| Re: IDL 6.0 [message #34520 is a reply to message #34519] |
Thu, 27 March 2003 15:01   |
Randall Skelton
Messages: 169
Registered: October 2000
|
Senior Member |
|
|
>> My *hope* is that IDL 6 will extend the current object code base and
>> support (1) operator overloading, (2) proper inheritance mechanisms (i.e.
>> avoiding name space clashes that occur in structures), (3) public /
>> private methods and (4) a C-API interface to object/heap variables. I'm
>> also keen to see an object interface to map drawing...
>
> Oh, dear. Maybe in 6.1. :-(
OK, I realize that I am being a tad greedy here but I really do think that
IDL objects are too limited in their current scope and it is beginning to
become a real problem for me. I do a fair amount of multivariate analysis
where a given vector or matrix is really a composite of various different
data sets. I solve large systems of this sort and have a variety of
routines in c/fortran that work on these composite objects efficiently.
In many cases it is as simple as defining an object that holds a vector or
array, with a hash table that describes what various parts of the vector
or array correspond to, and some ancillary stuff about the measurements.
I quite like operator overloading as it makes working with these composite
objects (i.e. solving them as linear/non-linear systems) easy code and
read. I do this all the time in c++, fortran and (recently) Matlab. I
have some great object graphics routines written in IDL to visualize my
data but I really have no way of actually operating on my data without
resorting to writing very cryptic code. Try re-writing:
xh = xh + invert( (1+gamma) * SaI + Kts # k) # $
( Kts # (y_obs-y_f) - SaI # (x_h-x_a) )
when each addition is 'obj->add(a,b)', scalar multiplication is
'obj->shift(1)', and matrix multiplication 'obj->mm(A,B)'. IMHO, data
encapsulation should not prevent people from actually working with their
data in a logical way!
My message to RSI is that I don't really need any more GUI tools,
spread-sheet views or slow pointy-clicky interfaces. What I would like is
a more complete set of core object programming interfaces so I can
program more logically and manipulate/visualize my data more efficiently.
I hate to admit it, but I have been using Matlab a lot recently as it's
object model is more suited to how I code using data objects. Now if only
it had pointers :(
I promise this will be my last 'IDL objects' rant for a while.
Cheers,
Randall
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
