Warning: this is an htmlized version!
The original is across this link,
and the conversion rules are here.
 (find-eevanim "channels.swf" 600 400)

The animation above shows Emacs sending commands to xterms one line at
a time through "communication channels".

When the user types <F9> Emacs checks if the current line starts with
a "*"; if it does, then the rest of the line is executed as Lisp code,
and if doesn't then the line is sent to the program at the end of the
current communication channel. The user only had to type a series of
<F9>s at the Emacs window for all this to happen.

The line

  * (eechannel-xterm "A") ;; create

creates a xterm "listening to channel A", and sets the default channel
to A; similarly,

  * (eechannel-xterm "B") ;; create

creates a xterm "listening to channel A" and sets the default channel
to B. After that some lines are send to the xterm at channel B, then

  * (eechannel "A") ;; change target

sets the default channel to A; the next lines are sent to A, and so

After executing/sending all those lines the user switches the focus to
an xterm and types "exit" there to close it, and then does the same
with the other xterm. Note that these xterms "listen" to the user's
keystrokes as usual, but _in addition to that_ they also listen to
input from Emacs.

In diagrams
Here's a diagram that shows roughly what we have when there's an emacs
and an xterm windows running on X. Many details have been omitted -
for examples, the real communication happens through fifos and ptys,
that are not shown - but it's easy to build a complete diagram from

             keyboard  mouse      display	    	  
                 |       |           ^		   
                 v       v           |		   
            |                            |	   
            |             X              |	   
            |                            |	   
   key/mouse | ^ display    key/mouse | ^ display   
      events v | commands      events v | commands       
         +---------+              +---------+       
         |         |              |         |       
         |  emacs  |              |  xterm  |       
         |         |              |         |       
         +---------+              +---------+       
                            chars and | ^ chars and 
                              signals v | signals   
                                  |         |       
                                  |   sh    |       
                                  |         |       

When we do (eexterm "A") in emacs, what happens is this:

               |         |
               |    X    |
               |         |
             /  ^       \  ^
            v  /         v  \
   +---------+            +------------+
   |         | initiates  |            |
   |  emacs  |:::::::::::>|   xterm    |
   |         |            |            |
   +---------+            +------------+
                            ::   |  ^	
                            \/   v  |   
                          |            |
            eeg.A.pid <-- | eegchannel |
                          |            |
                            ::   |  ^	
                            \/   v  |	
                          |            |
                          |     sh     |
                          |            |

  emacs runs: (find-bgprocess "xterm -T 'channel A' -e eegchannel A /bin/sh")
  xterm runs:                                          eegchannel A /bin/sh
  eegchannel saves its pid at ~/.eev/eeg.A.pid and runs:            /bin/sh

At this point the xterm is running a shell that is "listening on
channel A"; eegchannel pretends to be transparent and passes all
characters and signals in the vertical arrows transparently - but when
it receives a certain signal (a SIGUSR1) it reads the contents from
the file ~/.eev/eeg.A.str and passes it to the shell, as if those
characters were coming from the xterm - i.e., as if the used had typed

Here are the details of this "protocol":
when we type F9 on a line containing "echo foo",

				    |            |
				    |   xterm    |
				    |            |
   +-----------+                         |  ^	  
   |           | --> eeg.A.str           v  |     
   |   emacs   | <-- eeg.A.pid      +------------+
   |           | -----------------> |            |
   +-----------+     eeg.A.str ---> | eegchannel |
                                    |            |
                                         |  ^	
				         v  |	
				    |            |
				    |     sh     |
				    |            |

  emacs saves the string "echo foo\n" into ~/.eev/A.str,
  emacs reads the pid of eegchannel from ~/.eev/eeg.A.pid (say, "1022"),
  emacs runs "kill -USR1 1022",
  eegchannel reads the string from ~/.eev/A.str, and sends it to the shell.

NOTE: one frequent question is: "why are you using two normal files
and SIGUSR1s for the communication between emacs and eegchannel,
instead of making them talk through a fifo?" - the answer is: try to
explain the details of fifos - including creation - to someone who
knows little about the inner workings of a *NIX kernel and who is
uncomfortable to accept another kind of black box... The way with two
files and SIGUSR1s is simpler, works well enough, and I've been able
to write all the code for it in a few hours - and I still don't know
enough about fifos to implement a fifo version of this.

For more information:
  The code that runs in the movie:     (find-eevex "anim.e" "movie1")
  The preparatives for the recording:  (find-eevex "anim.e" "movie1_A0")
  The code for F9:             (find-eev "eev.el" "eechannel-do-this-line")
  An auxiliary program:        (find-eev "eegchannel")
  The program that produced the animation:
  Compiling vnc2swf (and some auxiliary libraries and programs):
                                (find-eevex "anim.e" "compile-ming")
  More about SWF:               (find-eevfile "anim/README")

Update (2012): eepitch
eechannel is non-trivial to set up: it invokes expect and xterm - as
external programs -, it saves temporary files into the directory
"~/.eev/" (by default), and it calls this Expect script, which should
be in the PATH:

                        (find-eev "eegchannel")

Usually we can rely on the script "eev-rctool" to take care of all the
copies of files and patches in rcfiles,

                        (find-eev "eev-rctool")

but there is a much simpler alternative. eechannel has been mostly
superseded by another tool, `eepitch', which needs no setup at all,
can be loaded with just `(load "eepitch.el")', does not depend on
other files (it doesn't even depend on eev), and runs entirely inside
Emacs - it sends lines to shell buffers instead of to xterms, and it
comes with support for interacting with more than 50 languages and
other interactive programs (eechannel by default interacts with a
shell). For more information on eepitch.el, please see:

                        (find-eev "eepitch.el")