/attribute names, or for text nodes?
ParsedXMLToConsumer does use a string table for element/attribute names.
Using a string table for attribute value or text nodes would be an
advantage in some applications; I don't know if it would make up
for the extra overhead of a string table.  It might be useful to
make it an option - just as it might be useful to provide an option
for creating indexes.

> It seems that you could get fragmentation in the TreeList object array if 
> deleting were possible.  Then you might want maintain a freelist.
>
> In a lot of ways memory management is much easier in C/C++.

You don't have garbage collection, but C++ destructors is a
very powerful tool, and you don't have to heap-allocate everything.
--
        --Per Bothner
address@hidden   http://per.bothner.com/