www.digitalmars.com         C & C++   DMDScript  

D - Multimethods

reply Mark Evans <Mark_member pathlink.com> writes:
http://nice.sourceforge.net/visitor.html

Good advice from the Nice language,
http://nice.sourceforge.net/index.html

Mark
Feb 19 2003
parent reply "Walter" <walter digitalmars.com> writes:
"Mark Evans" <Mark_member pathlink.com> wrote in message
news:b314qm$sns$1 digitaldaemon.com...
 http://nice.sourceforge.net/visitor.html

 Good advice from the Nice language,
 http://nice.sourceforge.net/index.html

Do all the incarnations of a particular multimethod function all need to be done at once (like methods in a class, or members of a template)?
Mar 07 2003
next sibling parent reply Mark Evans <Mark_member pathlink.com> writes:
Do all the incarnations of a particular multimethod function all need to be
done at once (like methods in a class, or members of a template)?

What does that mean, 'done at once'? Mark
Mar 07 2003
parent reply "Walter" <walter digitalmars.com> writes:
"Mark Evans" <Mark_member pathlink.com> wrote in message
news:b4b6sq$2fpg$1 digitaldaemon.com...
Do all the incarnations of a particular multimethod function all need to


done at once (like methods in a class, or members of a template)?


All be in the same scope, just like overloaded functions must be.
Mar 07 2003
parent reply Mark Evans <Mark_member pathlink.com> writes:
Walter says...
Do all the incarnations of a particular multimethod function all need to
be done at once (like methods in a class, or members of a template)?

What does that mean, 'done at once'?

All be in the same scope, just like overloaded functions must be.

Off the top of my head, no. The (first-class) generic function has some scope but the multimethods it 'owns' may live anywhere. It's just a question of how the call is dispatched. That determination is made by the ("multi") argument specializers and the design of the dispatch rules in the language. Those rules could be as wild as your imagination. Of course a major benefit of multimethods is that they can (should?) live outside class scope. This technique solves exactly Bill's problem of how to 'add members to multiple existing classes.' In terms of scope they might, e.g., live in several different modules, even if none of them are class members per se. Always subject to correction, Mark Possibly helpful links, and a quote: http://www.openbg.net/sto/newsread.php?grp=comp.lang.clos&msg=msg00246.html http://www.openbg.net/sto/newsread.php?grp=comp.lang.clos&msg=msg00244.html http://www.op59.net/cmm/cmm-0.15/users.html http://www.functionalobjects.com/resources/white-paper.phtml http://monday.sourceforge.net/wiki/index.php/MultiMethods http://www.cyberdyne-object-sys.com/oofaq2/body/basics.htm#S1.19 http://www.psg.com/~dlamkins/sl/chapter14.html http://www.cis.ohio-state.edu/~gb/Brew/Publications/HalfNHalf.pdf http://www.research.ibm.com/people/d/dgrove/papers/phd-thesis.pdf http://www.cs.washington.edu/homes/todd/research/ecoop99.ps 'In object-oriented languages with multimethods...the appropriate method to invoke for a message send can depend on the run-time class of any subset of the message arguments, rather than a distinguished receiver argument. Multimethods unify the otherwise distinct concepts of functions, methods, and static overloading, leading to a potentially simpler language. They also support safe covariant overriding in the face of subtype polymorphism, providing a natural solution to the "binary method" problem and a simple implementation of the "strategy" design pattern. Finally, multimethods allow clients to add new operations that dynamically dispatch on existing classes, supporting a form of what we call "open objects" ... that enables easy programming of the "visitor" design pattern and is a key element of aspect-oriented programming. Open objects also relieve the tension observed by others between ease of adding operations to existing classes and ease of adding subclasses.'
Mar 07 2003
parent reply Dan Liebgold <Dan_member pathlink.com> writes:
Correct me if I'm wrong, but aren't multimethod dispatch methods always some
sort of runtime search at the time of the method call?  Even if the search is
optimized by the compiler (to a binary search or somesuch), doesn't this
implementational requirement cross the line in terms of performance?

Dan
Mar 10 2003
next sibling parent Bill Cox <bill viasic.com> writes:
Dan Liebgold wrote:
 Correct me if I'm wrong, but aren't multimethod dispatch methods always some
 sort of runtime search at the time of the method call?  Even if the search is
 optimized by the compiler (to a binary search or somesuch), doesn't this
 implementational requirement cross the line in terms of performance?
 
 Dan

That sounds right to me. Bill
Mar 10 2003
prev sibling parent reply "Walter" <walter digitalmars.com> writes:
"Dan Liebgold" <Dan_member pathlink.com> wrote in message
news:b4ip35$ao7$1 digitaldaemon.com...
 Correct me if I'm wrong, but aren't multimethod dispatch methods always

 sort of runtime search at the time of the method call?

Yes.
 Even if the search is
 optimized by the compiler (to a binary search or somesuch), doesn't this
 implementational requirement cross the line in terms of performance?

It still should work out better than a sequence of if(dynamic_cast<...
Mar 10 2003
parent yuqianzhou yahoo.co.uk writes:
In article <b4jrb6$cvc$2 digitaldaemon.com>, Walter says...
 Even if the search is
 optimized by the compiler (to a binary search or somesuch), doesn't this
 implementational requirement cross the line in terms of performance?

It still should work out better than a sequence of if(dynamic_cast<...

Great! Walter, sounds like you're interested in implementing multimethods in D? If you keep your minds open like this, I'm sure D will have a bright future! BTW, the following is what I wish to see before version 1.0: -- Object persistence: save/load objects to/from memory/disk/network. -- covariant (or anchored) type not just on function return type, but also on parameters: e.g. even without template, we can still write the correct deepCopy for any object as: like this deepCopy(like this anObj); (for anyone who read Eiffel, you know where I come from :-).
Mar 11 2003
prev sibling parent reply "Craig Black" <cblack ara.com> writes:
Hey peoples,

I would like to thank Mark Evans for his numerous contributions to this
mailing list.

Multi-methods are great!!

Implementation can be done in a number of ways.

O(1) complexity can be achieved by indexing types and storing an
N-Dimensional lookup table of function pointers.

If the memory footprint for the N-Dimensional table gets to be too large,
then you could store the function pointers in a binary tree.

If multi-methods are implemented correctly, they will enhance performance,
and save a lot of coding in many instances.  Writing your own
multiple-dispatch is a pain, and usually involves long switch statements
(yuck)!  I know from experience.

Craig
Mar 24 2003
parent reply "Mike Wynn" <mike.wynn l8night.co.uk> writes:
"Craig Black" <cblack ara.com> wrote in message
news:b5o0du$iar$1 digitaldaemon.com...
 Hey peoples,

 I would like to thank Mark Evans for his numerous contributions to this
 mailing list.

 Multi-methods are great!!

 Implementation can be done in a number of ways.

 O(1) complexity can be achieved by indexing types and storing an
 N-Dimensional lookup table of function pointers.

 If the memory footprint for the N-Dimensional table gets to be too large,
 then you could store the function pointers in a binary tree.

Mar 24 2003
next sibling parent reply "Craig Black" <cblack ara.com> writes:
 why not compile the binary tree into a set of cmps and jmps ?

Hmmm ..... sounds difficult to implement but a good idea! Craig
Mar 31 2003
parent reply Ilya Minkov <midiclub 8ung.at> writes:
Craig Black wrote:
why not compile the binary tree into a set of cmps and jmps ?

Hmmm ..... sounds difficult to implement but a good idea!

Why not optimise out all of the cmps and jmps altogether then. ;> -i.
Apr 09 2003
parent "Craig Black" <cblack ara.com> writes:
There is no way to "optimize out" the logic in the binary tree.  It must
exist in some form.  Somehow, the multi-method dispatcher is going to have
to find the method that corresponds to the identity of the objects it is
dealing with.  This can be done in a number of ways, none of which vaporize
into thin air when you optimize them.

Craig

"Ilya Minkov" <midiclub 8ung.at> wrote in message
news:b726ev$1ea5$1 digitaldaemon.com...
 Craig Black wrote:
why not compile the binary tree into a set of cmps and jmps ?

Hmmm ..... sounds difficult to implement but a good idea!

Why not optimise out all of the cmps and jmps altogether then. ;> -i.

Apr 10 2003
prev sibling parent "Craig Black" <cblack ara.com> writes:
I was just thinking, you could also use a hash table.

Craig
Mar 31 2003