In my experience Meta-templates are really fun (when your compilers are compliant), and can give good performance boosts, and luckily I'm surrounded by seasoned C++ programmers that also grok meta-templates, however occasionally a new developer arrives and can't make heads or tails of some of the meta-template tricks we use (mostly Andrei Alenxandrescu stuff), for a few weeks until he gets initiated appropriately.
So I was wondering what's the situation for other C++ programmers out there?
Should meta-template programming be something C++ programmers should be "required" to know (excluding entry level students of course), or not?
Edit: Note my question is related to production code and not little samples or prototypes
If you can you find enough candidates who really know template meta-programing then by all means, require it. You will be showing a lot of qualified and potentially productive people the door (there are plenty of legitimate reasons not to know how to do this, namely that if you do it on a lot of platforms, you will create code that can't compile, or that average developers will have trouble understanding). Template meta-programming is great, but let's face it, it's pushing C++ to the limit.
Now, a candidate should probably understand basics (compute n! at compile time, or at least explain how it works if they are shown the code). If your new developers are reliably becoming productive within a few weeks, then your current recruiting is probably pretty good.
Yes, but I would not personally place a high priority on it. It's a nifty feature, but it's a bit situational, and good C++ code can be developed without it. I've personally used it once or twice, but haven't really found it to be valuable enough in my work to regularly use it. (Maybe that's a function of my lack of C++ production experience, though)
The only use I've ever made of template metaprogramming in production code is to unroll a critical loop which read a hardware register N times, followed by another M times, N, M different for different hardware and known at compile time. In general, the techniques don't seem a natural fit for our codebase, and I'd never get them through a code review.
Required? As always, it depends. For those of us in embedded land who are just now getting semi-decent C++ compilers for our tiny little DSPs and what not, we're just happy to be able to use classes.
However, if you've got a halfway decent C++ compiler, say gcc 3.3ish+, then yes, you should look at template metaprogramming. A good start is the boost library, of course, because it covers most of the templates you seem to look around for when the STL runs out of gas. It also serves as a great jumping off point.
However, sometimes I find that the advantages of template metaprogramming (lots of nice type safe code with a few lines of < and >) aren't worth the cost that it's going to take. Sometimes, a good old for( container::const_iterator iter = ... ) does what you need just fine.
18 months later and this topic is still very relevent. I would still say that template meta programming is not required knowledge, but you need to be able to at least read and explain the basics such as conditionals and the curiously repeating template pattern (looping). Beyond that, as long as you have a few people who can write a good interface to it, then just basic to intermediate template knowledge is all that is really required, though YMMV.
As someone who makes reasonable (although not extensive) use of templates and meta-programming, I go out of my way to try to make the interfaces (and by that I mean the internal usage interfaces) as normal as possible. Not everyone can understand templates, and even those that can sometimes cannot understand complex or obtuse meta programming paradigms.
With that said, if you want to dig in a modify my low-level libraries, you're going to have to know quite a bit. However, you should not even have to know templates (aside from baseline knowledge) to use them. That's how I draw the line at least, and the knowledge level I would expect in other developers (depending on how they are using the code).
I wouldn't consider template programming required, but it's definitely good to know. You should know enough about the subject to be able to effectively use template libraries such as the STL or Boost.
When I interview someone, I will always ask some questions about template meta-programming. If the candidate doesn't know about the subject, I would never hold that against them. But if they do, then it's a big plus in their favor.
It's not absolutely necessary to know how to use C++ templates. You can do most things without them. They are however a fantastic feature.
Since you roll your own templates, anyone new is going to have to come up to speed with them just like the rest of your code which is going to be the bigger chunk of the learning.
I encourage people to learn to use some of the features of the STL. I have used this library in production code and it does save time and simplify things quite a bit. I also roll my own when the need arises.
I've also heard good things about the boost library.
If I need to write portable code then I'll generally stick away from templates because many compilers still don't support them properly. If you need a portable STL then STLPort is the most portable.
Related
First a little background ...
In what follows, I use C,C++ and Java for coding (general) algorithms, not gui's and fancy program's with interfaces, but simple command line algorithms and libraries.
I started out learning about programming in Java. I got pretty good with Java and I learned to use the Java containers a lot as they tend to reduce complexity of book keeping while guaranteeing great performance. I intermittently used C++, but I was definitely not as good with it as with Java and it felt cumbersome. I did not know C++ enough to work in it without having to look up every single function and so I quickly reverted back to sticking to Java as much as possible.
I then made a sudden transition into cracking and hacking in assembly language, because I felt I was concentrated too much attention on a much too high level language and I needed more experience with how a CPU interacts with memory and whats really going on with the 1's and 0's. I have to admit this was one of the most educational and fun experiences I've had with computers to date.
For obviously reasons, I could not use assembly language to code on a daily basis, it was mostly reserved for fun diversions. After learning more about the computer through this experience I then realized that C++ is so much closer to the "level of 1's and 0's" than Java was, but I still felt it to be incredibly obtuse, like a swiss army knife with far too many gizmos to do any one task with elegance. I decided to give plain vanilla C a try, and I quickly fell in love. It was a happy medium between simplicity and enough "micromanagent" to not abstract what is really going on. However, I did miss one thing about Java: the containers. In particular, a simple container (like the stl vector) that expands dynamically in size is incredibly useful, but quite a pain to have to implement in C every time. Hence my code currently looks like almost entirely C with containers from C++ thrown in, the only feature I use from C++.
I'd like to know if its consider okay in practice to use just one feature of C++, and ignore the rest in favor of C type code?
The short answer is, "This is not really the most effective way to use C++."
When used correctly, the strong type system, the ability to pass by reference, and idioms like RAII make C++ programs more likely to be correct, readable, and maintainable.
No one can stop you from using the language the way you want to. But you may be limiting yourself by not learning and leveraging actual C++ features.
If you write code that other people will have to read and maintain, they will probably appreciate the use of "real C++" instead of "C with classes" (in the words of a previous commenter).
Seems fine to me. That's the only part of C++ that I really use as well.
Right now, I'm writing a number cruncher. There's no polymorphism, no control delegation, no interaction. <iostream> was a bottleneck so I rewrote I/O in C.
The functions are mostly inside one class which represents a work thread. So that's not so much OO as having thread-local variables.
As well as vector, I use <algorithms> pretty heavily. But the heavy-duty data structures are written in plain C. Mainly circular singly-linked lists, which can't even easily have distinct begin() and end(), meaning not only containers but sequences (and for-loops) are off-limits. And then templates help the preprocessor to generate the main inner loop.
The most natural way of solving your problem is probably right. You don't want solutions in search of a problem. Learning to use C++ is well and good, but object orientation is suited to some problems and not others.
On the other hand, using bsearch from stdlib.h in a C++ program would be wrong.
You should use C++ in whatever way makes the most sense for you.
I'm curious to find out if and when C++ meta templates are a good design choice for systems small to large. I understand that they increase your build time in order to speed up your execution time. However, I've heard that the meta template code is inherently hard to understand by many developers, which could be a problem for a large group of people working on a system with a very large code base (millions of lines of code). Where do you think C++ meta templates are useful (or not)?
Metaprogramming is just another tool in a (C++) programmers' toolbox - it has many great applications, but like anything can be mis- or over- used. I think it's got a bad reputation in terms of 'hard to use', and I think this mainly comes from the fact that it's a significant addition to the language and so takes a while to learn.
As an example of real-world use; I've used template metaprogramming to implement Compile-time asserts and shim libraries in the past; implementing these without templates would either have been impossible or required significantly more code than I had to write.
In the case of the shim library it could have been implemented in a classic object-orientated fashion which could have been designed to have a similar (low) level of code duplication as the templated implementation; however it's runtime performance would have been significantly worse.
If you want to see some good examples of how it can be used, I suggest you read Modern C++ Design by Andrei Alexandrescu (there is a sample chapter on the publisher's website) - IMO this is one of the best books on the subject.
Template metaprogramming doesn't make your code "inherently hard to understand". It moves the complexity around. The underlying metaprogramming code may be a pain to understand, but at the same time, it typically simplifies client code.
If its effect was just to make code harder to read, it wouldn't be used. The reason it is used from time to time is exactly that it simplifies the code as a whole.
Of course, programmers who are not familiar with metaprogramming will have trouble reading or maintaining the code, but isn't that just an argument against working with programmers who don't know their stuff?
Programmers who don't know about a for-loop wil find that hard to read too.
Fairly simple meta-programming is used throughout the standard library, in the form of traits. The standard library seems to be pretty well received, so I think we can say that meta programming is useful there.
I did face a situation where I had to tackle a not-so-big system which heavily used template metaprogramming (specifically static polymorphismus, SFINAE and maybe other techniques). And I can tell you that this will make it harder for the developers. If template metaprogramming is used a lot, every developer must be familiar with the techniques otherwise they won't be able to work productively.
On the other hand, some uses of templates are quite easy to understand and use. For example generic containers (vector), smart pointers, ...
As always, you need to balance the pros and the cons. If performance is your concern (there are other uses for template metaprogramming too), you should first demonstrate that there are significant and important performance gains to be had by using template metaprogramming techniques. (It's even quite possible to make a program that runs slower by hindering the compiler with excessive use of templates, so always measure!)
C++ templates were not originally designed for metaprogramming so even relatively simple problems solved with metaprogramming can produce code that is difficult to understand, especially to people who are not familiar with common template metaprogramming techniques (since they are usually not very intutitive). Also depending on your particular problem you might consider code generation vs. template metaprogramming.
This depends on who you're working with, and what they like and are familiar with. In the absence of any information I'd suggest the following concrete list of uncontroversial (and not very meta...) template 'things' for a new project:
handwritten smart pointers for whatever sorts of things your app uses
STL containers
static assert
And, for advanced users only:
traits used for some obvious standard application (script language bindings and serialization spring to mind)
This is perhaps a bit conservative but it seems to be easy to convince everybody of the value of these things. If well put together, they don't bloat compile times and mostly work pretty well with commonly-found code browsing facilities. And most of the templates shouldn't take much explaining, even to that large subset of C++ programmers who don't really understand templates all that well.
(Regarding boost, and any other template libraries that it has yet to merge with, it seems to be pretty adventurous by the standards of many (most?) C++ programmers at the moment. So it seems to me prudent to avoid it for now.)
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 4 years ago.
Improve this question
Lots of the answers to C++ questions here contain the response:
"You should be using boost::(insert
your favourite smart pointer here) or
even better boost::(insert your
favourite mega complex boost type
here)"
I'm not at all convinced that this is doing any favours to the questioners who, by and large, are obvious C++ novices. My reasons are as follows:
Using smart pointers without
understanding what is going on under
the hood is going to lead to a
generation of C++ programmers who
lack some of the basic skills of a
programmer. Pretty much this seems to
have happened in the Java field
already.
Deciding which type of smart pointer
to use depends very much on the
problem domain being addressed. This
is almost always missing from the
questions posted here, so simply
saying "use a shared pointer" is
likely to be at the least unhelpful
and possibly totally wrong.
Boost is not yet part of the C++
standard and may not be available on
the specific platform the questioner
is using. Installing it is a bit
painful (I just did it using Jam) and
is way overkill if all you want are a
few smart pointers.
If you are writing FOSS code, you
don't want the code to be heavily
dependent on external libraries that,
once again, your users may not have.
I've been put off using FOSS code on
a number of occasions simply because
of the Byzantine complexity of the
dependencies between libraries.
To conclude, I'm not saying don't recommend Boost, but that we should be more careful when doing so.
Few points:
Using anything without understanding is considered harmful. But it is only the ignorant technology user (and his manager) who gets burned in the end.
You don't have to install boost to get the smart pointers - they are header only. And installation itself is rather straightforward, in the simplest approach just typing one or two commands.
Many of the Boost libraries and solutions are present in TR1 or will be present in C++0x
You will always depend on external libraries... Try to choose the one that have a bright future in terms of maintenance and support.
Unless you want to roll-out your custom solution - which would have some advantages and disadvantages.
C++ is not a novice-friendly language. With apologies to Scott Meyers, a beginner isn't learning just one language with C++, but four:
The C parts
Object Oriented parts: classes, inheritance, polymorphism, etc.
The STL: containers, iterators, algorithms
Templates and metaprogramming
I would argue that if the beginner is already climbing this mountain, they should be pointed towards the more "modern" aspects of C++ from the start. To do otherwise means that the beginner will learn C-ish C++ with regular pointers, resource leaks, etc. Find themselves in a world of pain, and then discover Boost and other libraries as a way to stem the hurt.
It's a complicated picture no matter what, so why not point them in a direction that has a positive pay-off for the invested mental efort?
As for dependencies, a great deal of Boost is header-only. And Boost's liberal license should permit its inclusion in just about any project.
Do you know how the compiler works ? Do you know how the OS works ? Do you know how the processor works ? Do you know how electronics works ? Do you know how electricity works ?
At some point you are using a black box, the question is, "is my ignorance problematic for what I am currently doing?".
If you have the taste for knowledge that's a great thing - and I clearly consider that a plus when interviewing engineers - but don't forget the finality of your work : build systems that solve problems.
I disagree. No-one would suggest that you should dive in to smart pointers without a thorough understanding of what's going on behind the scenes, but used sensibly they can remove a whole host of common errors. Moreover, Boost is high-quality production code from which a C++ novice can learn a great deal, in terms of design as much as implementation. It's not all hugely complicated, either, and you can pick and choose the bits you need.
It's impossible to understand everything thoroughly all the time. So take the word of many professional C++ developers for it that many parts of boost are indeed very useful things to use in your day-to-day development.
The inclusion of quite a lot of boost in C++0X is testament that even the team that manages the evolution of the language thinks that boost is a Good Thing (tm)
C++ is a weird, tough language. It's relatively easy to learn compared to how incredibly hard it is to master. There's some really arcane stuff you can do with it. Boost::mpl builds on some of those arcane things. I love boost, but I cringe every time I see someone in my organisation use boost::mpl. The reason: even quite seasoned C++ developers have trouble wrapping their head around how it works, and the code that uses it often reflects that (it ends up looking like someone banged code out until it worked). This is not a good thing, so I partially agree that some parts of boost should not be used without caution (boost::spirit is another example).
The C++ standard is also a weird thing. Most common compilers don't implement all of the existing standard (e.g. template exports). It's only a guideline of what to expect.
If your developer doesn't have the savvy to decide which smart pointer to use in a particular situation, perhaps they shouldn't be messing around in that part of the code without senior guidance.
There are always external libraries, starting with the run-time. A lot of boost is header-only so it does not introduce new external dependencies.
Quite frankly, for beginners I think boost isn't that well-suited. I think a beginner is better off understanding how the basics work before moving up the food chain using higher level tool/libs like boost or even STL. At the beginner stage it is not about productivity, it is about understanding. I think knowing how pointers work, being able for instance to manually create a linked list or sort one are part of the fundamentals that each programmer should learn.
I think boost is a great library. I love it. My favourite library is boost::bind and boost::function, which make function pointers much more flexible and easy-to-use. It fits in very well with different frameworks and keeps the code tidy.
I also use different Boost classes. For example, I use boost::graph to create graph classes and I use boost::filesystem for working with files inside directories.
However, boost is very complex. You need to be an experienced programmer to know its worth. Moreover, you need to have atleast some experience in C++ to understand how Boost works and implications of using Boost here or there.
Therefore, I would highly recommend looking at Boost for experienced programmers, especially if they are trying to re-invent the wheel (again). It can really be what it says on the tin: a boost towards your goal.
However, if you feel that the person asking a question is a beginner and tries to understand (for example) memory allocation, telling him to try boost smart pointers is a very bad idea. It's not helpful at all. The advantages of smart pointer classes, etc. can be comprehended only when the person experienced how standard memory allocation techniques work.
To finish off, Boost is not like learning to drive a car with automatic gearbox. It's like learning to drive on a F1 racing car.
I fully agree with you. It is the reason that i first explain them how it should be done (i.e when recommending boost::variant, i explain they should in general use a discriminated union. And i try not to say it's just a "magic boost thing" but show how they in principle implemented it. When i recommend boost::shared_ptr, i explain they would need to use a pointer - but it's better to use a smart pointer that has shared ownership semantics.). I try not to say just "use boost::xxx" when i see the questioner is a beginner. It is a language that's not just as simple to use as some scripting language. One has to understand the stuff one uses, because the language does not protect the programmer from doing bad things.
Of course it's not possible for novices to understand everything from the start on. But they should understand what their boost library solves and how it does it basically.
You can't compare this with learning processors or assembly language first. Similar it's not important to know how the bit-pattern of a null-pointer looks like. Knowledge of those are irrelevant in learning programming with C++. But pointers, array or any other basic things in C++ is not. One doesn't get around learning them before using [boost|std]::shared_ptr or [boost|std]::array successfully. These are things that has to be understood first in order to use the boost utilities successfully in my opinion. It's not about details like how to manually implement the pimpl-idiom using raw pointers - that's not the point I'm making. But the point is that one should first know basic things about pointers or the other parts a boost library helps with (for pointers, what they are and what they are good for, for example). Just look at the shared_ptr manual and try to get it without knowing about pointers. It's impossible.
And it's important to always point them to the appropriate boost manual. Boost manuals are high quality.
The consensus among almost all the answers is that boost is very valuable for experienced developers and for complex, real world, C++ software. I completely agree.
I also think that boost can be very valuable for beginners. Isn't it easier to use lexical_cast than to use ostringstream? Or to use BOOST_FOREACH instead of iterator syntax? The big problem is lack of good documentation of boost, especially for beginners. What is needed is a book that will tell you how to start with boost, which libraries are simple libraries that simplify tasks, and which libraries are more complex. Using these libraries together with good documentation will IMO make learning C++ easier.
We should encourage the use of standard canned libraries (and Boost is almost as standard as they get) whenever possible.
Some people seem to think that beginners should be taught the C side of C++ first, and then introduced to the higher-level stuff later. However, people tend to work as they're trained, so we're going to see a lot of production code written with badly managed raw pointers (well-managed raw pointers are awfully difficult sometimes), arrays (and the inevitable confusion between delete and delete []), and stuff like that. I've worked with code like that. I don't want to do it again any more than I have to.
Start beginners off with the way you want them writing code. This means teaching them about the STL containers and algorithms and some of the Boost libraries at first, so the first thing they think about when needing a group of things is a vector<>. Then teach them the lower-level constructs, so they'll know about them (or where to look them up) when they encounter them, or on the very rare occasions when they need to micro-optimize.
There's basically two types of programmers: the coders, who should be taught languages the way they should be writing them, and the enthusiast, who will learn the low-level stuff, including principles of operating systems, C, assembly code, and so on. Both are well served by learning the language they're going to use up front, while only the enthusiasts will be well served by learning from some arbitrary level of fundamentals.
I think you are mixing a lot of different concerns, not all of them related to Boost specifically:
First, should programmers (or C++ novices specifically) be encouraged to use libraries, idioms, paradigms, languages or language features they don't understand?
No, of course not. Every programmer should understand the tools they use, especially in a language like C++. However, I don't see a lot of questions here on SO where people are encouraged to not understand the code they're using. When people say they want to do X in C++, I think it's find to say "Boost has an implementation of X which works, which is more than a homebrewed solution would do, so use that".
Of course if the question is "how does X work", the question can't be answered with "use Boost's implementation". But I really don't see the problem in recommending Boost for the former kind of questions.
I also don't see how it's even possible to use Boost without understanding what's going on under the hood. C++, with or without Boost, is not Java. Using Boost in no way protects you from the complexities of the language. You still have to worry about copy constructors, pointer arithmetics, templates and everything else that can blow up in your face.
This is nothing like what happened in Java. They designed a language that removed all the subtleties. Boost doesn't do that. Quite the contrary, it has pioneered new idioms and techniques in generic programming. Using Boost is not always simple.
About the availability of Boost, I think that's a non-issue. It is available on the platforms used in the vast majority of questions, and if they're not able to use Boost, the suggestion is still not harmful, just useless.
Further, most Boost libraries are header-only and don't require you to install the whole thing. If you only want smart pointers, simply include those headers and nothing else.
About FOSS, you have a point in some cases But I'd say this is a problem for less universal libraries that users do not have. But Boost is extremely common, and if people don't have it, they should get it, as it is applicable to pretty much any problem domain. And of course, the license is compatible with any FOSS project you care to mention.
I'd rather work on a OSS project that used Boost to do the heavy lifting than one which reinvented its own (buggy and proprietary) wheels, with steep learning curves that could have been avoided.
So yeah, in some cases, recommending Boost is unhelpful. But I don't see how it can be harmful.
In any case, I don't see how it can be even half as harmful as teaching novices to roll their own. In C++, that's a recipe for disaster. It's the sole reason why C++ still has a reputation for being error-prone and produce buggy software. Because for far too long, people wrote everything from scratch themselves, distrusting the standard library, distrusting 3rd party code, distrusting everything that wasn't legal in C.
I'm not at all convinced that this is doing any favours to the questioners who, by and large, are obvious C++ novices. ...:
Using smart pointers without understanding what is going on under the hood is going to lead to a generation of C++ programmers who lack some of the basic skills of a programmer.
Do we tell novice programmers that they must learn assembly language before they get to read up on modern programming languages? They clearly don't know what's going on under the hood otherwise.
Should "Hello World" include an implementation of the I/O subsystem?
Personally I learned how to construct objects before I learned how to write classes. I think I learned how to use STL vectors before I learned C-style arrays. I think it's the right approach: "here's how to refer to several nearly identical variables using a std::vector, later I'll show you what's swept under the rug via C-style arrays and new[] and delete[]."
I disagree. Of course you will always know more about the internal workings of everything when coding it from scratch than when using 3rd party libraries. But time and money are limited, and using good 3rd party libraries such as boost is a very good way to save your resources.
I can see your point, but understanding something does not mean that you have to rewrite everything from scratch.
They are not "standard" but they are as standard as a library can get.
It is true that deploying them can be painful (but not all of the sublibraries require compilation); on the other hand they do not have further dependencies on their own, so I wouldn't be too worried about that part neither.
I agree with you, high level libraries hide things from you. It might be a good idea in the short run, but in the long run, the novice will have severe gaps in their understanding of the language.
It's easy for us non-novices to say "just use this library" because we've been down that long hard road of learning things the hard way, and naturally we want to save someone else the trouble of doing the same.
Novices SHOULD have to struggle with rolling their own low-level solutions to problems. And then, when they've got a better understanding of how their own solution worked, they can use the third-party solution, confident that they have some idea of what's going on under the hood. They'll use that library better!
I think this is a broader subject than just being about Boost. I completely regret picking up VB as my first language. If I had just started with ugly, hard to learn c, I'd be years ahead of where I am now.
I would agree with the point about smart pointers. I am a C++ beginner, and when asking a simple question about pointer syntax, one answer suggested smart pointers were the way to go. I know I'm not ready for boost (I'm not really ready for the STL either), so in most cases I steer myself away from that type of suggestion.
Scoped and dynamic resource ownership are general basic neeeds and boost's implementation of'em is very good an highly recommended. I use them a lot and they work fine.
Boost is a great library. I really hope that it grows in breadth and acceptance. Use it, extend it, and promote it.
One of the great things about the .NET community is that it has a great base class library. One of the fundemental problems with C++, I believe, is the minimalistic C++ standard library. Anywhere you go to develop code, FOSS or corporate, there is some selection of libraries that are used since there isn't a broad standard library. So you end up being a INSERT_YOUR_COMPANY_HERE C++ programmer and not necessarily too transferrable. Yes, you design/architecture skills transfer, but there is the learning curve with picking up familiarity with whatever set of libraries the next place is using. Where as a .NET developer will basically be using the same class library and can hit the ground running. Also, the libraries that are built (and reused) have a broader base to build on.
Just as an aside, you can use http://codepad.org for a code paste bin and it supports boost!
I have worked for companies who have viewed boost as library to avoid due in part to its past reputation as a poorly managed project. I know things have changed with the project, but commercial projects who want to use boost must be aware of the source of the code contained in the library, or at least be assured that they're not going to be liable for IP or patent infringements.
Sadly, the library has this reputation and it will take a while for it to break before it sees wide use in the commercial sector. I also feel this is a reason not to recommend it blindly.
It is a university task in my group to write a compiler of C-like language. Of course I am going to implement a small part of our beloved C++.
The exact task is absolutely stupid, and the lecturer told us it need to be self-compilable (should be able to compile itself) - so, he meant not to use libraries such as Boost and STL. He also does not want us to use templates because it is hard to implement.
The question is - is it real for me, as I`m going to write this project on my own, with the deadline at the end of May - the middle of June (this year), to implement not only templates, but also nested classes, namespaces, virtual functions tables at the level of syntax analysis?
PS I am not noobie in C++
Stick to doing a C compiler.
Believe me, it's hard enough work building a decent C compiler, especially if its expected to compile itself. Trying to support all the C++ features like nested classes and templates will drive you insane. Perhaps a group could do it, but on your own, I think a C compiler is more than enough to do.
If you are dead set on this, at least implement a C-like language first (so you have something to hand in). Then focus on showing off.
"The exact task is absolutely stupid" - I don't think you're in a position to make that judgment fairly. Better to drop that view.
"I`m going to write this project on my own" - you said it's a group project. Are you saying that your group doesn't want to go along with your view that it should morph into C++, so you're taking off and working on your own? There's another bit I'd recommend changing.
It doesn't matter how knowledgable you are about C++. Your ability with grammars, parsers, lexers, ASTs, and code generation seems far more germane.
Without knowing more about you or the assignment, I'd say that you'd be doing well to have the original assignment done by the end of May. That's three months away. Stick to the assignment. It might surprise you with its difficulty.
If you finish early, and fulfill your obligation to your team, I'd say you should feel free to modify what's produced to add C++ features.
I'll bet it took Bjarne Stroustrup more than three months to add objects to C. Don't overestimate yourself or underestimate the original assignment.
No problem. And while you're at it, why not implement an operating system for it to run on too.
Follow the assignment. Write a compiler for a C-like language!
What I'd do is select a subset of C. Remove floating-point datatypes and every other feature that isn't necessary in building your compiler.
Writing a C compiler is a lot of work. You won't be able to do that in a couple of months.
Writing a C++ compiler is downright insane. You wouldn't be able to do that in 5 years.
I will like to stress a few points already mentioned and give a few references.
1) STICK TO THE 1989 ANSI C STANDARD WITH NO OPTIMIZATION.
2) Don't worry, with proper guidance, good organization and a fair amount of hard work this is doable.
3) Read the The C Programming Language cover to cover.
4) Understand important concepts of compiler development from the Dragon Book.
5) Take a look at lcc both the code as well as the book.
6) Take a look at Lex and Yacc (or Flex and Bison)
7) Writing a C compiler (up to the point it can self compile) is a rite of passage ritual among programmers. Enjoy it.
For a class project, I think that requiring the compiler to be able to compile itself is a bit much to ask. I assume that this is what was meant by stupid in the question. It means that you need to figure out in advance exactly how much of C you are going to implement, and stick to that in building the compiler. So, building a symbol table using primitives rather than just using an STL map. This might be useful for a data structure course, but misses the point for a compiler course. It should be about understanding the issues involved with the compiler, and chosing which data structures to use, not coding the data structures.
Building a compiler is a wonderful way to really understand what happens to your code once the compiler get a hold of it. What is the target language? When I took compilers, it took 3 of us all semester to build a compiler to go from sorta-pascal to assembly. Its not a trivial task. Its one of those things that seems simple at first, but the more you get into it, the more complicated things get.
You should be able to complete c-like language within the time frame. Assuming you are taking more than 1 course, that is exactly what you might be able to do in time. C++ is also doable but with a lot more extra hours to put it. Expecing to do c++ templates/virtual functions is overexpecting yourself and you might fail in the assignment all together. So it's better stick with a c subset compiler and finish it in time. You should also consider the time it takes for QA. If you want to be thorough QA itself will also take good time.
Namespaces or nested clases, either virtual functions are at syntax level quite simple, its just one or two more rules to parser. It is much more complicated at higher levels, at deciding, which function / class choose (name shadowing, ambiguous names between namespaces, etc.), or when compiling to bytecode/running AST. So - you may be able to write these, but if isn't necessary, skip it, and write just bare functional model.
If you are talking about a complete compiler, with code generation, then forget it. If you just intend to do the lexical & syntactic analysis side of things, then some form of templating may just about be doable in the time frame, depending on what compiler building tools you use.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 4 years ago.
Improve this question
Recently, I've got a dangerous idea into my head after reading this blog post. That idea can be expressed like this:
I don't need most of what the C++ standard library offers. So, why don't I implement a less general, but easier to use version?
As an example, using the STL spits out reams of incomprehensible and mangled compiler errors. But, I don't care about allocators, iterators and the like. So why don't I take a couple of hours and implement an easy to use linked list class, for example?
What I'd like to know from the StackOverflow community is this: what are the dangers, possible disadvantages and possible advantages to "rolling my own" for most of the existing functionality in C++?
Edit: I feel that people have misunderstood me about this idea. The idea was to understand whether I could implement a very small set of STL functionality that is greatly simplified - more as a project to teach me about data structures and the like. I don't propose re-inventing the entire wheel from the ground up, just the part that I need and want to learn about. I suppose what I wanted to figure out is whether the complexity of using the STL warrants the creation of smaller, simpler version of itself.
Re-using boost or similiar.
Most of what I code is for University and we're not allowed to use external libraries. So it's either the C++ standard library, or my own classes.
Objectivity of this question.
This question is not subjective. Nor should it be community Wiki, since it's not a poll. I want concrete arguments that highlight one advantage or one disadvantage that could possibly occur with my approach. Contrary to popular belief, this is not opinion, but based on experience or good logical arguments.
Format.
Please post only one disadvantage or one advantage per answer. This will allow people to evaluate individual ideas instead of all your ideas at once.
And please...
No religious wars. I'm not a fan boy of any language. I use whatever's applicable. For graphics and data compression (what I'm working on at the moment) that seems to be C++. Please constrain your answers to the question or they will be downvoted.
So, why don't I implement a less
general, but easier to use version?
Because you can't. Because whatever else you might say about C++, it is not a simple language, and if you're not already very good at it, your linked list implementation will be buggy.
Honestly, your choice is simple:
Learn C++, or don't use it. Yes, C++ is commonly used for graphics, but Java has OpenGL libraries too. So does C#, Python and virtually every other language. Or C. You don't have to use C++.
But if you do use it, learn it and use it properly.
If you want immutable strings, create your string as const.
And regardless of its underlying implementation, the STL is remarkably simple to use.
C++ compiler errors can be read, but it takes a bit of practice. But more importantly, they are not exclusive to STL code. You'll encounter them no matter what you do, and which libraries you use. So get used to them. And if you're getting used to them anyway, you might as well use STL too.
Apart from that, a few other disadvantages:
No one else will understand your code. If you ask a question on SO about std::vector, or bidirectional iterators, everyone who's reasonably familiar with c++ can answer. If you ask abut My::CustomLinkedList, no one can help you. Which is unfortunate, because rolling your own also means that there will be more bugs to ask for help about.
You're trying to cure the symptom, rather than the cause. The problem is that you don't understand C++. STL is just a symptom of that. Avoiding STL won't magically make your C++ code work better.
The compiler errors. Yes, they're nasty to read, but they're there. A lot of work in the STL has gone into ensuring that wrong use will trigger compiler errors in most cases. In C++ it's very easy to make code that compiles, but doesn't work. Or seems to work. Or works on my computer, but fails mysteriously elsewhere. Your own linked list would almost certainly move more errors to runtime, where they'd go undetected for a while, and be much harder to track down.
And once again, it will be buggy. Trust me. I've seen damn good C++ programmers write a linked list in C++ only to uncover bug after bug, in obscure border cases. And C++ is all border cases. Will your linked list handle exception safety correctly? Will it guarantee that everything is in a consistent state if creating a new node (and thereby calling the object type's constructor) throws an exception? That it won't leak memory, that all the appropriate destructors will be called? Will it be as type-safe? Will it be as performant? There are a lot of headaches to deal with when writing container classes in C++.
You're missing out on one of the most powerful and flexible libraries in existence, in any language. The STL can do a lot that would be a pain even with Java's giant bloated class library. C++ is hard enough already, no need to throw away the few advantages it offers.
I don't care about allocators,
iterators and the like
Allocators can be safely ignored. You pretty much don't even need to know that they exist. Iterators are brilliant though, and figuring them out would save you a lot of headaches. There are only three concepts you need to understand to use STL effectively:
Containers: You already know about these. vectors, linked lists, maps, sets, queues and so on.
Iterators: Abstractions that let you navigate a container (or subsets of a container, or any other sequence of value, in memory, on disk in the form of streams, or computed on the fly).
Algorithms: Common algorithms that work on any pair of iterators. You have sort, for_each, find, copy and many others.
Yes, the STL is small compared to Java's library, but it packs a surprising amount of power when you combine the above 3 concepts. There's a bit of a learning curve, because it is an unusual library. But if you're going to spend more than a day or two with C++, it's worth learning properly.
And no, I'm not following your answer format, because I thought actually giving you a detailed answer would be more helpful. ;)
Edit:
It'd be tempting to say that an advantage of rolling your own is that you'd learn more of the language, and maybe even why the STL is one of its saving graces.. But I'm not really convinced it's true. It might work, but it can backfire too.
As I said above, it's easy to write C++ code that seems to work. And when it stops working, it's easy to rearrange a few things, like the declaration order of variables, or insert a bit of padding in a class, to make it seemingly work again. What would you learn from that? Would that teach you how to write better C++? Perhaps. But most likely, it'd just teach you that "C++ sucks". Would it teach you how to use the STL? Definitely not.
A more useful approach might be utilizing the awesome power of StackOverflow in learning STL the right way. :)
Disadvantage: no one but you will use it.
Advantage: In the process of implementing it you will learn why the Standard Library is a good thing.
Advantages: eating your own dogfood. You get exactly what you do.
Disadvantages: eating your own dogfood. Numerous people, smarter than 99 % of us, have spent years creating STL.
I suggested you learn why:
using the STL spits out reams of
incomprehensible and mangled compiler
errors
first
Disadvantage: you may spend more time debugging your class library than solving whatever university task you have in front of you.
Advantage: you're likely to learn a lot!
There is something you can do about the cryptic compiler STL error messages. STLFilt will help simplify them. From the STLFilt Website:
STLFilt simplifies and/or reformats
long-winded C++ error and warning
messages, with a focus on STL-related
diagnostics (and for MSVC 6, it fully
eliminates C4786 warnings and their
detritus). The result renders many of
even the most cryptic diagnostics
comprehensible.
Have a look here and, if you are using VisualC, also here.
I think you should do it.
I'm sure I'll get flambayed for this, but you know, every C++ programmer around here has drunk a little too much STL coolaid.
The STL is a great library, but I know from first hand experience that if you roll your own, you can:
1) Make it faster than the STL for your particular use cases.
2) You'll write a library with just the interfaces you need.
3) You'll be able to extend all the standard stuff. (I can't tell you how much I've wished std::string had a split() method)...
Everyone is right when they say that it will be a lot of work. Thats true.
But, you will learn a lot. Even if after you write it, you go back to the STL and never use it again, you'll still have learned a lot.
A bit of my experience : Not that long ago I have implemented my own vector-like class because I needed good control on it.
As I needed genericity I made a templated array.
I also wanted to iterate through it not using operator[] but incrementing a pointer like a would do with C, so I don't compute the address of T[i] at each iteration... I added two methods one to return pointer to the allocated memory and another that returns a pointer to the end.
To iterate through an array of integer I had to write something like this :
for(int * p = array.pData(); p != array.pEnd(); ++p){
cout<<*p<<endl;
}
Then when I start to use vectors of vectors I figure out that when it was possible a could allocate a big bloc of memory instead of calling new many times. At this time I add an allocator to the template class.
Only then I notice that I had wrote a perfectly useless clone of std::vector<>.
At least now I know why I use STL...
Disadvantage : IMHO, reimplimenting tested and proven libraries is a rabit hole which is almost garanteed to be more trouble than it's worth.
Another Disadvantage:
If you want to get a C++ job when you're finished with University, most people who would want to recruit you will expect that you are familiar with the Standard C++ library. Not necessarily intimately familiar to the implementation level but certainly familiar with its usage and idioms. If you reimplement the wheel in form of your own library, you'll miss out on that chance. This is nonwithstanding the fact that you will hopefully learn a lot about library design if you roll your own, which might earn you a couple of extra brownie points depending on where you interview.
Disadvantage:
You're introducing a dependency on your own new library. Even if that's sufficient, and your implementation works fine, you still have a dependency. And that can bite you hard with code maintenance. Everyone else (including yourself, in a year's time, or even a month's) will not be familiar with your unique string behavior, special iterators, and so on. Much effort will be needed just to adapt to the new environment before you could ever start refactoring/extending anything.
If you use something like STL, everyone will know it already, it's well understood and documented, and nobody will have to re-learn your custom throwaway environment.
You may be interested in EASTL, a rewrite of the STL Electronic Arts documented a while back. Their design decisions were mostly driven by the specific desires/needs in multiplatform videogame programming. The abstract in the linked article sums it up nicely.
Advantage
If you look into MFC, you'll find that your suggestion already is used in productive code - and has been so for a long time. None of MFC's collection classes uses the STL.
Why don't you take a look at existing C++ libraries. Back when C++ wasn't quite as mature, people often wrote their own libraries. Have a look at Symbian (pretty horrible though), Qt and WxWidgets (if memory serves me) have basic collections and stuff, and there are probably many others.
My opinion is that the complexity of STL derives from the complexity of the C++ language, and there's little you can do to improve on STL (aside from using a more sensible naming convention). I recommend simply switching to some other language if you can, or just deal with it.
Disadvantage : You're university course is probably laid out like this for a reason. The fact that you are irritated enough by it (sarcasm not intended), may indicate you are not getting the paridigm, and will benefit a lot when you have a paradigm shift.
As an example, using the STL spits out
reams of incomprehensible and mangled
compiler errors
The reason for this is essentially C++ templates. If you use templates (as STL does) you will get reams of incomprehensible error messages. So if you implement your own template based collection classes you will not be in any better spot.
You could make non template based containers and store everything as void pointers or some base class e.g. But you would lose compile time type checks and C++ sucks as a dynamic language. It is not as safe to do this as it would be in e.g. Objective-C, Python or Java. One of the reasons being that C++ does not have a root class for all classes to all introspection on all objects and some basic error handling at runtime. Instead your app would likely crash and burn if you were wrong about the type and you would not be given any clues to what went wrong.
Disadvantage: reimplementing all of that well (that is, at a high level of quality) will certainly take a number of great developers a few years.
what are the dangers, possible disadvantages and possible advantages to "rolling my own" for most of the existing functionality in C++?
Can you afford and possibly justify the amount of effort/time/money spent behind reinventing the wheel?
Re-using boost or similiar.
Rather strange that you cannot use Boost. IIRC, chunks of contribution come in from people related to/working in universities (think Jakko Jarvi). The upsides of using Boost are far too many to list here.
On not 'reinventing the wheel'
Disadvantage: While you learn a lot, you also set yourself back, when you come to think of what your real project objectives are.
Advantage: Maintenance is easier for the folks who are going to inherit this.
STL is very complex because it needs to be for a general purpose library.
Reasons why STL is the way it is:
Based on interators so standard algorithms only need a single implementation for different types of containers.
Designed to behave properly in the face of Exceptions.
Designed to be 'thread' safe in multi threaded applications.
In a lot of applications however you really have enough with the following:
string class
hash table for O(1) lookups
vector/array with sort / and binary search for sorted collections
If you know that:
Your classes do not throw exceptions on construction or assignment.
Your code is single threaded.
You will not use the more complex STL algorithms.
Then you can probably write your own faster code that uses less memory and produces simpler compile/runtime errors.
Some examples for faster/easier without the STL:
Copy-on-Write string with reference counted string buffer. (Do not do this in a multi-threaded environment since you would need to lock on the reference count access.)
Use a good hash table instead of the std::set and std::map.
'Java' style iterators that can be passed around as a single object
Iterator type that does not need to know the type of the container (For better compile time decoupling of code)
A string class with more utility functions
Configurable bounds checking in your vector containers. (So not [] or .at but the same method with a compile or runtime flag for going from 'safe' to 'fast' mode)
Containers designed to work with pointers to objects that will delete their content.
It looks like you updated the question so now there are really two questions:
What should I do if I think the std:: library is too complex for my needs?
Design your own classes that internally use relevant std:: library features to do the "heavy lifting" for you. That way you have less to get wrong, and you still get to invent your own coding interface.
What should I do if I want to learn how data structures work?
Design your own set of data structure classes from the ground up. Then try to figure out why the standard ones are better.