Wednesday, March 11, 2009, 8:17:03 PM, Hussein Shafie wrote:
> Daniel Dekany wrote:
>>> The problem seems specific to DocBook 5, and more precisely to
>>> indentation when the document being saved conforms to a RELAX NG Schema.
>>> If you turn indentation off, you should be able to save your document in
>>> no time (less than 50ms).
>>>
>>> This is clearly a problem, but we currently don't see how to improve the
>>> performance[*] of XXE in this case and we currently do not plan to save
>>> documents in a background thread.
>>
>> I don't quite get what the problem is. Is it that in the most generic
>> case, an element of a given name can have totally different meaning
>> ("type") depending on what context it occurs?
>
> Yes.
>
>> But for most practical
>> XML schemas you had to indent all elements with the similar name on
>> the same way, regardless of where it occurs, so just looking up the
>> element name in a hash would be sufficient. (OK, in some mean schemas,
>> making a second decision based on the parent elements is needed, but
>> that's rare anyway, so it shouldn't affect the performance much.) Doesn't
>> this stand for DocBook 5?
>>
>
> XXE is not a DocBook editor. It has no intrinsic knowledge of DocBook 4
> or DocBook 5.
Yeah, but if 50% of your users use XXE for that, it better has *some*
intrinsic knowledge. Some cheating here and there... :)
> Everything it does depends on the schema of the document being
> edited.
>
> In the case of RELAX NG (DocBook 5), the content type of an element
> having a given name may depend on quite complex conditions. What follows
> is just a very simple example of what you can do with RELAX NG:
>
> <div class="section">
> <div class="title">Title of the section</div>
> <div class="paragraph">Blah, blah, blah.</div>
> <div class="paragraph">Blah, blah, blah.</div>
> <div class="paragraph">Blah, blah, blah.</div>
> </div>
>
> Indeed, DocBook 5 does not use the above scheme, but it uses other
> advanced RELAX NG features. However, whether a specific RELAX NG schema
> uses advanced features or not has no effect on XXE which always needs to
> cope with RELAX NG in its generality.
(Obviously, in principle, by analyzing the schema it can be told if an
element of a given name always occurs with the same xml:space
attribute value, and then for all elements with that name it's very
fast to decide how to indent. Even a pessimistic guess for that could
be good enough... But I have no clue how hard/easy it is to do that.)
> More information in "XMLmind XML
> Editor - Support of RELAX NG Schemas"
> http://www.xmlmind.com/xmleditor/_distrib/doc/rngsupport/index.html
>
> This being said, as always, we'll try to improve the situation.
--
Best regards,
Daniel Dekany
