At 11.40 +0200 2001-05-15, Javier Bezos wrote:
>>>The same should apply to, say, Greek. If I write "barbaros" [well,
>>>imagine it written in Greek] using the same beta, sometimes I would like to
>>>see the first one using a differenf glyph from the second one (a medial
>>>beta, not used currently).
>> And then I suggest that you do this by selecting a (top level) font in
>> which the beta has the medial form, not by using a special \medialbeta
>> command or by requesting that the LICR should incorporate something
>> equivalent to this.
>Thus, we need 2 virtual font for every font encoding. Don't forget
>that iota can be rendered below a letter or "in-line" (2),
>iota and upsilon can be rendered inverted (2), and there is
>the lunate sigma (2). Since these options are independent, are you
>suggesting the creation of 16 (!) vf files and tfm files for every
>font (and encoding)? (And regarding that, Greek is easy compared
>with scripts like Devanagari or Arabic.)
If you have four characters with to glyph variants each and you write 16
documents in which you realize each of the possible selections of glyph
variants (one selection per document) then you deserve to have to use
different fonts for each document. However:
At 11.24 +0200 2001-05-15, Robin Fairbairns wrote:
>isn't this sort of thing done by ligaturing a "boundary character", so
> "bnd" "beta" -> "initial beta"
> "sigma" "bnd" -> "terminal sigma"
>and so on?
>(i've never really understood boundary characters so i may have this
(Looks right to me, Robin.) If that is the kind of variants you are talking
about then of course they should all be in the same font, but the point is
that all these replacements of glyphs should be handled by the font, not by
the user or by LaTeX. In particular there shouldn't be an internal
character representation for the variants in LaTeX, because the variants
are semantically equivalent and therefore identical as characters.
The idea that OCPs could be used to select between variant glyphs in the
font is not without merit, though, _provided_ it does not mess up the LICR.
I'm also sceptical towards that it should be the language which selects a
certain variant form of a character, as that can become a severe drag on
font design. If anything, it should be the font (or its FD file) which
declares that "I provide the following variants of this character",
together with the code for choosing one or the other (this code could well
consist of pushing an OCP). Languages could by all means request a certain
variant form, but the font shouldn't always have to provide it!
I math fonts something like that could be used to handle the choice between
\epsilon and \varepsilon. As I understand it, these are semantically
equivalent---i.e., people will think you've done something wrong if you try
to use them both in the same formula to mean different things (but maybe
Barbara has some counterexample)---and so shouldn't have different internal
representations in LaTeX (as they do today).