Lippard.Multics 1986-10-13 16:29:06 mst Mon
Subject:  Automatic Rebinding
Date:  Wednesday, 8 October 1986 12:52 mst
From:  Barry Margolin <Margolin>
To:  {mbx >udd>m>jjl>misc>misc}

or:
          But I never learned to bind a manual...
                    Stephen Robbins

Dear Mom,

It's hard being a dynamic linker.  People want me to do so much work for
them.  And then they get upset when I follow their instructions to the
letter.  Sigh.  So to "solve" my having done something RIGHT, they
complicate my job AGAIN.  I wish I had taken your advice and become an
airline reservation system.

The job started out great.  When someone wanted to call procedure "B",
they'd call it first with a dummy address.  I'd notice, of course, and
load in the REAL routine "B." Then I'd replace the dummy pointer with
the B's real entry point.  After that, things'd run fine.  Since I'd
keep the real pointer in my reference name table, I could look up "B"
real fast the next time.  I was consistent, too.

"But what about recompiling?" they started to protest, "If B's address
changes, things will stop working!" Hell, I get things right the FIRST
time.  If I didn't I wouldn't keep my job long.  But if THEY don't I'm
expected to cover for them It burns me up!  I had to solve THEIR
problem.

Along with the routine entry point in my reference name table, I started
keeping a pointer to a block of memory.  In that block of memory I'd put
the address of every reference to "B" I had resolved.  Then on demand, I
could update all the references.  Did they appreciate that?  No.  "Too
much wasted space!" they cried.  Or sometimes, "There's not enough room
to record all my references to B!" Sheesh.

So I tried something else.  I got together with the compiler and had her
generate all indirect calls.  Instead of resolving "B" to point to the
routine itself, I had it point through an indirect entry in my reference
name table (where the pointer to that block of memory used to go, in
fact).  So when they recompiled "B," I'd just change that entry.  Voila!
All their calls went to the new "B." Were they satisfied?  Of course
not.  After that, all I hear is, "Procedure calls have slowed down!"
Don't they understand time-space tradeoffs?

Mother, there's just no pleasing some people.  I've been thinking about
it carefully, and I'm coming home.  Please fix up my old address space;
it'll be great to see you again!  MULTICS [sic] just isn't the place for
me.

                                        Love,
                                        Your son, D. Linker.