|
pasztorpisti wrote: Then we agree that C/C++ has obvious defects that shouldn't exist with todays
technology.
I didn't say that.
|
|
|
|
|
Well, up until now we basically havent listed any useful stuff for the OP about C/C++ benefits/defects. We just continued a totally unneded flame war that started from my statement about C/C++ headers so now I'm going to give a random list of some useful stuff about defects without attempting to be comprehensive because that would be too long.
Today it would be quite reasonable to use a "fixed/repaired" version of C without most of its defects - the same is true for C++. Of course no language is perfect but the number of C/C++ defects are unreasonable in a modern widely used general purpose language. There are several obvious bugs just alone in C that are directly inherited by C++. My favorite is: The Most Expensive One-byte Mistake[^]
The simple C language alone allows for a lot of programming errors most of which could be eliminated just by making the language more strict. I link a list of C programming errors that can be committed easily and accidently: http://www.andromeda.com/people/ddyer/topten.html[^]
Previously I mentioned header files that started a dispute. Some people would argue that headers are good because they separate the interface for reading. I think this is rather a personal preference - with todays tools you can esily extract such information on the fly for languages with better design like C# and Java as well where full runtime type information is available in the ide basically all the time. Even if you prefer header files, there are other languages that implement it in a much better way, I would compare C/C++ headers to assembly text file includes (containing full fledged C source code). (My favorite and silently compiling include related bug was when I found an anonymous namespace in a header, imagine what happens in this case...). Declarations in headers are not guaranteed to match cpp content if a mistake is made not to mention hidden dependencies. I would even question the readability of C++ header files that contain not only public/protected stuff but private and implementation details too (inlined functions -most notably templates - and data members are of course separated from actual implementation of the class methods). Headers and macros held back the development of C/C++ diagnostic tools a lot, this is why some modern (younger) languages has much better ide support even today.
Of course C++ inherited almost all mistakes and because of this immediately developed some necessary new ones. It also has its own unnecessary list of mistakes as well.
I would mention something that became immediately evident to me when I started using C++: compile speed is unreasonably slower than that of other languages even if we take the compile time of just individual compilation units in a relatively small or mid-sized project. One good comparison to this was when I used Borland Delphi and its stepbrother C++Builder that has the EXACT SAME ide but with C++ language. The compile times were so annoying in the C++ version that made iterations much worse despite the fact that I was tempted to switch to C++. C++Builder wouldn't be able to compete with Delphi even if we had the choice to use one of the best C++ compilers of the last decade and speed is gold when its about fast iterations and development time. Delphi compiled objectpascal so quickly you didn't even notice it - you modifed code, pressed the debug button, and half or one second later you saw the window of your debugged application to open - this is priceless. My favorite is when someone comes with the resoning that "C++ optimizes quite well, that takes time...". Other compilers optimize as well. This slow compile time has much more todo with reasons listed here: http://www.drdobbs.com/cpp/c-compilation-speed/228701711[^]. Please skip the first 3 links in the article becauase I think that has advertising purposes.
I'm open to any reasoning against the above list.
Another thing I simply can't understand: how does it come that you built large systems and you just say it doesn't matter what langues/tools you are using and most of the time the decisions are subjective. I completely disagree with these two points. I know that sometimes you are told to use XYZ technology becuase the customer wants it for this and that reasons (or just because there is a fashion wave and everybody wants to use XYZ technology to solve a problem beacause now that is cool) but this isn't always the case. In my opinion tools (including dev envs and languages) are very important. Depending on the type of project it often whort investing the time even to develop some additional project spcific tools because in time (and at the same time in cost) it often pays off well. At the same time if you automatize something with tools then you eliminate a lot of human errors that can increase the quality of the product a lot. You mentioned the importance of a regulated development process - I think every company that continues other than garage development has some kind of dev process, but that is another dimension of the problem of cutting development time and providing quality on a completely different level: management. The dev process is the tool of the manager, the language/ide is the tool of the coder. Without properly chosen tools both dev time/cost and quality suffers for sure but on a different level when using a bad software dev process.
modified 24-Sep-12 21:23pm.
|
|
|
|
|
C headers provide portability.
There are LOTs of C compilers. Headers provide an easy way to build a C compiler for any PIC, ARM, X86, (name your own) architecture.
It's not about the developer. It's about the CPU/Memory. About the C/C++ to machine conversion.
Headers give a lot of headaches to developers, but developers pay that price to see their code run "everywhere" and not have to rewrite ALL the code every 2 years (for other architecture).
Now you tell me that developers should have better tools. And they have. They have a lot of other languages that don't keep the "header" problem.
The slow, very basic, error prone compiling.
But that languages don't have the portability of C/C++ headers.
Portability cannot be enforced. It is the result of ease of implementation of a language for an architecture.
C/C++ compiler based approach is unmatched in this area.
it´s the journey, not the destination that matters
|
|
|
|
|
ErnestoNet wrote: C headers provide portability.
Not exactly sure what you are claiming but as stated it is wrong.
C, the language (not just headers) provides some portability but it is not guaranteed nor even easy and most of that is achieved through the standard libraries.
ErnestoNet wrote: C/C++ compiler based approach is unmatched in this area.
No idea what you are talking about.
Line for line, Java is more portable than C/C++.
C/C++ both have features that are specifically not portable in comparison to Java.
|
|
|
|
|
Headers provide an easy "standard" to build a compiler from.
If there are more compilers, there is more portability.
Java is not more portable than C. Java is (that I know) very portable, just not more portable than C.
C has compilers built from a lot of companies for a variety of products (PIC, PIC32, ARM, X86, etc). Java mostly from Sun for X86.
it´s the journey, not the destination that matters
|
|
|
|
|
It's not the C headers that provide portability. Nothing proves that better than the existence of crossplatform languages without headers.
|
|
|
|
|
Text based compilers are easier to write that binary based compilers.
The headers approach is easy to build a compiler from.
C is a very simple language to write a compiler.
There are LOTS of C compilers.
More compilers provide better portability.
Portability is also usually the result of compatibility. So the reason why the headers approach doesn't change...
There are not many cross platform/architecture languages.
Name a crossplatform language and I'll compare that to C in terms of portability....
it´s the journey, not the destination that matters
|
|
|
|
|
If we speak of writing a compiler that will be used to compile thousands of other programs, then its worth investing time to make the compiler as good as possible otherwise you will pay the price as many times as you compile/develop a program or more importantly a useful ide for the language.
I've never argued against the portability of ansi C, lets consider that as a positive trait of the language but that has really nothing to do with the header files. For example freepascal is crossplatform and it has no header files, the same is true for java+jvm and .net mono + its languages that have quite extensive and unified crossplatform library compared to that of ansi C.
|
|
|
|
|
Consider that C can work in microchips with very little memory and a very reduced CPU instruction set (for example http://www.microchip.com/[^]).
Java and .NET are very different that C. They are "platforms", not languages. Java and .NET runtime is in the 20Mb+ size libraries. Latest versions don't support old operating systems.
rt.jar is 42.6Mb (compressed!!). Java and .NET strings are inmutable, etc.
I'm not defending C, but most modern languages aren't languages, they are frameworks.
And C/C++ are few of the languages out there not owned by a corporation (.NET-Microsoft, Java-Oracle).
C/C++ do not target app developers. They target compiler developers, OS developers, driver developers and libraries developers.
I guess thats why C tops TIOBE (and C++ is 4º).
it´s the journey, not the destination that matters
|
|
|
|
|
Thats true, I'm still sad about how much better could C be (including the related devenvs) without losing its current power. I also treat both java and C# as fully fledged languages, they are just higher level (that is quite ok in the context they are used) and they aim a different target. Those languages provide very good development environment and incorporated very good ideas related to language design.
|
|
|
|
|
The problem with C (and C++) is that they have to be compatible with older code. And they are too flexible (too much for my taste).
Don't think of high or low level.
Templates in C++ are higher level than C# and Java generics (for sure).
Think it this way. C/C++ is compatible with new and old stuff. ansi AND Unicode. Asm, procedures, objects and templates.
Don't be sad for C/C++.
They will outlast C#, Objective-C and Java (Java has started to decline and C again tops the tiobe index).
You ask what makes C/C++ "good".
Speed, portability, flexibility, huge codebases and an absense of an owner.
But, most of all, a compromise of compatibility, no matter how complex the language becomes, how outdated the structure is. New features will be added, old features will not be replaced.
To be "better", C/C++ should be incompatible with older code or fork. Luckily that won't happen. It will sacrifice simplicity and ease of use for compatibility.
Oh, and C/C++, in certain cases, can be MUCH faster than C#/Java. I have an article where C/C++ is 10x faster that well written C#/Java.
it´s the journey, not the destination that matters
|
|
|
|
|
ErnestoNet wrote: Oh, and C/C++, in certain cases, can be MUCH faster than C#/Java.
And in certain cases I am faster than C/C++/C#/Java and any other language that you care to name.
Not to mention all of the functions that I do that no language can accomplish.
But since people don't pay me to digest lunch but they do pay me to write large systems I can be safe in the knowledge that the flavor of the month language (even when it is an old flavor) will have little impact on the performance of the systems that I write and will not make me more or less productive either.
|
|
|
|
|
It's good to know different languages and try to understand why they are the way they are.
Even more, it's good to know about different technologies (SQL, COM, OLAP, EJB, Corba, XML, JSON, CSS, HTML, etc).
It's not about being good, but keeping the right attitude towards developing knowledge.
People that know a lot about developing are the ones that admit they know nothing and keep learning, with an open mind.
About the pay...that usually involves other skills that have very little to do with technologies or programming... : http://www.theregister.co.uk/2012/03/21/how_to_get_paid_more/[^]
it´s the journey, not the destination that matters
|
|
|
|
|
ErnestoNet wrote: It's good to know different languages and try to understand why they are the way
they are.
Which is why I do in fact know quite a few.
ErnestoNet wrote: It's not about being good, but keeping the right attitude towards developing
knowledge.
People that know a lot about developing are the ones that admit
they know nothing and keep learning, with an open mind.
And know how to look for objective data as well.
|
|
|
|
|
ErnestoNet wrote: Consider that C can work in microchips with very little memory and a very
reduced CPU instruction set
You are wrong.
You are confusing language, compilation and executable. One uses a C compiler to create a program where the executable then runs on the chips that you cite.
compiler != program != executable
ErnestoNet wrote: Java and .NET are very different that C. They are "platforms", not languages
Wrong.
Java and C# very specifically are languages.
I would guess that you are unaware of real time Java which is specifically targeted at embedded devices.
You are confusing language with libraries and you will NOT be able to write a C program that does anything discernable without the C libraries. And it is quite possible to write a C program on windows that would never run on the chips that you cite.
ErnestoNet wrote: I'm not defending C, but most modern languages aren't languages, they are
frameworks.
What exactly do you think that the C++ Standard Library is? Where do you think 'printf' comes from in C?
ErnestoNet wrote: C/C++ do not target app developers. They target compiler developers, OS
developers, driver developers and libraries developers.
I guess
thats why C tops TIOBE (and C++ is 4º).
Nonsense. The people who write compilers, drivers, OSes and libraries are a small, small fraction of the developer base. If those were the only people using it then it would be far down the TIOBE list.
Developers creating business applications use it. That is why it shows up there.
You would be better off citing embedded development but there is also C++ and Java development in that domain.
|
|
|
|
|
It's not about being right, but if you want to be right, there you have it:
YOU'RE RIGHT!
That said, the STL, if you don't use it, you don't load it. The header approach only loads (and compiles) what you use.
In fact, when you compile and link, only the functions that you use are linked. In Java and others, you load everything (even if you don't use it).
About Embedded development, I know a lot of people that work with it and they work mostly with Assembler and C. Java and C++ may exist, but they are not used.
Today embedded has changed with ARM cellphones.
A 2Gb memory dual core CPU cellphone is closer to a PC that to embedded. There you can run Java, as in the PC.
A real, cheap embedded chip with 64Kb of memory and a few cycles processor can't run (with decent performance) Java or C++.
Java embedded only runs on ARM to start with. It requires at least 130Kb (with tweaking) and 700kb
by default. It requires a network connection! It does not support real time. You can check everything here:
http://www.oracle.com/ocom/groups/public/@otn/documents/digitalasset/1852008.pdf[^]
I'm going to say it again, just because Bjarne Stroustrup said it and I think he is right:
Java and .NET ARE PLATFORMS.
it´s the journey, not the destination that matters
|
|
|
|
|
ErnestoNet wrote: In Java and others, you load everything (even if you don't use it).
Wrong. In a number of ways.
First both C and C++ in modern compilers commonly rely on shared libraries. Modern OSes load the entire shared library even if one method is used.
Static linking although possible isn't normally used on desktop OSes and there is no reason to expect that Java/C# would need to do anything different on a desktop OS.
In contrast Java and C# do something similar but, at least for Java, that is done solely as a run time performance optimization. One can create a stripped down version of the library is one wants. And one can also do a load only on use case (at least in java) even if using the standard API jars. I suspect the same is true for C#. And that is true for a desktop OS.
If one has an embedded Java SDK then there ware going to be substantial differences when compared to a desktop OS. But the same thing apples to using C in embedded systems as well.
ErnestoNet wrote: Java and C++ may exist, but they are not used.
If it wasn't used then there wouldn't be companies creating compilers for just that purpose. Googling provides that. A company selling nothing but compilers wouldn't be viable if there wasn't a sizable market of people creating applications using it.
As another example the following is a link for a company that specializes in creating java embedded solutions.
http://www.k-embedded-java.com/[^]
Following is a device.
http://www.avidwireless.com/AVIDdirector.html[^]
ErnestoNet wrote: Java embedded only runs on ARM to start with. It requires at least 130Kb (with
tweaking) and 700kb
That is a product that Oracle offers. It is far from the only product out there.
|
|
|
|
|
|
ErnestoNet wrote: You said it yourself, in C and C++ you can statically link. Or not use. In Java/C# you can't. You load 42Mb rt.jar in Java.
Actually no I didn't say. Actually I said the opposite. Rather specifically.
What part of what I said didn't you understand?
ErnestoNet wrote: About embedded Java in http://www.avidwireless.com/Products.html[^], it looks rather expensive and not a custom solution.
Which is irrelevant.
ErnestoNet wrote: We are talking about different kind of machines here.
If you want to define "embedded" software in a specific way and get the developers who currently call themselves embedded developers to agree that they are not in fact embedded developers then go for it.
But until then my statement stands.
|
|
|
|
|
ErnestoNet wrote: Text based compilers are easier to write that binary based compilers.
The headers approach is easy to build a compiler from.
C is a very
simple language to write a compiler.
There are LOTS of C compilers.
At this point I am rather certain that you do not know what you are talking about.
C headers are part of the language. Period.
C compilers implement the C language. Period.
The first fact is only related to the second by the fact that headers are in the language. It has nothing to do with compilation.
There is no such thing as a "binary" compiler in common usage. Best I can suppose you are talking about is what occurs in a Java Virtual Machine when it process a Java class file. That process it best described as interpretation not compilation.
Your confusion about the above also has nothing to do with your confusion about what portabiity means.
ErnestoNet wrote: More compilers provide better portability.
Wrong. You have confused availability with portability. More compilers means you can use it on more platforms.
In point of fact almost all C code written for the Windows system will not work on any other platform (without extreme care but that is my point about portability in the first place.)
ErnestoNet wrote: There are not many cross platform/architecture languages.
You are wrong.
Java and Perl exist on many platforms. Excluding small unix platforms, perl exists on basically all unix systems that C does.
And again availability is NOT the same as portability.
ErnestoNet wrote: Name a crossplatform language and I'll compare that to C in terms of
portability....
Your term definition for portability is wrong.
http://en.wikipedia.org/wiki/Software_portability[^]
By definition of programming languages I can always (within resource limits) create compiler/interpreter that originated on one platform and implement it for another. You are using that for your definition and then claiming that because C exists in many places that that makes in "portable" but instead is what it proves that it is 'popular' and 'useful'. Which is different than portability.
And in addition Perl shows up on all those systems too.
|
|
|
|
|
Call it availability. C is the most available language.
There is an important point you make:
I can always (within resource limits) create compiler/interpreter that originated on one platform and implement it for another.
"within resource limits" is the key here. C is simple. Java JVM is complicated (and owned).
It's important that you know that I'm not defending C.
It's really ugly compared to other languages and prone to error.
But it's fast (predictable fast, no GC collection pause), very "available", flexible and stable.
Most of the programs I use everyday are built in C/C++.
None in Perl. None in Java.
it´s the journey, not the destination that matters
|
|
|
|
|
ErnestoNet wrote: "within resource limits" is the key here. C is simple. Java JVM is complicated
(and owned).
Not sure what you think "owned" means but far as I know the term "Java" is owned but the language is licensed in such a way that I can in fact take it and create a compiler and complete JDK fro a new platform free of charge.
See "Licensing" in the following.
http://openjdk.java.net/faq/[^]
If you had to start from scratch then C, as in C11, is not trivial. And I am talking about the language and not the libraries. Implementing everything from scratch would be a large undertaking.
Of course one would start with the existing source and modified for a new platform.
Which is the same thing one would do for Java.
ErnestoNet wrote: But it's fast (predictable fast, no GC collection pause),
Again - research Real Time Java.
Other than that modern business usage is not impacted by the modern GC.
ErnestoNet wrote: Most of the programs I use everyday are built in C/C++.
None in Perl. None in
Java.
Which is why *I* said that language choice is by user preference and not availability.
Perl is available on all of the C platforms so they don't choose C because it is available.
|
|
|
|
|
pasztorpisti wrote: I'm open to any reasoning against the above list
In terms of the errors - those are not errors that show up for me. The errors that I must find involve logic errors. Whether those errors are made by me or others.
And if you spend significant amounts of time day to day on errors like those then at best that I can judge you have a problem that can only be solved with a change in process.
pasztorpisti wrote: In my opinion tools (including dev envs and languages) are very important.
I can and I have achieve orders of magnitude increase in performance by changing requirements.
The only way I have ever achieved anything close to that at the implementation level is because the design itself that lead to the implementation was wrong. Other than that implementation improvements can only lead to small increases when everything else is held steady.
The following is what impacts performance and even project success.
1. Requirements - most impact
2. Architecture
3. Design
4. Implementation - the least impact.
1/2 are not impacted by language although 2 might be impact by technology.
3 can be impacted by technology but only minimally by language.
4 is impacted by language.
So the reason language as less impact is because it does.
pasztorpisti wrote: I think every company that continues other than garage development has some
kind of dev process,
Every company has a process. Some companies have a formalized process. The only measured and significant improvements in development have come from formalizing the process and improving that. Far as I can remember improvement measures consisted of fewer bugs (detected at various points in life cycle), short delivery time, reduced cost and reduced maintenance cost.
pasztorpisti wrote: but that is another dimension of the problem of cutting development time and
providing quality on a completely different level: management
No it isn't. Again the only impact from the studies that I read was based on process improvements. Tools of any sort had nothing to do with it.
|
|
|
|
|
Please understand that I never questioned the importance of a development process and the same is true for architectural decisions and design. Indeed, the biggest mistakes I have ever seen were architectural/design mistakes. Without good design or anything above that we have nothing to speek about. My answers assume good architectural conditions and this time focus on the beautification of the least important 4th step.
Perhaps some day I will work just with the first 3 steps and I will probably be less concerned about the language and toolset the coders have hard time with...  Seriously!
|
|
|
|
|
Richard MacCutchan wrote: When it was chosen, it was the only choice.
Really?
The Multics operating system was being programmed in PL/1.
The Burroughs B-5000 and subsequent machines had their operating systems coded in Algol with extensions. And a damn fine operating system it was.
My alma mater, Case Western Reserve University, implemented an operating system with an Algol-like language called Chili.
All of this while C was just beginning to happen and Unix was not yet born.
Richard MacCutchan wrote: I meant it was unfair in that you were judging a language developed in the 80s
by the standards of today's knowledge and technology.
Oh! Does that mean you will be saying nice things about Fortran and COBOL? Or, at least, refrain from saying nasty things about them? 
|
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
