Thierry,

 > » i'm not sure i agree on the split for the series attribute. and it doesn't
 > » really origins in the CM history --- it does origin in "Methods of Book
 > » design" by Hugh Wiliamson.
 >
 > interesting info, thanks. i stand corrected.

how should you know my thought process from looking at the code? :-)

 > » one idea with the axis' or attributes was that it should be desirable (and
 > » sensible) to change individual attributes while retaining al others. Now i
 > » claim that there is not much argument for changing individually width but
 > » retaining weight or the other way around
 >
 > That's a point.  Here is a (maybe bad--tell me) counterexample: I

i was replying to explain my original reasoning and to spawn discussion if
necessary (i don't doubt that there are counterexamples where the separation
of this axis into two is desirable). the question that i think one has to ask
is, are they relevant enough to implement the separation as a spearate concept
or are the problems resulting from it (eg sparse font table) are worse and
outweighting them? i'm not saying that my reasoning back then was correct
(though it did look reasonable to me)

to give another (perhaps even stranger example): would you think a colour axis
on fonts is appropriate? say colour in the sense of "modified" like outline
only, gray, ...  --- my feeling is not, even if you could produce comparible
examples like the one you made for width below --- what do you think in that
case?

 > define an abstract environment as a noarrower column using some narrow
 > sansserif font. Within this environment, I want to be able to typeset
 > anything that could be in the text, including weight variants. Another
 > similar use: in a bilingual (?)  document, i keep the translation in a
 > narrow version of the font, everything else affected by the same font
 > variations. Well, yes : my examples could be easily treated by
 > declaring a xx-narrow family, rather than having width & weight
 > separated. My problem is on the practical/genericity of the markup
 > side. something like \fontfamily{\f@family n} is, i believe, very
 > fragile.

a) i like your examples, they are worth thinking about

b) i don't like the concept of fake families, it is fragile as you say and
while it may be alright in a one-off case if you starting using it on regular
basis then something is wrong (either you need an extra axis or perhaps
something else is amiss)

c) i wonder if what you need here is an axis or perhaps you need a mechnism in
which you can control your defaults better, eg a way to say, that within this
environment/region the \bfdefault is "bc" and ... tossing in ideas

 > I see that your
 > » i mean proper classes (not a generic one like article et al) can't cater for
 > » more than a single font set anyway, can it?
 >
 > breaks my argument as well...

it breaks only the argument that the generic classes are of any use :-)

good night
frank