Re: Generating high-level formatting output

[Joerg Wittenberger <Joerg.Wittenberger@xxxxxxxxx>]

>>>>> "NG" == Norman Gray <norman@xxxxxxxxxxxxxxx> writes:

NG> than a TeX representation of the FOT).  The weakness of this is
NG> that it encourages you to revert to the bad-old-days of one
NG> stylesheet per output format.

Period.

NG> A better way, and one I might return to if I decide we need 'roff
NG> output, would be to do the `print formatting' step as a
NG> transformation from the document DTD to some DTD oriented towards
NG> high-level formatting, then further transform that to LaTeX,
NG> 'roff, or whatever using a simple transforming step which, because

I'd like to second that.  I remember a recent thread ("RE: About the
source library") which in turn came effectivly to the same conclusion
at how to design the internals for "Openjade" (or whatever).

NG> The output DTD could be something simple like:

NG> <div level="1"><title>Section title</title> <para>Here is some
NG> <span type="emph">emphasised</span> text.  </div>
...
NG> LinuxDoc DTDs were more-or-less a translation of LaTeX structures
NG> into SGML -- is that right?

True.  Some years back I went that way.  I wrote a DTD targeted to
minimal typing, totaly removing all formating related tags (like
<toc>).  Then I processed those documents adding even more taging than
LaTeX would require (implying all those infos).  The result of that
stage correspondent approx. to your example.  This intermediate
representation was fed to some backends writing high level LaTeX,
Lout, *roff -man, html and GNU info.  Some glue code runs the
appropriate formaters for .ps, plain ascii, xfig to eps/gif converters
etc.

At that point I added support for Linuxdoc exactly the way you
described - I simply introduced another transformation from Linuxdoc
to my LaTeX-alike-with-all-that-rearranging-done.  It took < 2h.

Last week I added another front end reader for IBM bookmaster syntax
and now I'm happily converting from bookmaster to docbook DTD (not yet
completly automatic).

You see: the way works.  For me it seems to work more efficient that
writing DSSSL style sheets, that is I can get much more work done than
by writing DSSSL and fighting jade/tex.

NG> With that developed, and a suitable set of output elements
NG> identified, the next step could be to designate these as
NG> DSSSL-type FOs, and support them directly in back-ends.

NG> Does that make any sense?

I believe there is nothing wrong with the dsssl FOs even though it's
sometime easier to simply use the "backend-FOs" in place than convince
the backend to handle DSSSL-FOs.  But I don't believe that that's
worth to go into any international standard.

At the other hand you are right.  I think it would be worth to agree
upon a "formating oriented DTD" (guess the FOT output of jade or an
extension of that would do) and implement those FOs for a couple of
backends with an xml syntax as front end.  Another group could
concentrated on DTD-transformatio tools...

/Jerry

PS: if you are curious; my server is broken currently, but you can
find my abovementioned work under:

ftp://ftp.debian.org/debian/dists/unstable/main/source/text/sdc_1.0.8beta.orig.tar.gz


 DSSSList info and archive:  http://www.mulberrytech.com/dsssl/dssslist