🚀 go-pugleaf

My previous followup got clobbered somehow; sorry for the empty posting.

Rick Kuhn



Joseph F. (Jeff) Hull (jhull@thor.NoSubdomain.NoDomain) wrote:
: 
: Are there any generalizations we can make (with any degree of 
: assurance) about what constitutes a "good" standard interface?
: Or a bad one?
: 
One characteristic of a good interface (i.e. good w.r.t. a particular
problem) is that it provides a collection of services that are "close
to the problem domain", i.e., neither too high or too a level of
abstraction.  (Lots of people have written about this, so
what I've said isn't particularly new.)  For example, Xlib might be
considered a good interface for dealing with the X protocol, but not
for doing higher level things like writing a menu driven package, where
you want widgets to do menu bars, etc.

..


: |> 
: |> These are major research questions, of course, but one problem with 
: |> such interfaces today is lack of rigorous definition of the services 
: |> provided.  ...  Formal descriptions would be even better.  
:  
: Who is proposing formal systems for such descriptions? How can I 
: get in touch with them?  How do we get them involved in this thread?  
: How do we get copies of their proposals posted here so we can read 
: them?

There are a variety of formal description techniques that are suitable
for these interface descriptions; VDM and Z are becoming well known.
The programming research group at Oxford has specified some UNIX
interfaces using Z.  However, the IEEE Tech. Committee on Operating
Systems (TCOS), the parent body of the POSIX standards, does not plan
to write formal descriptions of the POSIX interfaces.  TCOS has decided
that formal interface descriptions would, in most cases, require too
much effort to write, and there are relatively few people knowledgable
enough to either write them or use them if they were provided.
Although I think formal descriptions would be beneficial, I think the
TCOS view is probably valid.  What might make formal descriptions cost
effective would be tools to generate conformance tests from the
descriptions, since building conformance tests is currently very labor
intensive.  I've heard talk of some efforts in that direction and have
been meaning to track down the details, but don't know much more now.

The situation is different for protocols.  There are some formal
descriptions of OSI protocols; NIST and other organizations have built
tools for generating code from specifications written in Estelle.  
Some of this stuff can be ftp'ed from osi.ncsl.nist.gov.

--

Rick Kuhn                                   Telephone: +1 301 975 3337       
Natl Institute of Standards & Technology    Fax:       +1 301 590 0932      
Technology Bldg. B266                       Internet: kuhn@swe.ncsl.nist.gov
Gaithersburg, Md.  20899  USA                       DRKuhn@dockmaster.ncsc.mil