Re: More thoughts

[Jamshed Kakar]

Hi,

> I think I know enough Javascript after that to make it so that you can
> move icons where you'd like and click 'save', and have it do the Right
> Thing.

How are you finding the speed of Javascript?  Even simple applications
(calendar for example) I've cooked up have been fairly slow.  I would
be somewhat concerned about speed once you got a tonne of SVG objects
to take care of.  It'll be interesting to see how this pans out.

> I'm noticing some quirkyness which different devices (even different
> Cisco devices).  MAC addresses, for instance, don't seem to be format
> specified.  Sometimes it's a string of octets all mashed together,
> sometimes :, or space, or dash seperated.  I've been trying to think
> of how "quirks" should be handled.  Basically a sysdeps/ hierarchy
> that has code in it to handle different devices?  Is it reasonable to
> write this in Scheme?

What language(s) do you plan to write in?  Is it worth having code to
handle different devices; ie: are they sufficiently different to
warrant this or would it be better to have one set of "smart" code to
handle the whole thing?  Look at the "Strategy" pattern in Design
Patterns- this may be a better way to mimic a sysdeps/ hierarchy.

Something I've been experimenting with which I think will be crucial
is how to actually test a network application.  Thus far I've been
working in C++ and have found that the only way to test such
applications is to *totally* encapsulate all modules- client, server,
etc.  This means main (), aside from error checking, shouldn't be much
more sophisticated than:

int
main (int, char**)
{
  Radar::Server server;
  server.run ();

  return 0;
}

So far I've been able to produce a server object that spanws itself as
a daemon when run () is called.  The next trick which I haven't quite
gotten around to implementing is having it return the pid_t of the
daemon process to the main thread... having the code designed in such
way means it can be automatically started and stopped in unit tests.
Without having unit tests require more than one machine I see this as
the closest way to mimick the distribution of responsibility that
occurs in the real world.

> I also may start writing a MIB compiler soon.  CMU's snmpwalk takes
> forever to read all of the MIBs in, I have a feeling that their
> package is going to be unsuitable for linking in.  If not, I'll have a
> bunch of design work to do on the Ideal SNMP agent. =)

MIB?

Take care,
Jamu.

-- 
 Premature optimization is the root of all evil. -- Donald Knuth