Here are some thoughts about templates and font selection. Perhaps I
thought about it a bit too long, because the mail got a bit long-winded,
but I hope you'll endure it.

There are several levels of font selection in LaTeX. At the bottom there is
the primitive TeX level, where fonts are specified through TFM file names.
On top of that, there is the NFSS level where fonts are specified using
attributes (encoding, family, series, shape, size, and once NFSS3 is
released case). On top of that, there is what I will call the author level
(since that is the lowest normal authors are expected to go), where fonts
are selected using commands such as \textit, \small, and \bfseries. On top
of that, finally, is the markup level where fonts are indirectly selected
by the markup commands.

The template.dtx file contains a couple of examples of how some template
keys can contain the font selections made by the template, so that in a
sense the markup level to author level mapping of fonts is controlled
through templates. The point I would like to make is that the author level
to NFSS level mapping should perhaps also be controlled through templates;
at least it might solve some problems with the current state of things.

These problems generally stem from that macros that make author level font
selections are usually unaware (i.e., the one who wrote them were, and they
make no attempt to examine) of the actual situation at the lower level, so
one should be prepared to make compromises while mapping the selections to
the NFSS level. In the current setting however, this mapping stays fixed
throughout a document and can only be (responsibly) changed by code which
take complete responsibility for all document fonts. Since there usually
isn't any such code, except in classes that are specially designed for
certain fonts, font implementors have taken the only option they
have---modifying the NFSS to TeX level mapping so that it suits the
standard settings---since that is under each font implementor's control. I
think it's the wrong way to go however, and problems certainly do arise
when a font family has a bx series, but it is the b series which should be
used for normal boldface.

One way to use templates for getting around this is to define a template
for "author fonts", something like

\DeclareTemplate {noparams}{author-fonts}{0}{
   medium      =n [m]   \mddefault,
   boldface    =n [b]   \bfdefault,
   light       =n [m]   \ltdefault,
   upright     =n [n]   \urdefault,
   italic      =n [it]  \itdefault,
   slanted     =n [sl]  \sldefault,
   smallcaps   =n [sc]  \scdefault

(Or perhaps author-fonts should be a type and the template should be named
default or something. I'm not sure, but it seems odd that all templates
without parameters (i.e., having type noparams in the above setting) should
always be affected by the same \UseCollection commands.)
Now if FD files are allowed to declare instances of this template, one
could use those to specify which values the \XXdefault macros should have
when the current font family is declared by that file. ot1cmr.fd would for
example say


since it (unlike most font families) has bx as its standard boldface
series. The AvantGarde (pag) family has a weight named "Extra Light" (el)
but no light weight, and has obliques but no italic or smallcaps. Hence
t1pag.fd would say


And so on. The reason I took m as the default for light above is that I
suspect most families don't have a light series.

Of course, this requires that this template (or an instance of it) is used
each time the family is changed. Hence the definition of \rmfamily should
become something like

   %A (see below about this)
   %B (see below about this)

but I think that is reasonable.

At this point one may think that this turns the power over author to NFSS
level mapping completely over to font implementors, so that class designers
are left without a saying in this, but that is not the case. Everything a
class designer who is creating a class for use with a fixed set of fonts
has to do to resume control over for example what is the standard boldface
series, is to create collection instances for the font families in question
and then \Use that collection throughout the document. A user who tries to
use the specialized class with fonts it wasn't made for will however see
the default settings for those fonts. Hence it works out for the best
either way!

The reasons for the %A and %B in the suggested \rmfamily definition above
do however add a complication. The problem is that when changing from from
one family to another using an author level font selection command, you
expect the author level font selections of series and shape to be
preserved, not the NFSS level selections, and hence one must (i) remember
the author level font selections and (ii) map them anew each time the
family is changed. Supposing that the \XXdefault macro for author level
series and shape are stored in \author@f@series and \author@f@shape
respectively, a command like \bfseries would have to be given the definition


and the %B line above would have to be replaced by


The %A line is because of another technicality. If the author's last font
selection was at the NFSS level then it is more reasonable to expect the
NFSS series or shape to be preserved; this can be done by including

   \expandafter\ifx \author@f@series \f@series \else
   \expandafter\ifx \author@f@shape \f@shape \else

(I suspect there is some better command for doing these tests somewhere in
the experimental LaTeX3 code) at the %A line above. Finally, one also has
to make sure that the FD file in question is loaded before the
\IfExistsInstanceTF test is made (otherwise the fonts might come out wrong
the first time they are selected). Hence one would also have to do some
kind of


(where the hypothetical \EnsureFDFileLoaded would be very similar to the
existing \try@load@fontshape macro---just add a group and don't use
\f@encoding and \f@family to pass data) at the end of %A.

A final thought connected to this: I considered having emphasis (default
value it) and it-emphasis (default value n) keys in the author-fonts
template as well, but now it seems more like selecting fonts for
emphasising text is something which belongs in the markup level. In any
case however, LaTeX2e* shouldn't have these shapes hardwired into the
command, as they are in LaTeX2e.

Lars Hellström