Eric Mahurin
3/11/2008 5:05:00 PM
On Tue, Mar 11, 2008 at 9:18 AM, Robert Klemme <shortcutter@googlemail.com>
wrote:
> 2008/3/11, Eric Mahurin <eric.mahurin@gmail.com>:
>
> > Ruby, unfortunately is really lacking on this front.
>
> Did you see my references to libs in RAA? Do they lack something that yo=
u
> need?
Sorry, I didn't see your post. I jumped in the thread later. I have seen
that red-black tree API before.
> When doing geometric
> > stuff, I find every ounce of C++ STL useful. An STL-like ruby lib
> would be
> > useful. External iterators are also very important. Ruby's normal
> internal
> > iterators (what others might call visitors) don't have the same
> flexibility
> > (caller can go forward/backward/read/write/erase/insert at will).
>
> That's an interesting point you make. I haven't noticed the lack of
> external iterators (there *is* in fact an implementation, albeit with
> limited functionality) so far. For me the most pressing disadvantage
> of #each and family so far was that you cannot iterate several
> collections at the same time with this.
Yep. Forgot about that one. That is quite useful for "external" iterators
too.
> For random access you can use
> an Array with index. And for Arrays it would be fairly easy to write
> a thin wrapper around such an index which will allow all the movements
> and operations you are wanting.
>
For deletions there are also methods that will do the job with O(n).
> Alternatively, when using Arrays, you can use #splice! and []=3D.
>
For arrays, there is little need for external iterators, since random
access is O(1). The array index itself can act as an external iterator and
be no less efficient than a full external iterator (which could access an
array element without needing the array).
Can you leak a bit more information about the scenarios where you need
> more / different behavior?
Here are a few applications I can recall:
* sweep-line algorithms. I've done several implementations and used
others. To have an efficient O(n*log(n)) one, these are some of the things
I use from sorted-tree structures and their iterator:
- multi set/map: allows for duplicate entries (to deal with multiple
coincident edges)
- find: gives an iterator to an item
- lower/upper bound: more specific find when dealing with duplicates or i=
f
the item doesn't exist yet
- ++/-- on an iterator: go to neighbor above/below
- erase using iterator: O(1) erase an item you've already found
- insert (w/ and w/o iterator hint): with an iterator hint, the insert is
O(1), w/o it is O(log(n))
* range/region query. Iterators can be useful for looking to neighbors or
inserting/deleting in O(1) time. Unfortunately, the STL (multi) set/map
typically isn't sufficient to to implement some of these structures. But,
an API similar to what's in STL (including external iterators) is quite
useful. Google libkdtree++ to see one implementation.
* MST (minimum spanning tree). Kruskal's can get away with just a heap, bu=
t
Prim's algorithm needs a deletions. Iterators may not be needed. This is
also similar to Dijkstra's shortest path algorithm. Also, building steiner
trees can use various sorted tree structures (don't remember if external
iterators are needed). Geometric MST's or steiner trees also take advantag=
e
of additional structures like sweep-lines or region query structures.
From my experience, many/most algorithms you implement on large geometric o=
r
graph (as in graph theory) data sets can take advantage of sorted structure=
s
and associated (external) iterators. I'm sure there are many other
applications when you are dealing with large data sets (where simple O(n**2=
)
or slower algorithms aren't good enough).
At one time Gonzalo Garramu=F1o made an C++ STL to ruby i/f using swig. I
also did something similar for python using swig and used it for some
sweep-line/MST/steiner. I only learned/used python because there was an ap=
i
into the db I was using (in addition to C++).
Note that I still would prefer using ruby compared to C++, but the librarie=
s
in C++ are more in line with what I do unfortunately (and I need C++ for
performance). But, ruby has clearly improved my C++. I use templates
regularly to get my duck-typing (arbitrary typing). I use both external
iterators and internal iterators (visitors) where appropriate in C++. And,
I think my general OO skills in C++ have improved because of ruby. The
syntax is very ugly and I have to deal with too many details, but the
majority of the concepts in C++ are there in some form.