[Rpm-metadata] Debian metadata (was Re: format misc)

[Jeff Licquia]

On Tue, 2003-10-07 at 16:07, Darrin Thompson wrote:
> You'd have to think carefully about how you structure it. This probably 
> wouldn't work:
> 
> <depends>
>    <deb:orentry>
>      <entry name="blah"/>
>      <entry name="second-alternative"/>
>    <deb:orentry>
> </depends>
> 
> If a client who doesn't understand the debian namespace ignores orentry 
> then they either have two entries or none at all. It depends on how you 
> define "ignore".
> 
> Personally I like defining ignore as "ignore the attribute or element 
> and all the element's children."

Very good point.

> Perhaps a better solution would be:
> 
> <depends>
>    <entry name="blah"/>
>    <deb:orentry>
>      <entry name="second-alternative"/>
>    </deb:orentry>
> </depends>

Ick.  I hate imbibing context with semantic meaning.

I'd probably implement it something like this:

<depends>
  <entry name="alt1">
    <deb:altentry name="alt2" flag="GE">
      <version ver="1.0"/>
    </deb:altentry>
    <deb:altentry name="alt3"/>
  </entry>
</depends>

That's still a bit ugly, but it would seem to failsafe under nearly all
the parsing situations I can think of.  If a parser didn't drop subnodes
of unrecognized nodes, then the <version> tag would cause problems, but
that could even be solved with an <altversion> tag in the namespace or
some such.