[XML-SIG] Preventing DLL hell is very expensive

Tim Diggins subscribed@red56.co.uk
Monday, 03 March 2003 11:37:57 +0000

> The problem is that so few OSS projects have such manpower.  
> Of course, I 
> don't know what Microsoft's excuse is ;-).  Anyway, I believe 
> that you can 
> only ever get 2 of the following:
> 1) A package in active development
> 2) Freedom from DLL hell
> 3) Nice modularity for integration with other subsystems 
> (i.e. APIs of broadly 
> useful granularity)
> Unless you have IBM-type bucks.  Pick two, if you're lucky.  
> You're not going 
> to get all 3.  I think PyXML offers 1 & 3.  

As a PyXML user, I found this a very useful observation, giving a lot of insight on the tradeoffs. While up to now I've been leaning towards a change top-level namespace, maybe the issues with _xmlplus that have been discussed would be solved by some explicit explanation of the "special relationship" between the external PyXML package and the xml core package. This documentation addition should be in 2 places:

1) the install documentation with PyXml (to solve the kind of misunderstandings and hard to diagnose errors I've had about the problems I've had as a user "moving" code from xml core to PyXML - as Martijn has laid out) maybe this could point out where PyXML is installed and why (apologies if this is already in the documentation and I never read it carefully enough) and also some of these "typical errors" that people have 

2) the xml core documentation with Python (to point out the relationship with PyXML which is special enough to warrant upfront advice).

If xml core/PyXML users were aware of the relationship then they are more likely to be able to diagnose problems and report them correctly.

(Also, having never contributed to an active OSS project before, but used lots of them, and then suggesting some work to do, I feel a responsibility to volunteer to write some of these documentation additions myself - but I'm not sure if that's appropriate - maybe someone can contact me offlist ito give me a suggestion of how to start if it's appropriate).

Side thought:
I presume there's no (performance-efficient) way to have both packages (xml core and PyXML) installed side by side (e.g. pyxml.dom... and xmlcore.dom...) and then have the toplevel xml.dom... be an "alias" to those real packages (a bit like os.path). If this WAS possible, then this would be seem like a solution... (it would certainly make tracing problems in upgrading from core->pyXML easier)



> -----Original Message-----
> From: xml-sig-admin@python.org 
> [mailto:xml-sig-admin@python.org] On Behalf Of Uche Ogbuji
> Sent: 03 March 2003 05:53
> To: Brian Quinlan; xml-sig@python.org
> Subject: [XML-SIG] Preventing DLL hell is very expensive