🚀 go-pugleaf

GNU EMacs 18.59.1 of Wd Dec 9 1992 on woody (berkeley-unix)
  [really.. Sun OS 4 on Sparc2]

I have an apparantly repeatable problem.

I have a large *shell* buffer (4.5Mb) which was running a
compute-bound Unix process producing intermittent output; the only
other buffers were *scratch* and *Buffer List*.  I went to the top of
the *shell* buffer, and did "^S***" to find "***", an indicator
produced by the Unix process.  Having found the first instance, I went
to the next instance by typing "^S".  Such instances are 10's of lines
apart, and Emacs must update a significant fraction of the screen for
each instance found.  I repeated *just* "^S" about 400 times in rapid
succession (I was interested in how many occurrences there were of the
indicator; yeah, I could have built a macro to count, but one gets
started on one approach and just carries through if it is a one time
event).  At this point, instead of displaying the next instance, Emacs
said "Back to top level" in the command line area at the bottom of the
screen.  In fact, most keys (including "C-g" and "C-h l", "Meta-x")
responded "Back to top level". Only a few other keys worked (C-l, C-f
and C-b).  I figured I had lost Emacs completely to weirdness.  After
a few minutes of trying lots of things, surprise, surprise, suddenly
everything is back to normal. It is interesting that (display-time),
which I always start via my .emacs file, did not update the time on
the status line during the "weirdness" period.

The problem seems to be repeatable; get a big buffer with irregular
text structure, "^S" to find a relatively common string, and then
simply lean on the "^S" key and let auto-repeat generate lots of
continue-incremental-search requests.  I don't know how to reliably repeat the
recovery from the wierdness, but my few successful experiments hint
that the computing process finally producing some more output to be
displayed in the shell window is probably the key.  I also once
got the message "Garbage Collecting", and that seem to un-confuse Emacs.

Guessing: Perhaps the Emacs display logic gets behind, so ^Ss aren't
processed fast enough, and I consequently overflow an input buffer
with ^Ss?  Is there is a timeout involved somewhere?