Re: [gentoo-dev] Antwort: Re: [gentoo-dev] YaST will be GPL [Virus checked]

[John Nilsson]

On Sun, 2004-03-21 at 20:21, Stuart Herbert wrote:
> It'll never happen.  XML isn't something that everyone accepts is always a
> good thing.  In many languages, handling XML content programmatically is a
> RightRoyalPain(tm).  And we currently lack adequate XML-aware diff and grep
> tools - two things essential to making this work.

So why would you handle XML? Use a library or other tool to parse the
file. Part of this project would be to develop the tools needed to get
and set configuration in a simple way.

> So how would the config tool know how to generate the XML file?  A schema is
> nowhere near enough.

Are you sure? 

> > This schema would also contain the documentation for the configuration
> > file. From this Schema a man-page could be generated. Thus config tools
> > and man-pages would have the same source of information.
>
> A bit like Perl and their POD files?

I don't know Perl... i gather POS is to Perl as reStructuredText is to
Python.
The idea is the same, but it would be silly to not use a xml syntax for
docs too.

> XML is a reasonable choice for marking up data such as configuration files.
> Personally, I prefer to use it when you're feeding structured data from one
> system to another - because of the ability to validate the input.  I'd rather
> use a relational database technology than XML in static files on disk,
> because it's much more searchable and shareable.  But that's just me.

If you mean that configuration should be saved in binary format, I don't
agree. I believe that a configuration file must be readable/editable in
the most basic format still humanly comprehensible, and backwards
compatible. This means ASCII (Or equivalent subset of any compatible
charmap).
In any way, configurations is inherently more hierarchal then relational
in nature. Just take a look at the evolution of the configuration files
for some major apps. Most enhancements seems to be about the  hierarchal
structure, hint: XF86Config.

> How many packages are there in the Gentoo tree?  Over six thousand.  Let's say
> you could convert each one in an average of one man day's effort.  I think
> that's optimistic, but it keeps it simple.  Now, a man year is around 200
> days of effort, allowing for weekends, holidays, illness, and overheads.  So
> that's 30 man years of effort to convert what we have today.
>
> If you had RedHat's money, maybe it could be done.  But Gentoo people are
> volunteers.  On average, volunteers on activities outside normal work and
> family commitments manage a man month's worth of effort over 12 months of
> elapsed time.  So you're really looking at 360 years to get it done by
> volunteers.

Did I mention dream? ;-) But just to adjust your math, forking also
means merging packages providing similar services so it's not quite 6k
packages...
Gentoo is about choice I hear people screaming. I believe some choices
should be made by developers and some by users. Making a clear
separation of whom should make which might even give booth more choice
with fewer packages.

> How are you going to get the authors of 6,000+ packages to agree to follow a
> single XML schema?

Flexibility, ease of use, and quicker market adoption of the package.
Less newbie support?

> I can't think of a way in which you could prevent duplication of validation
> logic except one - the validation logic would have to be written in some form
> of scripting language such as JavaScript, and be part of the XML schema
> itself.  XML Schema itself is insufficient, and the moment your shared
> library of code starts needing callbacks (think of the getopt function as an
> example of what I mean) then you end up with duplicated logic.

Am I mistaken in that you can express any constraint in an XSD?

ex. from http://www.w3.org/TR/xmlschema-2/ Section: "4.3.4 pattern"
<simpleType name='better-us-zipcode'>
  <restriction base='string'>
    <pattern value='[0-9]{5}(-[0-9]{4})?'/>
  </restriction>
</simpleType>

Are there any reasons for the validation to be part of the application
logic?
The whole point in using XSD is that the XSD IS the validation logic and
that the validation engine is completely separate from the application.

> I think it would help if you had a clearly defined problem to direct your
> research at, and a wide background study of how things are done today and the
> myriad of reasons why.
>
> Best regards,
> Stu

I'm in the process of implement a kind of laboratory for this kind of
research. A kind of crossbreed between a wiki and phpBB, a RFC factory
if you will. The XML/Configuration matter is the perfect test case and
will be initialized as soon as I have something ready.

Even though I feel you are a tad negative, I value your input.

-John