Lars Hellström writes: [snip] continuing where I left off last night > An argument for separating these functions is that doing something > innovative in one of the three areas is not making you more likely to > do something innovative in the two others. > E.g. a designer might need > to commision a new template for headings because the design does > something that cannot be achieved simply by setting parameters in the > standard template, but it's only the "heading on page" that needs this. 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. The only thing a heading object should do beside receiving document input for toc or running head, is potentially reconstructing the data that it passes on (but not manage its formatting). Of course it is a grey area here, especially as in LaTeX we are all used to mix document structure with formatting data, e.g. from the examples that Ulrich gave: we do add \\ to a heading to force a line break somewhere (which is effectively tying the document to the used formatting) but then we clearly need to pass something else to the toc as we don't necessarily want the \\ there as well or not in the same place. but abstractly: say for a running head the template may have a specification possibility to say what should go into the running head, ie the title text + a number or just the title text or, the title this some other data but the heading would not describe the formatting (which would belong to some page style object receiving that data). so in summary I don't think the separation here is necessary or worth while > Creating the new template then becomes much more complicated if it also > has to deal with the technicalities of running heads and TOCs. One > might also want to combine one package's nice "heading on page" > features with another package's feature to include author names in the TOC. actually I think we have to drop the believe that such packages will exist - they don't in that form exist in 2e which has a simpler model, and will not work in the l3 model either. What I mean is that there will be no packages that enhance existing templates by adding something to the templates. What you will have is packages that offer new templates for the same object type with different features. just like titlesec is a reimplementation of LaTeX's \@startsection template (if you want to call that a template) but there is no package that adds "quotes" to headings written with titlesec. There is only a package that replaces the chapter commmand with a new one that supports quotes on top > > Unless you start overloading the arguments in a way that \NoValue for a > > TOC title means use the main title for TOC while an empty TOC title mean > > do not use one, or something like that. > > [Shiver] Only a last resort. no not :-) I guess we agree here > > > > what else is missing or which of the above aren't sensible in the > > > > first place? > > > > > > One might consider making one argument a keyval dictionary and collect > > > many of the above in that instead. > > > > > > that would be a document level syntax decision, ie how to write a heading > > command in your document. > > No, I meant it as an *implementation level* syntax. oh, I see what you're getting at > > For commands that do complicated things (and it certainly looks as > though the "make heading" instances fall in this category), it is > generally a fact of life that the number of things you /might/ want to > specify as arguments to them grows much faster than the number of > things you would /typically/ specify as arguments to them; maybe a > typical call has use for $\sqrt{n}$ arguments if there are $n$ > arguments in total; for e.g. $n=16$ that would come out as $12$ you > don't care about. Having a separate mandatory argument for each of them > quickly becomes unfeasible, as one pretty much has to look up the > specification every time writing a call just to remember which order > the arguments were in. > > A better approach, following *nix command line tools, is to have > *options* -- and in particular options that can take values, like > --unified=3, --src-prefix=foo/bar/, or --pretty=full (to mention 3 of > several dozen options of git-diff-tree). TeX doesn't support variadic > commands that well however, so the natural way for us to implement > "options with values" is to have a single mandatory argument whose > value is a dictionary mapping option names to values, and then some > other mandatory arguments for the few things you really need to specify > in almost every call. well, it may be something to think about. but a few comments here - 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 - 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 - one of the reasons to limit ourselves to mandatory args initially was speed, but another one (and we reflected on that one) was that it would force us to keep the semantics simple. I still think that this is basically the right decision, but we'll see. if we do turn up with object types that we have to limit but don't really want to than yes that needs revisiting. As it stands now, the object types that I came across require a small number of input supplies the biggest ones are those for headings and toc entries. Of course, abstractly it is easy to envision an object that requires 20 different input supplies and that formats them somehow, especially in database query -> format type of situations (such as your example with bibtex). but the question is: would/should that still be formatted then with a template model? or should there be some preprocessing through some different method. 'night frank