[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: question about placements in /usr/share/xml



Mark forgot to send his reply to the list, so I am quoting it in its
entirety here. It definitely sets my mind at ease. :-)

On Feb  2, Mark Johnson (mrj@debian.org) wrote:
 > On Mon, Feb 02, 2004 at 06:49:04AM -0500, Neil Roeth wrote:
 > > On Feb  1, Adam Di Carlo (adam@onshored.com) wrote:
 > >  > Mark Johnson <mrj@ibiblio.org> writes:
 > >  > 
 > >  > > On Sat, Jan 31, 2004 at 06:07:40PM -0600, Ardo van Rangelrooij wrote:
 > >  > >> So, the root SGML catalog file only knows about package
 > >  > >> SGML catalog files that live in /etc/sgml which on their turn only
 > >  > >> know about local SGML catalogs living under /usr/share/sgml.
 > >  > >
 > >  > > I, again, don't understand the need to put SGML catalogs under
 > >  > > /usr/share/sgml. There's nothing inconsistent about doing it this way.
 > >  > [...]
 > >  > > There's no way that I'll ever agree with what you're proposing. Someone
 > >  > > else needs to weigh in, so we can get some sort of consensus.
 > >  > 
 > >  > I'm with Mark here.  If I have XML data going in
 > >  > /usr/share/xml/docbook/4.0/ for instance, obviously I'll have my XML
 > >  > catalog, /usr/share/xml/docbook/4.0/catalog.xml.  Doesn't it also make
 > >  > sense to have my SGML catalog, /usr/share/xml/docbook/4.0/catalog ?
 > >  > 
 > >  > Seems to make sense to me, as a user or as a package maintainer.
 > > 
 > > Not to me, in either role.  The FHS says:
 > > 
 > > "/usr/share/sgml contains architecture-independent files used by SGML
 > > applications, such as ordinary catalogs (not the centralized ones, see
 > > /etc/sgml), DTDs, entities, or style sheets."
 > > 
 > > I infer from this and the corresponding XML section that if /usr/share/sgml
 > > and /usr/share/xml both exist, then I would find all the files used by SGML
 > > applications in /usr/share/sgml and all the files used by XML applications in
 > > /usr/share/xml.  I would expect all file references in the SGML files to be
 > > into the /usr/share/sgml hierarchy, or into a package specific subdirectory,
 > > but not into the /usr/share/xml hierarchy.
 > 
 > This is indeed the case. No SGML resources (i.e. DTDs, ...) would get installed 
 > /usr/share/xml. 

Good.

 > > If I installed nothing but SGML packages, I would be surprised if
 > > /usr/share/xml even existed, and even more surprised if it contained the bulk
 > > of the information for my SGML system, which sounds like the direction this is
 > > heading. 
 > 
 > Nope, I think you've got the situation reversed. XML resources would get installed
 > under /usr/share/xml, as would their SGML catalogs, when appropriate. OTOH, DocBook 
 > SGML would not install a single file under /usr/share/xml. All of its stuff would
 > go under /usr/share/sgml. There is no XML Catalog for DocBook SGML, so there's no 
 > reason to install any of its files under /usr/share/xml.

This was not clear to me from the prior messages, I thought for some reason
SGML catalogs for SGML resources were ending up under /usr/share/xml.  Thanks
for the clarification.

 > > Also, in the proposed Debian XML/SGML policy (perhaps an obsolete
 > > version) it says that the xml-core package will create the /usr/share/xml
 > > infrastructure, so in your example, where the DocBook 4.0 SGML catalog is in
 > > the /usr/share/xml hierarchy, the docbook package would have to depend on
 > > xml-core as well as sgml-base, which seems totally nonsensical. 
 > 
 > Not true. DocBook SGML would have no dependency on xml-core, as it cannot use the XML 
 > Catalog system.
 > 
 > > It would make
 > > more sense if the docbook-xml package depended only on xml-core and the
 > > docbook package depended only on sgml-base.
 > 
 > I disagree. Some XML stuff can indeed use the SGML catalog system, and some XML stuff 
 > definitely can't (e.g. docbook-xsl). So, yes, docbook-xml would indeed depend on both
 > xml-core and sgml-base, which IMO seems perfectly natural. OTOH, docbook-xsl would 
 > depend only on xml-core. There's really no inconsistency.

I'm convinced.

 > > I thought the problem was that our current setup has only /usr/share/sgml into
 > > which all XML apps are shoehorned, and that the solution involved creating
 > > /usr/share/xml so that everything would make sense from an XML perspective.
 > > ISTM that the current solution is heading in the direction of doing that, at
 > > the expense of making things nonsensical from an SGML perspective.
 >  
 > The motivation for /usr/share/xml is that there is a growing number of xml and 
 > related resources that simply don't qualify as SGML. For example, as soon as you 
 > add an 'xml:base' attribute to an XML element, it's no longer SGML. SGML is more 
 > general in some ways, XML (and related specs) are more general in others. The 
 > important point being that XML will continue to grow in both resources and in the
 > number of related specifications. This latter point is the primary reason for 
 > /usr/share/xml.

That what was I thought the motivation was, until I read (perhaps misread)
this thread.

 > >  > Symlinks should only be provided to support *legacy* locations --
 > >  > e.g., the file used to be in /usr/share/sgml/docbook/4.0, but now it's
 > >  > /usr/share/sgml/docbook/4.0 .
 > >  > 
 > >  > Lets not overcomplicate matters please.  I don't see any reason why
 > >  > SGML catalogs cannot reside in /usr/share/xml/....  Remember the
 > >  > catalog is just the registration of some content (a DTD or entity
 > >  > file).  That is to say, its metadata.  Whether the *content* itself is
 > >  > SGML or XML should determine whether it goes in /usr/share/sgml or
 > >  > /usr/share/xml .
 > > 
 > > I think anyone who tried to understand the SGML structure that had some mix of
 > > references to both /usr/share/sgml and /usr/share/xml instead of just to
 > > /usr/share/sgml would think the former is overcomplicated relative to the
 > > latter.
 > 
 > Again, the SGML structure will have no references at all to /usr/share/xml. 
 > 
 > Hope this clarifies some things...

Absolutely, thanks.

-- 
Neil Roeth



Reply to: