General notes:

   * Setting the lisp-find-definition-verbose-p switch to non-nil
enables some extra debugging output.  There is no pause after each
such message, so the cost for enabling this is low, but it does mean
that you have to switch to the *Messages* buffer to see what it said.

   * The *Edit-Definitions* buffer now has a rudimentary
lisp-edit-definitions-mode of its own, which makes it easier to find
multiple definitions via "random access."

   * The lisp-fall-back-on-find-tags switch turns off the "Can't find;
use tags?" query, which I find annoying.  The way I see it, if I had
wanted find-tag, I would have invoked find-tag directly.

   * I have tested ACL, CMUCL, and SBCL, all of which use the new
protocol.  Other CL dialects should still work, though with some loss of
function, but I haven't tested the backward-compatibility code lately.
[SBCL seems to return no file, or the wrong file, for functions
sometimes, but methods and classes work, so for the time being I'm
assuming this is an SBCL problem(s).  -- rgr, 31-Mar-03.]

Some bugs:

   1.  Finding a structure slot accessor in CMUCL sometimes fails the
first time.  It tends to succeed afterwards, so I haven't been able to
track this down.

   2.  ACL won't tell us whether an explicit defgeneric form exists; it
always gives us a source file, presumably the source of the first method
for implicit generic functions.  (method ilisp::object-source-file
(standard-generic-function)) in find-src.lisp implements a workaround: A
defgeneric is reported iff there is a documentation string, which of
course means that explicit defgeneric forms without doc strings are
missed.

   3.  If M-. finds a definition that is already visible and positioned
so that the comments above it are showing, reposition-window-lisp hides
the comments again.  (More of a glitch than a bug, but still annoying.)

   4.  It may be necessary to clean up lisp-setup-edit-definitions,
lisp-next-file, and the other search-based stuff in the ilisp-src.el
file.  In particular, search-lisp and replace-lisp shouldn't use
*Edit-Definitions*; there could be bad interactions with M-. if these
aren't changed.

Some possible enhancements:

   * Package DWIM.  This would be really nice; it wasn't even possible
while the Lisp was restricted to returning a list of file names.  Since
there may be more than one alternative, it would be best to define a
response message that enumerates the alternatives, so the user can
choose.

   * Mouse interface.  I have the code to make M-left on a symbol (or
reasonable facsimile) find the source, but I need to disentangle it
from my other hacks first.

   * The way the code works now, it expects to find a definition
exactly once in the given buffer.  It should look for more hits,
notify the user if any were found, and permit stepping through them in
source order.  (Ideally, it should be able to filter out "false hits"
such as definitions with "#+ignore" in front of them, but that's more
ambitious, and requires going back to the Lisp to ask it to "vet" each
hit.)

   * When it finds the next definition, the ACL eli M-. command tells
you how many known definitions are left; this would be nice to have,
and probably not too tough.  (When the previous enhancement is done,
it should probably only do this for the last hit in a buffer.  The
user would see either "N more definitions" or "N more occurrences of
this definition", but not both.)