Slashdot Mirror
Can OO Programming Solve Engineering Problems?
"I appreciate the concepts of OOP and see its applicability in managing records, GUIs, and possibly standard function libraries. I cannot, however, convince myself that there is a clean way to use these concepts to solve the type of procedural problems that I have encountered in the past (finite difference solutions to differential equations, finite-volume computational fluid dynamics, iterative solutions to non-linear equations, Monte-Carlo simulation of radiative heat transfer, etc.)
Am I just being close minded to the ideas of OOP or do my problems just require 'procedural' solutions, which are better solved using procedural techniques? I'll even be happy with the answer 'Your problems are two small and specialized to realize any significant advantages of OOP.'
I'd be interested in hearing comments from anyone else who has this problem, anyone who has worked through it, or anyone who can send me an example of an engineering application of C++ and OOP."
Your problems are too small and specialized to realize any significant advantages of OOP.
This is a really difficult question to answer. I guess my reply would be that no, pure "OO" programming wouldn't hold any advantage to simple procedural programming in this case. However, you can easily add procedures to objects, so OO language is really just a superset of procedural.
Personally, I like the organizational potential of OO. You can have a Pipe object, and call function like Pipe.calculateFlowrate(12,43,23), then you could have a subclass SquarePipe, and call the function SquarePipe.calculateFlowrate(12,12,43,23).
Essentially, you could have 10 objects each with 10 functions to do 100 different tasks, rather than 100 different functions to do 100 different tasks.
Moderation: Put your hand inside the puppet head!
There are a number of ways to approach this, but I've found the following useful:
It's more a matter of thinking in terms of telling your processes to do stuff than creating a road for them to walk down. If you know what I mean
All your Qaeda are belong to US
C is a procedural language, not a functional language. Examples of the latter include LISP, Scheme, and ML.
As to whether an object-oriented approach is suitable for engineering applications, I think it depends on the applications and problems. How important is abstraction and re-use? If you deal with matrices, would you benefit from a matrix object which implements all the properties a matrix has? With differential equations, are there objects suitable for implementation?
As with all general questions I am afraid the answer is "It depends."
OOP doesn't solve engineering problems of the kind you describe. The programs solve the problems.
OOP is a way to think about structuring the program, how you organize data and operations on the data, how you build reusable components which you can apply in some future problem you need to write a program to solve. In other words, OO helps solve software engineering problems. You'll still need to write the program which solves your particular problem.
And before the flames commence, yes, OO isn't the only solution to this class of problems. It's just one.
Take a look at BLITZ++, then tell me OOP is not useful for scientists and engineers. :-)
I think the safest thing to say is that whatever your programming needs, whether you're doing pure matrix/BLAS number crunching or writing complex simulations/models, you should think twice before using FORTRAN. Well-written code in, say, C++ will be more maintainable and accessible to other people you work with (and who have to touch your code in future).
The only thing which keeps people using FORTAN that I've seen is that the optimising compiler support is fantastic compared with the equivalent offerings in C/C++ compilers. But that's not much of an excuse for general day to day problem-solving.
Just my 2p.
These sigs are more interesting tha
First off you're examples Monte Carlo simulations and differential equations aren't "engineering problems". They are math problems that are a component of an engineering problem. Engineers does things. It builds a bridge, drives a car, prints a document. The examples you give are parts of that solution. Every GUI needs to do things like interpolate colors and position, which are the same class of problems -- if orders of magnitude simpler -- as differential equations and Monte Carlo simulations. So it seems to me that you need to figure out what your real problem is and figure out what programming paradigms might help you out with that.
As others have pointed out OOP doesn't let you solve things that are otherwise unsolvable. It is a way of solving problems that may be better in some situations and worse in others. The examples you give of solving equations are all better suited to a functional language than a procedural one like Fortran, so rather than asking "why use OOP" you might want to first ask yourself "why am I using Fortran?" Obviously there are a number of factors that go into how we pick our solutions. Maybe OOP per se isn't a good fit for your problem but you need it to be multiplatform and have a huge amount of available code and standard libraries so you end up going with Java. Or whatever.
The kinds of problems you're talking about don't have a great mapping onto the traditional ways of describing OOP. However, OOP is really just a somewhat formalized kind of way of dealing with abstraction and data encapsulation. You can make a difference equation a class. Maybe it'll only have one method that immediately finds all the finite solutions. But once it is a particular datatype you can also do things like compare whether two user-entered difference equations are identical and just returned the cached solution. You can curry them. You can return partial solutions and then come back later and ask the object for more solutions.
Don't you already have discrete data types for these things? And don't you already have functions that operate on those discrete datatypes? Then you're already doing OOP. Sure you're not using inheritance and multimethods and things like that. But not every OO program does.
OO languages exist at the "high" end of the language-level spectrum. They're geared toward managing code complexity in the face of a problem domain which is conceptually complicated, primarily by encapsulating bits of the problem domain into digestible and self-contained sub-problems. The overhead of all of the OO constructs is worth it if the reduction of your problem domain into smaller chunks is neccessary to solve the problem (or advantageous in terms of directing the efforts of multiple team members in areas where some decoupling is possible).
However, if you problem is "low-level" or conceptually simple (though not neccessarily computationally simple) -- a recipe like "apply transformation x to dataset y, then transform again w/ algorithm z", the OO features simply serve as a distraction from thinking about your actual problem domain and it's solution.
So yes, IMHO, there are problems for which OO techniques are not ideally suited, and most importantly, if the techniques get in *your* way they are not the right tool for *you*. Rememer, languages and tools don't solve problems. People do. If a tool makes you task easier, use it. Otherwise, save yourself the time.
For simplicity's sake, let's think about quadratic equations for a second. You can solve them easily, but if you want to use them in a larger program, you could create a QuadEq object in OOP with the following properties:
An OO programmer would then add methods to set, retrieve, and calculate those properties based on what's been entered. And the QuadEq object would be entirely portable and easily amplified for future equations.
I don't think choosing OOP is a matter of being the only tool to solve certain problems. However, it is often the most efficient way to solve large, rapidly-changing problems. But like you said, other problems (like many of the ones I encounter in web development) will be small and uncomplicated enough that the overhead of OOP isn't worth the trouble.
I've seen some truly awful procedural code (lots of it was Fortran, BTW). I've seen some truly gorgeous procedural code (lots of it was Fortran, BTW). I've seen some some wonderful, and even more pretty awful, OO code (mostly C++, but with some Java).
Go ahead and study object oriented programming. You'll learn some new ways to do things. But I think it's the act of studying, and the act of learning, that will be the most valuable thing you get out of the process. Too many people never study how to program, how to document, how to design code. They learn one or more languages. Their code shows it.
A few people have a natural tendency to write elegant code. A few more, but still not very many, study and try to learn how to write elegant code.
But don't expect the object-orientedness to make much difference. A dozen or more years ago, a young whippersnapper got hooked on objected-oriented design. He derided all the existing Fortran we had as spaghetti code. To some of us, though, his "object-oriented" code was lasagna code. No overriding structure, code spread out all over the place, a single function scattered over three files. And this was still Fortran; I've seen C++ coders who took six files for a similar, simple function.
I've also surprised myself! when some of my OO C++ code needed four lines to add new functionality. But it was carefully designed, after years of programmer improvement and study.
Go ahead and try it; it can only help.
I think the "answer" to your question is that, yes, your examples are too simple. When you're doing some simulation where the "finite difference solutions to non-linear equations" all interact with each other, and with computed fluid dynamics and heat transfer, etc, you'll immediately see the benefit of separating all these concerns into encapsulating object types.
[
Imagine for a minute that you don't know what problem you want to solve yet. You know that that you want to apply a Galerkin Finite Element Method (for instance, though this particular method isn't required) to a whole class of problems on unstructured grids on a whole class of distributed and shared memory parallel computers. Imagine that you want your user base to be able to specify their equations like they would in LaTeX or some other markup language. Now try imagining that you have only FORTRAN77. Not a pretty picture. We're in the process of completing a rewrite of major sections of our parallel code to do exactly this. Our code started out (7 years ago) as an extremely efficient parallel (3D) C/F77 code for Navier-Stokes + Heat Transfer and is quickly growing into a multi-purpose, multi-physics code written largely in C++.* We extract considerable advantage form C++'s ability to hide implemenations so that as long as interfaces don't change the guts can. We also make good use of the ability to run code before main() in order to register the exisitence of routines (hash tables are your friend). If the routine isn't there you can't call it, but the code still compiles and runs otherwise. We also make use of base class/derived class relationships and polymorphism to allow, for instance, one base mesh class for the rest of the code to interact with, but with two separate derived classes: one to generate meshes internally, and one to read meshes from other programs. Etc., etc. I'm not sure our website can take the /.'ing, but you can look here for some hints.
* I say largely b/c there's a few struct's still left over from the code's C days, but all the F77 is gone. There are still calls to assembly coded (vendor supplied) BLAS routines, though.
Yes...I am a rocket scientist.
Preface: I've been an embedded systems software engineer since 1987. I've been programming in C for almost all of that time, and C++ since '91. I've designed systems with over 250kloc, hard realtime.
The biggest single advantage of OOD is that you can say "This sort of thing will have a function that does something like this, but I won't make the final decision on exactly HOW to implement that function until late, when I have more information on what needs to happen."
Let me give you a concrete example. I had to design the code to reprogram the flash in one of our devices. I faced the following problems:
1) the way the data got to us could either be Xmodem or Ymodem.
2) The file format could be either OMF or raw binary.
3) The types of memory devices could change - they could be Intel type 1 flash, AMD type 1 flash, or Intel type 2 flash (or RAM), all of which have different programming requirements.
3a) In addition, the types of flash could be changed by the production line (based on what was available - most of those parts were "on allocation", meaning "you take the quantities we give you and you like it - you ain't IBM so you don't matter".
3b) The types of flash on a given unit could be mixed - you had to probe at runtime to figure out what you had.
So, here's how I decomposed it:
1) I had two file type objects: OMF and Binary. They each HAD-A Input object, which provided data , and HAD-A Memory object, which would program memory. The OMF object read OMF records from the Input object, parsed the OMF records, extracted the data, and commanded the Memory object to program regions of memory.
2) I had an Xmodem object, which WAS-A Input object. It read data from the serial port, handled the checksum or CRC verification, block counting, and made the data available to the user (the file object).
3) The Memory object HAD-A pointer to a Memory_driver object. When the Memory object was commanded to write to a block of memory, it verified that the write was in the current block, and if so passed the request to the Memory_driver object. If not, the Memory object HAD-A list of Memory_driver_list objects, which each HAD-A pointer to a function that would probe a given memory address and return either NULL or a pointer to a newly created Memory_driver object. The Memory object would iterate through the list, asking each Memory_driver_list object "Can you program this?". When one of them returned a non-null value, the Memory object would delete the old Memory_driver, and use the new Memory_driver.
4) I had Intel_series1, Intel_series2, AMD_series1 and RAM objects, all of which WERE-A Memory_driver. Each class HAD-A static instance of a Memory_driver_list object, which automatically linked itself to the driver list. Each driver HAD-A static member function to probe memory and return a pointer to a newly created driver if needed.
First, this let each routine be focused on what it needed to do - the Xmodem routines didn't worry about OMF format, the OMF didn't do Xmodem handshakeing, the Memory routine didn't care about the specifics of programming memory. I could test each object out in isolation, get it working, and move on. Now, you can do this with proceedural programming techniques too.
Second, when a new type of memory was added, I was able to just write the driver for it, and not touch the OMF object, the Memory object or the file transfer objects. They automatically picked up the new driver. Now, you can do this in a proceedural language like C, but how do you do so? You make each driver have a table of function pointers, and you have the upper level code keep track of a void * that contains your driver information. Guess what - that's OOP! The function table is the same as a C++ vtable, the void * is your this pointer. Except that in C, you have to track all that stuff yourself, in C++ you can let the compiler worry about the BS and you focus on the code design.
In short, OOD helps, but you still have to use it correctly.
www.eFax.com are spammers
I have always thought of OOP as fundamentally a procedural programming technique but extended to give you better ability to organise and manage larger projects. C++ is a superset of C - you don't have to use the OOP features but they can help you enormously in the right situation.
:-). In this case, all the code that used the object would remain the same, possibly saving you a load of work.
OOP also encourages good programming practices like producing well-defined interfaces, implementation hiding and code re-use. I actually think these are crucial to the effectiveness of OOP. These are all good things, and I'm sufficienctly lazy that I would never write anything other than a trivial program in purely procedural style.
However, OOP is still best suited to particular domains. Modelling, GUIs, enterprise applications, games etc. are perfect because they all require the manipulation of discrete and readily identifiable "objects" that map onto the OOP model very well.
If you're writing network drivers, you could use OOP, but as other posters have been keen to point out the (very small) overhead of most OOP languages actually becomes a problem for very low-level code. Hence OOP isn't so well suited for OS kernel code, network protocols, hand optimised inner loops and embedded applications.
The difference between procedural and OOP is mainly a trade-off between low level control of the machine and having a semi-automated system to manage the complexity for you. Interestingly, for larger projects, good programmers in procedural langauages often end up having lots of function calls that take the "object to be acted on" as an argument and are therefore effecively emulating method calls. Once they start putting in switch statements in those functions to branch on the type of object encountered, they have effectively emulated virtual methods.
Your case is an interesting one. You are dealing with mathematical/engineering problems that are often tackled in a procedural style.
If you're feeling adventurous, I would actually suggest you head off to look at functional langauges. Reason is that these langauges are ideally suited for representing functions and other mathematical constructs, just as OOP is geared towards representing and manipulating "objects". Haskell would be my no.1 choice if you're looking for elegance and flexibility, O'Caml if you are more concerened about raw performance.
If these seem too radical a departure, then I still think that OOP could make sense for a major project. You probably wouldn't need inheritance, but implementation hiding can be very useful. Let's say you're modelling fluid dynamics, and do it procedural style with a big array. Works fine until you decide that in fact you want to store the data in an octree so that you can get a couple of orders of magnitude better resultion in the particular areas you are interested in. At that point you have to change everything that accesses the data structure, including the 100 or so simulation programs you've been developing. Not fun. If OTOH you had encapsulated it all in an object, you would just change the object's private implementation, keeping all accessor methods the same (e.g. getFluidVelocity(x,y,z), getPressure(x,y,z) or whatever, I'm not an engineer
It's also often useful to have data represented as objects just so you can do useful things like pass the around, build larger data structures etc. Not that you can't do this in procedural style, but OOP takes away all the pain. I've lost track of the number of times I have taken advantage of all the useful features like object serialization in Java, which saves you all the hassle of having to write import/export functions to store your data structures on disk.
Ultimately, use what seems right for your application. But remember that OOP and other techniques such as functional programming aren't just fads, they are ways to solve problems that procedural style just isn't well suited for. If you find yourself spending too much time writing the "glue" to hold a big application together, then it's a good sign that you've actually picked the wrong tool.
First of all, for those individuals who refer to C++ as an OO language, please stop. You're wrong. C++ can be used for an OO project, but it is a multi-paradigm language. At least that's what Bjarne Stroustrup calls it. But what does he know about C++?
Second, use of C++'s STL does not equate to OO programming. It is an example of generic programming. Here's a hint: the STL has very little inheritance except for iterators and iostreams -- most evaluation is handled at compile-time. And even iterators and iostreams are just as much generic as OO.
Finally, please dispel the rumor that C is automatically faster than C++ because of C++'s excessive overhead. Need proof? Please read this article about treating C++ as its own language and not a variation of C. Yes, it's a PDF. Get over it.
Think the article is FUD? Prove it! Take the examples from the article and tune them better than he did. Compile them with trusty ol' gcc and g++ on your box. Measure the results. After you do so, can show that C is faster, does not contain any potential buffer-overflow bugs, handles error conditions, and wasn't at least five times more code to do it, then reply back with your results. I have a feeling I won't be getting any replies from people who actually try it.
That said, use whatever language you like best. Studies have shown that people will always perform better in languages they know intimately well than languages in which they have a general familiarity.
But if you want to use OO and C++, check out this numeric library
Have a nice day
- I don't need to go outside, my CRT tan'll do me just fine.
- small problems for which run-time is not a problem; and
- compute-intensive problems for which run-time is important.
For the first of these, an interpreted prototyping language with an embedded symbolic processor (like mathematica) may well be the best bet.For the second kind of problem, optimization can be of critical importance. It is inherent in the language that Fortran is more optimizable than C: due to the rules of the language, alias analysis in Fortran is a (relatively simple) polynomial-time problem. At present (though there is effort being put into the C 2X standard to try to change this), alias analysis for C and C++ is a NP-complete problem. And given the current processor trends (deeply pipelined superscalar processors running much faster than their memory interfaces), that gives Fortran a factor of 2 performance edge (at present; this promises to become even more marked in the future...)
That doesn't mean that you can't write object-based software in Fortran; I've been doing that for many years, on large scientific modeling codes. Bertrand Meyer's seminal book (The Construction of Object Oriented Software? -- not sure of the title; I'm snowed in at home and my copy is at the office :-( ) even has a section on writing object based Fortran.
Based on two decades of experience in computer modeling, my advice to ThChalm is:
And frequently this attitude will cause you to re-think the problem, and use far better algorithms than you would have otherwise.When I applied these ideas to various problems in the high-performance environmental modeling problem domain, I wound up with an emissions model that is 600 times faster than its fine-grained OO predecessor, and an atmospheric chemistry/transport model that is 10 times faster than a competing naively-written procedural-style model.
"My opinions are my own, and I've got *lots* of them!"