forrest-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Crossley <cross...@apache.org>
Subject skinconf versioning (Was: svn commit: rev 10172)
Date Fri, 23 Apr 2004 05:41:19 GMT
Nicola Ken Barozzi wrote:
> The fact is that somehow I was still getting errors in the logs
> about a missing DTD, so I got irritated and took them away.
> Ok, so this is not a technical reason,

Can't you just ignore the "error" until we get time to fix it. :-)

> ...but the more I think of it, the more it seems to make 
> sense. Do property files have a DTD? Do they need one?

Well i think that they do. I have never been able to understand
how haphazard configuration files can ever support compatibility
between versions.

User: Hey, my so-and-so is busted!
Support: Er, what version of the skinconf file do you have?
User: Shrug, dunno.
Support: You need to add the <so-and-so> tag.
User: What is that, how do i use it?
Support: Oh my god, not again.

> For example, let's say that somebody makes a skin that needs some extra 
> customization values... where do they put them? Without a DTD, they can 
> easily add them to the skinconf file.

The skinconf-*.rng could me more lenient, allowing them to
add extra optional stuff to their own skinconf.xml file.

> What are the reasons we should keep the DTD, given the above point and 
> the problems we have been having till now (not no mention having to make 
> users update the DTD at every change)?

One reason that i have heard, is that having a DTD declared
for the skinconf.xml file enables people to use their favourite
XML editor.

I think that we are getting into this bind because we do not
release Forrest often enough. The contracts with users via
releases are too lax.

Dave Brondsema wrote:
> The only users that have to update their DTDs frequently are people keeping up
> with SVN.  And in that case the DTD will make sure they keep up to date with any
> changes.

As users, we all need to keep our skinconf.xml files up-to-date
anyway, so with or without a DTD, user-level changes need to be made. 
 
> I'm ok without a DTD so long as we replace it with a (web) document that lists
> and explains each element & attribute.  It should be up-to-date all the time for
> both the latest release and current SVN.
> 
> And a [headsup] email to forrest-dev for any change to the skinconf that isn't
> backwards compatable. 

It will be impossible to manually document the many different
versions of the evolving skinconf structure.

If we had a series of well-defined schema, then we have tools
to automatically generate web-page documentation from them.

I think that it is very important to have well-defined structure
for this file and to validate it before proceeding with the build.
If we don't control it, then i think we would descend into chaos.

Would this procedure be okay ...

* Use a versioned namespace in the skinconf.xml

* Do not declare a DTD in the skinconf.xml

* Define a RELAX NG grammar with separate filename for
each substantial change.

* Document each change in status.xml

* Release Forrest more often.

* Generate a DTD from each RNG. Generate html doco too.

* Users can insert a DTD declaration into their skinconf.xml
if they want to (or configure their xml editor in some other way).

* The Forrest build system strips any DTD declaration from the
file before using it, so that the parser does not trip up on it.
Is this really needed?

* The Forrest build system uses the relaxng/schematron to
validate the skinconf.xml - Jing can evidently respond to the
namespace to apply the correct schema.

* Add some build targets to automate skinconf upgrade.

* Release Forrest more often.

--David




Mime
View raw message