LATEX-L Archives

Mailing list for the LaTeX3 project


Options: Use Classic View

Use Monospaced Font
Show Text Part by Default
Condense Mail Headers

Topic: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Sender: Mailing list for the LaTeX3 project <[log in to unmask]>
Date: Fri, 13 Nov 2009 23:44:24 +0100
Reply-To: Mailing list for the LaTeX3 project <[log in to unmask]>
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
In-Reply-To: <[log in to unmask]>
Content-Type: text/plain; charset=us-ascii
From: Frank Mittelbach <[log in to unmask]>
Parts/Attachments: text/plain (120 lines)

 > Frank Mittelbach skrev:
 > > 
 > > I was wondering about this. Is it true? In case of the toc I would say no:
 > > all a heading object does is passing data into a toc sub-system and the
 > > formatting of that data is then controlled by templates for that part
 > > 
 > > And the same is true for the running head.
 > Right. I was thinking that it is the heading object that formats text 
 > for TOC and head, but if it just passes data along to be formatted 
 > elsewhere, then a single object is fine. The tripartite model would 
 > anyway have problems with the fact that the whatsits corresponding to 
 > data going elsewhere should anyway be positioned somewhere, and that 
 > would have to be controlled by the on-page component.

well, assuming we end up implementing the galley model like the one in galley2
then this is not such a big problem as this takes care of the puttting
whatsits etc into the right places. But beside that there is the problem that
both the toc and the running header not only receive the title argument but
also some material that is generated e.g., the heading number and then you
have the question: where and when is that generated? and the answer is "before
they are called" and then you are back to a single model of some sort

 > [snip]
 > >  - an open-ended interface is also meaning that you will likely end with a
 > >    babel like confusion (and I don't mean 2e babel but just the biblical
 > >    problem) i.e., everybody is dreaming up new values that are then not
 > >    supported anywhere else etc etc ... and out of the window goes and
 > >    transparency in terms of replacing one design by the next
 > In my experience, it works quite well to just silently ignore 
 > unsupported options (probably because it's equivalent to knowing that 
 > it is there but never caring about its value).

that is good and fine, but only for those that are truely
unsupported/unknown. What I'm worried about is opening the door to
incompatible variants of object types that look like they can be exchanged
with each other (or rather instances build for them) while in reality they

Actually bibtex is not a bad example for it: depending on the bib style
different fields are required to make the style work and sometimes differently
named fields mean the same, so it is not as if you can easily reuse a .bib
file with different styles

for me an object type is described by "the semantics of the object" and the
inputs the the object takes, so an object type that "formats a heading" and
takes a "subtitle" as one of its inputs is a different type to one that
doesn't do this but instead takes an "author name" say

I'm not sure if and to what extend one should go away from that. I guess it is
possible to have some sort of partial order put on object types, like when the
arguments of object type X is a subset of the arguments of object type Y (and
obviously the object types both have the same basic semantics, eg both generate
a heading). In that case, if there is a request for an instance of type Y (the
one with more arguments) then you can probably satisfy it with an instance of
type X instead.

But is it worth doing (including the possible overhead for checking?)? Or is
it better to say an object type is and object type and if you want some
additional or other arguments to go with it then this becomes a different
object type, period. Dunno

 > >  - let's first see if we ever end up with huge object types. I don't get
 > >    believe it. the heading might be the most complex thing we ever see in this
 > >    respect and even here I guess what we end up with is less than 6
 > Already title, toctitle, headtitle, number-flag, toc-flag, and 
 > head-flag makes 6, and that's without author, subtitle, xref-id, 
 > abstract, and other stuff that's been discussed.

yeah, you are making your case I guess. perhaps we have to rethink this. But
to be honest I'm kind of torn here between keeping the interface on that level
lean and fast and making it wider. 

TeX has gotten much faster since I started using it (when it was typing a page
in about 30 seconds) but even though it is nowadays lightening fast in
comparison, I still think there there are limits one should be careful not to
make too slow. And those are essentially those of the "inner loop" of
typesetting. And if many many things end up being instances of some object
type then the input supplies to those instances should work with resonable
speed. There are only a small number of document level commands compared to
what  is happen internally (so xparse etc can  be fairly elaborate) but if
building boxes, setting up paragraph parameters, etc etc all first have to
build up an interface structure, store it, then retrieve it, then my fear is
we are going to cross that boundary fast.

Its a bit like NFSS really, where I came up in the end with a storage model
and an interface model which was flexible enough, but still with a small
footprint in space requirements and processing time - and both were
needed. That interface was ugly in places, I guess, but it got 2e successfully
through more than a decade and I'm fairly sure a more verbose model wouldn't
have managed that.

Maybe the right compromise is to use  #1 .. for major arguments and devote the
last argument (on some object types) to receive a dictionary of some sort as
you suggested. 

However, I'm very much convinced that the input supplies specified by that
dictionary have to be a part of the object type and that any change in them
means we are talking about different types that should be named differently.
Or perhaps not ... maybe that simply needs more thoughts whether it is good to
allow some (but which kind?) of flexibility in that part.

 > The purpose of using a dictionary would be to allow the data being 
 > passed around to have more structure. I also think passing a dictionary 
 > to the TOC entry formatter would be great, as that will allow *it* to 
 > decide whether to include an author name, subtitle, short description, 
 > etc. or not.

it sounds tempting, but it does require that the TOC entry formatter
understands what "author" means and that it means the same as what it thinks
about it, so either you need some very clear definitions about the semantics of
these dictionaries or you need to otherwise control that