part of Achim's reply to "templates for galley" tempted me to think about this once more and offer some thoughts for discussion. main point is i would like you to think about is, what are the invariants of that interface, ie what can be assumed to be the same from one template to the next (within a type) and what can differ. Here is the extract from Achim's mail that triggered this: > o pshape template. Perhaps it would be a good idea to add two arguments > to the pshape template for the horizontal and vertical dimension > (where the exact meaning depends on the template). Say, a template > for a rectangular cutout could interpret the horizontal dimension > as width of the cotout and the vertical dimension as number of lines > to be indented. > > o justification template. I'm not quiet sure how the galley templates > are intented to be used -- the package only defines the templates > themselve no commands to use them. But I think that the justification > templates shouldn't be able to choose whether they affect all > paragraphs or only the next one. Instead the user should be able to > tell the system which case is desired everywhere such a template is > used. to formulate those a bit more abstractly: 1) how much freedom has a template to interpret its arguments (ie data passed to it from the document)? 2) what is the scope a template instance apply itself? is it well-defined? let's tackle 1) first. Achim suggests that the pshape type should get two arguments those meaning should depend on the template. in my opinion this is likely to result in chaos. the main idea behind the template types is that logical structure of the document, i.e., the commands and environments therein, are transformed into instance calls of templates of certain types. this is defined via, say, an xpare interface and does define the document class (no formatting yet) structure. now the task of the designer is to declare such instances (whose names are fixed by the structural part above). He/she does that by declaring instances using \DeclareInstance{<type>}{<name>}{<template>}... and specifying values for the keys that belong to <template>. and that should be enough to design a layout (just a lot of \DeclareInstance) nothing more (we leave out the case that there isn't a suitable template for a design looking like ... and just assume that such templates are all written) so to make this work templates of the same type need to interpret data coming from the document (ie via arguments) in the same way. At best I see that they might be allowed to ignore it. to give an example: in xhj i defined two templates of type measure which change the measure of the galley. One of the templates does this relative to the current measure in the template and another one does this by specify absolute values in its keys. This type accepts an argument to overwrite the left margin width. Now if one template would interpret a value coming from the document as "add this to the left margin" and the other as "this is the left margin width" then guess how exchangable those templates get! as for allowing a template to ignore such values, again an example: suppose we have a list template type which as one of its arguments gets the left indentation of the list matter (or \NoValue if nothing was specified by the user in the document). The intention is to allow a document syntax like: \begin{description}[foobar] where the string "foobar" defines the width of the indentation (eg could be the widest label), in latex2e thebibliography is in fact an environment that is more or less of this type (except that the argument is not optional) Now suppose you have a design where nested lists at a certain level are formatted in a paragraph instead of making each item a separate paragraph. In that case a notion of "left margin" is no longer appropriate. Now one could argue that those two type of lists should not have the same list type, but I think they should since otherwise how do I switch from one such layout to the other for the same "document class" (ie document level syntax) So allowing to ignore argument values is a weak form of acceptance which (in most cases) allows to format a document in different layouts even if the document does contain some elements with "formatting like" data. Of course, another way to avoid this problem would be to disallow arguments to templates which do carry "formatting related" information. But that one the other hand would render the interface much less flexible and if it is too clumsy for the user to guide the layout in certain comon cases that i think something was done wrong. To return to the pshape template: so I don't think it should get arguments that are interpreted differently by different templates. In fact for that particular template type I don't think it should get arguments at all. I see this template as something (again) to be used only as part of other templates to define very specialised shapes in certain layouts --- but shapes which are independent of of document data but exist in their own right. For example, I came recently across a very nicely formatted book whose Capter openings have all be done by having the first paragraph with a triangle like cutout that was part of the design for the chapter head. That's the kind of area where I see applications for pshape. In contrast, if one wants to provide commands that allow the user to directly manipulate the paragraph shape somehow (one example in a restricted sense being Initials) then this should not be done via the templates provided in xhj but by making use of the lower-level interface that comes with galley2.dtx the module xinitials will do something like that =================================================================== do you remember what the second abstract question was? No? Here is another rephrase: to what elements should templates apply that do set parameters of underlying algorithms (in some way or other) for example the "std" template of type justification sets the justification for all upcoming paragraphs of the galley (potentially at least) while the "single" template sets it only for the next. Achim suggested that not the template but the user should decide on the scope. Now I agree with him but who is the "user"? The user is the designer of the document class layout and if this "user" says that the first paragraph of a caption should have the first line centered and any following paragraphs should be justified without indentation then he can do this now by setting a justification-setup =i key to the appropriate value. by making the (i agree) somewhat different feeling templates one and the same type I allowed for flexibility in design. In an earlier implementation I had two types but that resulted in either strange templates (making use of both) or in a lot of repitition, ie the need to have nearly identical templates differing only in one key since that key was of a different type. another way to look at it is to say that the "single" one is actually also setting all paragraphs in the galley (it sets the first to X and the rest to whatever was before X). yet another way is to get rid of "single" and implement what i called "complex" differently since what in practise will be needed are the instances from "std" and from "complex" does this make some sense? ------------------------------ a related question is whether it is permissable for templates to do inheriting of values while other templates of the same type do set all values. I think i should be permissable for the sake of flexibility, however, it is a fact that this makes looking at a design implementation (ie at a list of \DeclareInstance commands) a bit diffcult if not used with care. After all you may have difficulties to guess what is actually gonna be used if such template instances then follow each other. comments? frank