Why to use LV/C++/C?

[Greg McKaskle]

There are so many ways to do the comparison. Years ago I read a book on comparing languages, and I thought their system was a pretty good one.

My disclaimer -- I work for NI, on LV. That means I'm a C/C++ programmer, and I also program in LV. My opinion is that they both have their strengths, and that is as it should be. In Atlanta, the question came up, "What is the most important language for the kids to learn?" There are many answers this, but my favorite answer is that the most important computer language for you to learn is your second one.

Learning a language gives you a logic tool. You begin to look at problems from the perspective of that tool. It is the "If you are holding a hammer, everything looks like a nail" psychology. Learning a second language, especially one that is considerably different, results in having a second perspective, and learning the tradeoffs and interplay. Hopefully the second language also opens the eyes to the possibility of other language features, perhaps ones that haven't even been invented yet. To keep with my analogy, learning about other fasteners such as screws, bolts, clips, adhesives, etc. really gets started once you change your point of view and realize that the hammer and nail, while they are amazing, aren't the only tools in the modern toolbox.

The book on comparing programming languages was "Comparative Programming Languages" by Wilson and Clark. No, it doesn't cover LV, but it does break down the elements that they think are most useful to measure when looking at languages for a comparison. It is also useful to see the different ideas that languages implement, and to watch that change over time.

Inspired by their categorization, I normally condense this to, Expressive Power, Simplicity & Orthogonality, Implementation, Error Detection & Correction, Correctness and Standards.

If I were to go into each of these for both LV and C, it would be a mega-post, and besides, I think it is better for people to do it on their own and come to their own conclusions.

As an example, C++ scores better on expressive power than C does. Why? They both have the ability for the user to declare new functions, new variables, and even new types. But C++ goes further to allow for operators like + and - to be expanded for new types, and allows for the -> to be used to dispatch to class member functions. C++ also has additional keywords. This power comes at a price, however, as the sorts of errors you can receive from the compiler go way up, and the overall simplicity goes down.

Finally, while most new users initially focus on the text versus graphics difference, this is pretty superficial. There could be a graphics C. Arguably you've already seen that. There could be a textual LV, but defining parallel graphs in text is fundamentally hard and we have instead headed down the graphics road. A more fundamental difference is that C/C++ are procedural languages or imperative languages whereas LV is mostly a functional language. These terms may not mean much, but they are the more important ones to look up and think about.

I don't have a great analogy, but that won't stop me from trying one anyway. An imperative language to me is like driving a vehicle with a manual transmission. There are discrete gears, combined in various ratios to direct the engine rotation to the wheels. You control this with a clutch, a shifter, possibly more if in a semi or farm equipment where I can control other differentials. Fundamentally I'm exposed to the implementation and am expected to manipulate the system at that level. Not all settings make sense, and grinding of gears, getting stuck in neutral, and stalling the engine are all common occurrences, especially until you learn the system quite well. But once you understand the system, it is quite powerful and adaptable.

A functional language in this analogy is more like an automatic transmission system. Fundamentally the big difference is that the driver specifies what the outcome should be, but not the intermediate details. It may be as simple as F-N-R and a pedal. You still control the velocity of the vehicle, but you have traded control for simplicity. When you inspect the automatic transmission system you may find that it is really a manual with an automated shifter, or you may find a CVT or hydrostatic where it isn't even easy to compare them any longer. A CVT isn't limited to a small number of gear combinations, so is it better than a limited manual transmission? Or is it inferior because of the limits on the amount of power it can transfer?

Different approaches to the same problem. Isn't it better to focus on understanding both than to be dogmatic about it?

Greg McKaskle

[slavik262]

Originally Posted by Greg McKaskle

"What is the most important language for the kids to learn?" There are many answers this, but my favorite answer is that the most important computer language for you to learn is your second one.

I wholeheartedly agree with you on this. My first language was Visual Basic.NET, and then through FIRST robotics I learned C, then taught myself C++. The first language really taught me how to "think like a programmer," but the second language really broadens your horizons.

Originally Posted by Greg McKaskle

As an example, C++ scores better on expressive power than C does. Why? They both have the ability for the user to declare new functions, new variables, and even new types. But C++ goes further to allow for operators like + and - to be expanded for new types, and allows for the -> to be used to dispatch to class member functions. C++ also has additional keywords. This power comes at a price, however, as the sorts of errors you can receive from the compiler go way up, and the overall simplicity goes down.

Agreed, but once you learn how to wield that power I have found that C++ (in my opinion) can be used to make programming much easier to understand. Object-Oriented Programming in C++ such as allowing functions to be members of objects make code much less abstract and easier to follow since things are gouped together. Yes, learning OOP can be very challenging because it forces you to think completely differently, but once you become well-versed in it I have found that it makes life easier. Which brings me to...

Originally Posted by Greg McKaskle

A more fundamental difference is that C/C++ are procedural languages or imperative languages whereas LV is mostly a functional language. These terms may not mean much, but they are the more important ones to look up and think about.

I'm confused by what you're saying by this. While C is procedural or imperative, C++ is Object-Oriented. Of course you can write procedurally in C++ just like you would in C, but by taking advantage of the object-oriented approach offered in C++ you can make your programs much less procedural, like I mentioned above.

Originally Posted by Greg McKaskle

I don't have a great analogy, but that won't stop me from trying one anyway. An imperative language to me is like driving a vehicle with a manual transmission. There are discrete gears, combined in various ratios to direct the engine rotation to the wheels. You control this with a clutch, a shifter, possibly more if in a semi or farm equipment where I can control other differentials. Fundamentally I'm exposed to the implementation and am expected to manipulate the system at that level. Not all settings make sense, and grinding of gears, getting stuck in neutral, and stalling the engine are all common occurrences, especially until you learn the system quite well. But once you understand the system, it is quite powerful and adaptable.

A functional language in this analogy is more like an automatic transmission system. Fundamentally the big difference is that the driver specifies what the outcome should be, but not the intermediate details. It may be as simple as F-N-R and a pedal. You still control the velocity of the vehicle, but you have traded control for simplicity. When you inspect the automatic transmission system you may find that it is really a manual with an automated shifter, or you may find a CVT or hydrostatic where it isn't even easy to compare them any longer. A CVT isn't limited to a small number of gear combinations, so is it better than a limited manual transmission? Or is it inferior because of the limits on the amount of power it can transfer?

Different approaches to the same problem. Isn't it better to focus on understanding both than to be dogmatic about it?

That actually turned out to be a beautiful example. The reasons I love C++ are the ones you mentioned above. At first, things don't make sense, don't work, and seem to vex you at every turn. But once you get the hang of it, you realize that you have an incredibly powerful and flexible tool at your disposal. The downside is that programs can take longer to develop. I personally don't mind this, however, because I enjoy working "under the hood" and being more in control (perhaps it's just because I'm obsessive-compulsive ). I would also like to take the chance to look at LV though to see how it works and what advantages it has, even though our team will be coding in C++ this year.

[Alan Anderson]

(C/C++ is procedural, LV is functional)

Originally Posted by slavik262

I'm confused by what you're saying by this. While C is procedural or imperative, C++ is Object-Oriented.

You should probably look up what "functional programming" means (try Wikipedia). Being Object-Oriented doesn't appreciably change C++'s imperative nature. What you've written seems similar to "While an orange is a fruit, a navel orange is seedless."

[slavik262]

Originally Posted by Alan Anderson

(C/C++ is procedural, LV is functional)


You should probably look up what "functional programming" means (try Wikipedia). Being Object-Oriented doesn't appreciably change C++'s imperative nature. What you've written seems similar to "While an orange is a fruit, a navel orange is seedless."

Ah. Thank you for clearing that up.

[DonRotolo]

Originally Posted by Greg McKaskle

There could be a graphics C.

There is, we call it "Easy C".

[lasereyes]

Originally Posted by Don Rotolo

There is, we call it "Easy C".

Will there be an EasyC option next year?

[Gdeaver]

Our team has lost all the programers to graduation. The returning group wants no part of programming. The are several 9th graders who have experience with the lego platform. If they join the team it will be interesting to see how thier use of NXT-G aids them in the programming of the FRC in lab view.

[Michael Hill]

I may be sounding like one of the "older" mentors on this board (perhaps I am? lol, no), but my first programming language I learned was Fortran 90/95. My second was ANSI C. Now when I first started work, I was fairly proficient in either language (really, both languages are good because knowing them, mostly Fortran, is a lost art). However, most of our programs are programmed in LabVIEW. A fundamental difference between these languages is that Fortran and C are "top-down" programming, whereas LabVIEW is a "data-flow" programming language. I do know one thing, LabVIEW does things MUCH easier and faster than C or Fortran (programming wise). For example, you have absolutely no pointers, no worry of memory allocation or anything. It is much easier to use, but it can be quite difficult making the transition for your brain to think in data-flow vs. top-down.

[Alan Anderson]

Originally Posted by Michael Hill

[With LabVIEW]...you have absolutely no pointers, no worry of memory allocation or anything.

That contradicts much of what I've heard about both the specific FRC tools we are told to expect (e.g. DYNAMICALLY ALLOCATED!) and the general style of LabView programming I learned a couple of years ago (e.g. "open" a resource and pass the resulting pointer along to everything else that uses it, remembering to "close" it when finished).

I look forward with a mix of anticipation and dread to actually having this stuff in front of me to work with. My optimistic expectation is that I'll quickly realize what simple concept is keeping me from completely understanding what I'm reading, and that I'll finally "get it". (My nagging fear is that I'm a programming dinosaur, stuck in a procedural tar pit and doomed to extinction as the dataflow mammals take over.)

[JaneYoung]

Originally Posted by Alan Anderson

(My nagging fear is that I'm a programming dinosaur, stuck in a procedural tar pit and doomed to extinction as the dataflow mammals take over.)

If I may, Alan, you will never be a dinosaur of any type, stuck in a tar pit of any type, and doomed to extinction - because you are sticking with it and will see it through - evolving: curious, asking questions, seeking to understand, comprehend, gain skill sets and how to apply them. I think that is a big part of advancing technology and science in any area.

Your written thoughts created an incredible visual.

[Michael Hill]

Originally Posted by Alan Anderson

That contradicts much of what I've heard about both the specific FRC tools we are told to expect (e.g. DYNAMICALLY ALLOCATED!) and the general style of LabView programming I learned a couple of years ago (e.g. "open" a resource and pass the resulting pointer along to everything else that uses it, remembering to "close" it when finished).

I look forward with a mix of anticipation and dread to actually having this stuff in front of me to work with. My optimistic expectation is that I'll quickly realize what simple concept is keeping me from completely understanding what I'm reading, and that I'll finally "get it". (My nagging fear is that I'm a programming dinosaur, stuck in a procedural tar pit and doomed to extinction as the dataflow mammals take over.)

A lot of things ARE dynamically allocated, most noticeably, arrays. HOWEVER, the biggest difference is that all the dynamic allocation takes place behind the scenes in LabVIEW. The end-user of LabVIEW, i.e. the programmer, i.e. you, never has to worry about using malloc, calloc or anything like that. The only type of thing that I can think that you should have to "close" would be a TCP connection in LabVIEW. You don't have to "close" any other references or anything like that, I least I never have had to in the years I've been programming.

[slavik262]

Originally Posted by Michael Hill

For example, you have absolutely no pointers, no worry of memory allocation or anything. It is much easier to use, but it can be quite difficult making the transition for your brain to think in data-flow vs. top-down.

Aren't there certain advantages to manually managing memory though? Perhaps I'm also stuck in the procedural tar pit (although at age 16 I would hope I'm not a dinosaur ). I realize that LV may provide faster development, but doesn't C++ provide you with slightly finer control?

[EricVanWyk]

Originally Posted by Michael Hill

For example, you have absolutely no pointers, no worry of memory allocation or anything.

I am not a software engineer. I am not a LabVIEW expert (barely proficient is stretching it). However, I have (at some point) used more than 10 programming languages, and I have spent enough time with LabVIEW to become frustrated and (later) elated.

I will defer to Greg as to whether or not Michael's assertion is true 100% of the time, forever and ever. I simply don't know.

What I do know is that typical LabVIEW code reminds me of LISP/Scheme in how it handles basic data types. The Liopleurodons among us may benefit of approaching it from that angle - that was the mental leap that made everything "click" for me. Now I think of LabVIEW as Graphical PythonLISP.

To over generalize:

Python -adjective
1) readable and approachable
2) utilitarian / lacking blind devotion to a particular ideology.
3) pretty

I will admit that "pre-click", LabVIEW was frustrating. I kept trying to force imperative programming methods which just weren't appropriate. Now I am rather excited to use it for `09 (especially since I haven't mentored a programming team in FIRST in 2 years).

[Greg McKaskle]

Several things to comment on.

First -- Eric gets it! Yay.
LV is different, not always better or always worse, but because of dataflow, almost always different. It really isn't magic either. Like most industry languages, LV is a mixed bag. A functional core with grafts to allow easier looping and structured programming, more grafts to allow for I/O APIs that use references, and of course the implementation is still missing many of the things that we want it to have.

Second -- Michael is correct.
LV isn't an ideal realtime language, because it does do allocations implicitly. Some languages alloc almost everything in sight, others make it very explicit. LV has many scalar types, or flat types as we refer to them that are statically allocated, but the string and array in particular, are dynamic. There are in fact bounded and fixed variations of them, which will give you better determinism, but they are not used that often. More on LV memory model later in the post.

Third -- I agree with slavik262.
There are certainly situations that benefit from tightly controlling memory allocation as well as other resources. But not only is there a tradeoff in the number of details you need to take on for explicit control, but the power of C's syntax in particular is way into the "user beware" camp. Is x=rand(); *x=5; ever what you meant? The time spent finding an outright nasty memory corruption is quite the tradeoff to make.

Finally -- Alan, you don't sound like a dinosaur, but perhaps a skeptic. I find skepticism to be a critical element in making new things work.

Getting back to the LV memory approach.
I'm not sure this statement will help, but the primary LV types have by-value semantics, even if their implementation is by reference. What this means is that each wire acts as an unnamed temporary storage element. It has one writer and as many readers as you like. So if a wire simply runs between the output of + and the input of =, it is really just binding those together functionally. If the wire forks and delivers the data to multiple inputs, there are no side effects to worry about. The data value is delivered to each. It is impossible for another location on the diagram to modify the wire, no matter what order they are scheduled in, how long it takes to execute, etc. This doesn't seem like a big deal at first, but in a parallel language, it is a cornerstone. This is why variables, global or local ones weren't even in the language for quite some time, and why they need to be used very carefully. Multiple writers allow for race conditions, and the parallel dataflow scheduling makes the race conditions a common occurrence.

If the actual implementation used a by-value approach, LV would be quite a bit slower and bigger than it is. Instead, the compiler analyzes the graph and allows for temporaries to be folded together. It also takes statements that are more parallel than needed by the computer architecture and serializes them into a statically scheduled snippet to help further with memory reuse. Finally, arrays and string wire forks mean the same as a Boolean, but their implementation is different. They are really alloc'ed buffers, and the buffers are commonly shallow copied to improve efficiency, but not at the cost of simplicity to the user.

The exceptions to the by-value semantics are the reference types. Primary ones are I/O such as TCP, files, and UI elements. Copying an entire file around with by-value semantics just wasn't feasible in '83 when they started working on LV, so some APIs expose a reference type which should be closed when completed, similar to most POSIX libs. But trying to improve it slightly, LV does shadow these refs and will close them at program completion or termination, even if you don't.

LV certainly isn't magic, and I'll be happy to explain how things work if you ask the questions.

Greg McKaskle

[EricVanWyk]

Originally Posted by Greg McKaskle

First -- Eric gets it! Yay.

LabVIEW is like the big connectors on the PD: A total pain until you figure out the right way to use them, at which point it is like a godsend. Remember, Slide Not Stab!