Re: Re;: NSXMLParser on GS (was Re: Books / GNUStep / XML)

[Dr. H. Nikolaus Schaller]

Am 18.10.2004 um 17:22 schrieb Adrian Robert:
> On Oct 18, 2004, at 3:49 AM, Dr. Nikolaus Schaller wrote:
>
> > Dear all,
> > we have a first (90%) implementation of NSXMLParser (and
>
> I like the sound of this, but it also must be said that implementing a 
> fully standards-compliant XML parser with good performance is not a 
> minor task.  The

Yes, performance is good by the architecture we have chosen - but 
mainly depends on the speed of generating NSStrings from UTF8 character 
sequences.

> use of libxml2 has the benefit that it is a mature implementation that 
> receives lots of testing and dots/crosses all or most of the i's and 
> t's in the specification.
>
> I appreciate that the focus of myStep is different from GNUstep.  Is 
> the goal of your project's NSXMLParser code to provide an XML 
> implementation to this level, or is it intended more to be used in 
> special cases, such as archive I/O, where the XML formats and needed 
> behaviors are a controlled subset of the full range of possibilities?

Currently it does not read and process any !DOCTYPE or !ENTITY. And 
handling of errors is not foolproof. But scanning&parsing tags with 
attributes, parsing values between tags and handling the 5 standard 
entities (e.g. &) works. This is sufficient to properly parse Cocoa 
Property Lists. The goal is to be more and more compatible to 
NSXMLParser - but I don't know if that class is a full XML parser that 
can handle all special cases.

I have e.g. not yet found out if it really loads a remote !DOCTYPE to 
handle additional entity definitions - and what happens if there is no 
network connection available. And my experiments have shown that it 
does not handle lineNumber and columnNumber properly. Finally, I think 
the NSError returns of NSXMLParser are not very helpful...

> If the latter, does your design make it possible to implement some 
> type of runtime switching behavior so that libxml2 could be used 
> (behind the same API) instead of the native implementation if needed?  
> Runtime parser selection is the norm in the Java world, and while it 
> can cause inconvenience at times, it often comes in handy as well.

I think GNUstep should provide both: NSXMLParser being compatible to 
Cocoa - and a real libxml2 wrapper that can be conditionally compiled 
in. So users can decide if they want to be platform independent (i.e. 
compile their apps with Cocoa) and have restricted XML parsing or to 
use some GSXMPParser (that is not readily available on the Mac).

Nikolaus