[cad-linux-dev] Re: polymorphic inheritance
- From: Bruno Postle <bruno@xxxxxxxxxx>
- To: CAD linux development <cad-linux-dev@xxxxxxxxxxxxx>
- Date: Wed, 25 Aug 2004 11:47:03 +0100
On Tue 24-Aug-2004 at 15:22 -0500, Eric Wilhelm wrote:
>
> >polymorphic inheritance.
>
> What exactly do you mean by that? Is the inheritance in types or
> references?
References. Sorry, was just throwing words around without
explaining what I mean.
Say you have a drawing showing the elevation of a door, it has a
window, some trim and a doorhandle. In a project you've reused this
as a standard library item using xref (ignore the fact that AutoCAD
is slow as hell when using xrefs this way).
Suddenly you have a need for a drawing of a door that has a
pushplate instead of a handle. With a normal CAD system you have
the following options:
1. Create a fresh new door drawing and use that throughout.
Drawback: there is no "code-reuse", the two doors are now not
related in any way.
2. Go back and fix the original door drawing so that it is itself a
combination of two drawings - A blank door and a handle. Then
reassemble a new door-with-pushplate drawing from these bits and
use that.
Drawback: you are tinkering with an item that might be being used
in a unknown number of locations by other people - Better get it
right or this could be *very* expensive (for this reason, you
probably don't have write-permissions on the file in the first
place).
This is similar to the kind of problem programmers face. With
object-oriented programming, if you have a method that shouldn't be
there or needs to behave differently, just subclass the object and
rewrite the method.
With a directory-based CAD file-format you can have this 3rd option,
because every element that is used to define the door-handle is an
independent file with a persistent address.
With Draft, a reference (block/xref) element has a list that can
point to one or more 'drawings' (hence "multiple inheritance"). It
also has an 'ignore' list of elements - So a door can be subclassed.
> If you are addressing individual files outside of a drawing, you
> are going to kill the "grep litmus test."
>
> I think anything to do with inheritance and references takes us
> too far towards relational drafting. That doesn't fit the
> rhizopod model.
Yes, none of this has any relevance when using a hub as an
intermediary between existing CAD tools.
Though the other litmus test of a "good idea" is that it makes
possible things that were never possible before.
> So, are you already designing the "sardine" spec here?
No, I'm designing an extension to the spec for the 'reference'
object, I want everything else to be identical to rhizopod. If I
can tag these extended reference elements with "version: foo", then
your library will know not to bother interpreting them - It will
spit out a warning and won't core dump on people.
--
Bruno
Other related posts:
