David Kastrup writes:

 > >  a) this doesn't help for LICR objects like \" as they potentially execute
 > >  \accent and that bombs in math
 >
 > Nope.  We would have the case where \"a has a proper letter
 > equivalent.  In that case we would write the following (excuse the
 > illegal notation
 >
 > \def{\"a}{\ifmmode \relax \ifmmode \ddot a\else ä\fi\else ä\fi}

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 (so that we evaluate to a \char and
therefore get an active math definition in the first place)


 > >  b) and even if \"a expands to a \chardef then the chardef number
 > >  depends on the outer fontencoding so might vary from case to case
 >
 > That's the real problem.

second real problem

 > How about squashing this particular problem with
 > \protected\def{\"a}{\ifmmode ...}?

yes, that is indeed possible a way to go (with eTeX version 2 that is)

i think i did say last night that etex doesn't havesomething to stop scanning
for \omit andthe like, but my memory played tricks on me.

it did not have such a thing in version 1 but with version 2 \protected was
extended to do exactly that stop the scan of an \omit

so that would be one point in favour of an eTeX based solution, as suggested
in the policy discussion

frank