"This is an attempt to implement Emacs's (dabbrev-expand) in classic vi.
"
"Author: Will Mengarini <seldon@eskimo.com> <http://www.eskimo.com/~seldon>
"
"Everything in this file should be directly includable in a .exrc file except
"for the control characters in the final line (which is the only line of
"code; everything else is documentation).  To put those robustly on a web
"page I put, for example, "^" then "[" to represent <Escape>; you'll need to
"edit all the caret notation in the final line of code to make it work.
"If you try to mess with any of the intermediate versions, don't forget to
"edit their caret notation too.
"
"This code works, and I still actually use it when I'm using classic vi,
"which happens on dial-ups to Unix shells where I don't want to run Emacs.
"However, it's not a full-fledged (dabbrev-expand), as you've already
"guessed if you know anything about classic vi.  Most of this file is
"basically a war story, although it ends with working code.
"
"This was the original plan:
"
"  ^V^[
"     Exit input mode
"  my
"     Set mark y
"  b
"     Move to beginning of word (the abbrev) from which we were invoked
"  "yyw
"     Yank abbrev to register y
"  o?\<^V^[
"     On next line, code start of backward search for word beginning
"  "yp
"     Append register y as word to search for
"  0"ydd
"     Go to beginning of line that has that code & delete to register y
"  `yb
"     Move to beginning of abbrev
"  @y
"     Execute register y (error out if not found)
"  "yyw
"     Yank expansion to register y
"  `y
"     Return to position y (after the abbrev)
"  bdw
"     Move to beginning of word & delete it
"  "yp
"     Paste expansion in place of it
"  A
"     Return to input mode
"
"That would have produced this macro:
"  map! ^V^K^V^D ^V^[myb"yywo?\<^V^["yp0"ydd`yb@y"yyw`ybdw"ypA
"Before you keep reading, see whether you can figure out why
"that abends the .exrc with the error message "missing rhs".
"
"[beat]
"
"The problem is the choice of keymapping; "rhs" stands for right
"hand side, & this is failing because ^D is the EOF char!
"
"That exemplifies the crap I had to put up with to get this
"to work.  I couldn't map it to ^C^J, my Emacs mapping
"for (dabbrev-expand), because ^C always just beeps, even
"though the .exrc gives no error message when it's mapped.
"
"The next problem was that, amazingly, using the same letter for
"the mark & the register was failing.  I still don't really believe this,
"but when I replaced mark x with mark y in the development version,
"the replaced version fails.  Unfortunately (I guess) I later improved
"the code by taking out a brain fart--yydd instead of just dd--& when
"I did that, the x/y nonsense stopped happening.  Just in case you
"don't believe me, here
"  map! ^V^K^V^K ^V^[mxb"yywo?\<^V^["yp0"yyydd`xb@yyw`xbP dwA
"is a copy of the final working version with the brain fart still
"there, & with mark x used instead of y.  If your vi has the same
"dain bramage as mine, you should be able to make that the active
"version, verify that it works, then change the "x"s to "y"s, &
"it should fail.  But don't try to read that working version yet;
"there's more to the horror story.  After I gave up on a mnemonic
"(for suitable values of "mnemonic"--same as I use in Emacs, anyway)
"keymapping, I just picked an ergonomic one instead, ^K^K.  This
"  map! ^V^K^V^K ^V^[mxb"yywo?\<^V^["yp0"yyydd`xb@y"yyw`xbdw"ypA
"would've been the working version if the x/y swap had sufficed.
"No such luck, though.
"
"Mysteriously, I started getting "can't yank inside global/macro"
"where I try to yank the expansion into register y, overwriting
"the search expression; & switching to an unused register had no
"effect.  It's some random failure inside vi/ex.  I hacked the
"problem by doing an anonymous yank, & that works, but requires
"contortions for deleting the abbrev, since that'd overwrite the
"anonymous yank.  Instead I do the paste before the deletion, then
"move a char forward (I use the space char for that command instead of
""l" since I've remapped "l"), and delete what remains of the word.
"
"Here are the 3 versions for review.  The final one "works".
"  map! ^V^K^V^K ^V^[mxb"yywo?\<^V^["yp0"ydd`xb@y"yyw`ybdw"ypA
"  map! ^V^K^V^K ^V^[mxb"yywo?\<^V^["yp0"ydd`xb@y"wyw`xbdw"wpA
"  map! ^V^K^V^K ^V^[mxb"yywo?\<^V^["yp0"ydd`xb@yyw`xbP dwA
"
"That's for suitable values of "works", because it'll fail on a
"one-char abbrev, & I can't find a way around that.  I did try.  What
"I needed is a synonym for "b" that won't move off the first char of
"a word.  Doing "?\<" (search backward for word beginning) behaves
"like "b".  Searching backward for a word separator will work unless
"the abbrev begins a line, so inserting a search sentinel at the
"beginning of the line is necessary.  (Remember it needs to be removed
"before the macro errors out if the search for the expansion fails.)
"So I spent another hour and a half trying to get a helper macro to
"work that would do b-without-changing-words.  It got as far as this
"  map ^V^K~ mz0i~^V^[`z ?[^A-Za-z0-9_]^V^[ mz0x`z
"which I think is correct, but it errors out on the final "`z" because
"mark z somehow gets cleared & never reset.  Changing the name to something
"previously unused has no effect.  The macro works correctly when executed
"by hand, but when run as a macro, the final "mz" just doesn't happen; this
"can be verified by attempting "`z" by hand after the macro fails.
"This might work under a less flaky vi; nvi, perhaps.  If you do that,
"you should be able to replace all the "b" commands in my (dabbrev-expand)
"with "^V^K~"s & get a (dabbrev-expand) that'll work on 1-char abbrevs.
"
"As for vi classic, there've been enough failures already
"on macros that'd work if vi vi vi worked that further hacking
"doesn't seem worth it.  So abbrevs must have at least 2 chars.
"
"Also, using (dabbrev-expand) is un-undoable.
"
"I don't feel like implementing the rest of elisp in vi tonight.
"
"But anyway, hey, it's vi, & now it has a working (dabbrev-expand).
"Here's the final result:
"
map! ^V^K^V^K ^V^[myb"yywo?\<^V^["yp0"ydd`yb@yyw`ybP dwA