[plt-scheme] re: question about symbol data type implementation in mzscheme

From: Matt Dawson (mdawson at mninter.net)
Date: Sun Nov 21 09:40:33 EST 2004

Richard Cleis wrote:
>I have developed software for controlling devices...
>...I see one blade of grass and two weeds, perhaps because I don't 
>understand your problem.

This is good "pushback". The problem comes from my own delusions of
granduer. Instead of developing a protocol very specific to one devie
I'm trying to develop an generic protocol that could work with many
different types of devices -- something that could become and open

Here is a little more detail to explain why I think that there will have
to be a proliferation of symbols. 

Industrial automation protocols are usually broken down into two parts.
There is the "application layer" and the "device profile".  The
application layer provides generic infrastructure for communicating with
a device but is not specific to any particular device class. The
application layer could work equally well for a refrigerator or a
digital servo drive. The device profile is a set of rules/conventions
for talking to a particular class of device. My company has a number of
products that use a protocol called CANOpen. The CANOpen application
layer views devices as "things with setup parameters". Each setup
parameter has an a numeric address. The "device profile" for a device
like a digital servo drives specifies what the actual setup parameters
are and what there addresses are. For instance it might specify that
parameter 7 is the set speed for the motor. 

In my scheme inspired protocol symbols like "new", "send", and "define"
are part of the application layer. My application layer models all
devices as "things with objects". There are a generic set of mechanisms
for invoking methods and creating new objects. The device profile for a
particular class of device specifies what the actual objects and classes
are. The device profile for a device like a refrigerator might define a
"refrigerator%" class and an "ice-maker%" class. To determine whether
the ice-maker was working correctly you would send an expression like 

   (send (send device get-ice-maker) operating-normally? )

Here the root device is an instance of a "refrigerator%". The
"get-ice-maker" method returns a reference to an "ice-maker%" object.
The "ice-maker%" class has a method called "operating-normally?".

Symbols like "get-ice-maker" and "operating-normally?" are part of the
refrigerator device profile. Each profile will require its own set of
symbols. A symbol like "operating-normally?" might get used in serval
different device profiles. 
>I have developed software for controlling devices; the paradigm is 
>based entirely on sending scheme expressions to devices that return 
>only scheme expressions.  I keep symbol tables from growing like weeds 
>by organizing the functions into truly primitive operations and 
>grouping information into reasonable lists.  In your example...
>(send sensor serial-number)
>...I see one blade of grass and two weeds, perhaps because I don't 
>understand your problem.
>is one blade of grass that could perhaps return
>'(serial-number driver-build-number width height)
>or whatever grouping makes sense for your project.
>Ironically, I would pay real money for a "sensor" that talks scheme... 
>so I could use it with my "devices" that are controlled with scheme.

Posted on the users mailing list.