Donald,

 > > nope nope (imho:-)
 > >
 > > that's for the case where \"a is precisely *not* executing an \accent but is
 > > actually a glyph in the current font
 >
 > But that is the only case you have to handle!  \accent combinations
 > don't have the right kerning anyway, so just stick \relax before
 > the \accent.

you both are right and i was wrong (for the case of accents). however you would
have to identify that \"a is a glyph first, even more, you would need to do the
right thing concerning any text character eg some are fetched from a different
encoding so all that would be very messy indeed

 > The (killer) problem I see has already been alluded to: The inputenc
 > characters are already active, so you have to have a single definition
 > that works for both the initial expansion of the input text and as the
 > math-active character, without recursion.
 >
 > David's usage of "ä" is probably part of the "illegal notation", but if
 > I may either clarify or fix:
 >
 > Definition to convert from LICR to glyph:
 >
 > \def{\"a}{\ifmmode \relax % make sure we are in math mode to stay
 >           \ifmmode \ddot a%
 >           \else \string ä\fi
 >           \else \string ä\fi}

sorry, perhaps i'm still dumb from my cold or else dumb anyway, but i don't
get you here. what is this supposed to tell me?

one of the problem is that pressing key ä (umlaut-a) on the keyboard maps to
\"a alright in the LICR but that is not equiv to doing

 \string ä

for typesetting ---  the slot to use varies from encoding to encoding. so if i
interpret your definition above correctly then you end up with exactly
typesetting \char `ä always for \"a ... or what?

frank