DEV Community

Cover image for Getting Cozy With C++
Ben Lovy
Ben Lovy

Posted on • Updated on

Getting Cozy With C++

Swallowing The Pill

Some Background

C++ is my self-learning white whale. I've tried many times over the years to make it a bit further though one of the big Stroustrup bibles, but inevitably flame out embarrassingly quickly. It's just huge, and kinda complicated. I was always in awe of it and wanted to be able to leverage its power, but as it was easier to get going with other languages I never got around to taking the time. I never had quite enough self-discipline.

I'm back in school now, so I took advantage of the opportunity and specifically enrolled in the C++-focused track. I figured a good jolt of structured academic instruction might be just the ticket to force myself to put in the time and energy. I'm finally doing a significant month-long final project with it, so it's my first real off-the-rails C++ test drive having now learned more about the language than I'd ever managed before.

It turns out I think I even like it, but boy is it an awakening coming from Rust and JavaScript and C and company.

What This Is

This is an extremely beginner-level look at some stuff I've learned to un-stick myself while implementing this homework assignment. I'm making no claim that I've discovered the best solutions to these problems, this is more a journal of what's ended up working for me. Si vez algo, di algo.

The closest language analogue I'm comfortable with now is Rust (or maybe C, I'm not sure - they're similar for different reasons), which also happens to be what I've been using the most of lately, so I'm approaching this project more or less as I would a Rust project for better or for worse. Quickly I could tell the idioms are pretty different, but you've gotta start from somewhere.

For context, the project is a CLI game of Battleship. The code can be found on GitHub.

Using statements

My first confusion came from namespace etiquette. I knew I didn't like using namespace std, so I decided to go with scope-level using statements:

std::string someFunc()
{
    using std::string;

    string myString = "";
}
Enter fullscreen mode Exit fullscreen mode

This keeps it explicit in the global scope but allows me to pull specific things into a function without sacrificing clarity - you can see where it's coming from right there.

Then I got confused about #include statements - sometimes a header is included through multiple layers of other includes, because the preprocessor is literally just pasting code into other code resolving these. It can be tough to see where a specific function is actually included.

I was pointed to this article, which is worth a read. The biggest takeaway for me was that if you just use a pointer to a specific object, you don't need to actually include it, you can (and probably should) just forward-declare it.

Debugging

Until now, I've mostly been a println debugger. I've known how to use gdb but never saw that as easier than just adding a debug!() output somewhere.

I have now come to very much appreciate gdb. My programs had simply never gotten large enough. A quick rundown of literally everything I need:

  1. Compile with the -g flag.

  2. Invoke gdb my_executable

→ gdb build/volley 
GNU gdb (Gentoo 8.3 vanilla) 8.3
Copyright (C) 2019 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
// blah blah blah
Reading symbols from build/volley...
(gdb)
Enter fullscreen mode Exit fullscreen mode

You enter commands at the (gdb) prompt.

  1. Break on the function you want to inspect:
(gdb) break runFiring
Breakpoint 1 at 0x40e913: file src/game.cpp, line 38.
Enter fullscreen mode Exit fullscreen mode
  1. Use commands to navigate through your program:

a. run/r: run the loaded program for debugging until next breakpoint
b. next/n: step through execution one line at a time, not stepping into functions.
c. step/s: step through execution one line at a time, stepping into all functions
d. print/p: print the value of a variable
e. examine/x: examine memory of a variable
f. continue/c: stop stepping line by line and resume execution to the next breakpoint (or program completion)
g. kill/k: kill the program being debugged without exiting gdb to take it again from the top (use quit/q to get back to your shell)

You can add breakpoints any time, and remove them with delete. Use the up and down arrows to access command history, and leaving it empty and hitting Enter will just repeat the last command - useful for stepping line by line.

There's lots and lots more, there's a great PDF cheatsheet here. I find myself using info locals a lot, which shows you all the variables in the current stack frame.

It is so much better than adding and removing println statements and recompiling. It's much more exploratory and interactive, and a million times more efficient. I still only just barely know how to use it, too.

Clean Up After Yourself

There's a super quick way to check if you've done your job memory-leak wise: valgrind.

This is another tool I do not know how to use but already get immense benefit from:

± |master U:13 ✗| → valgrind build/volley 
==7744== Memcheck, a memory error detector
==7744== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==7744== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
==7744== Command: build/volley
==7744== 

            Battleship!!!

// ... etc - play a game

Game over!
==7744== 
==7744== HEAP SUMMARY:
==7744==     in use at exit: 672 bytes in 8 blocks
==7744==   total heap usage: 3,970 allocs, 3,962 frees, 177,854 bytes allocated
==7744== 
==7744== LEAK SUMMARY:
==7744==    definitely lost: 0 bytes in 0 blocks
==7744==    indirectly lost: 0 bytes in 0 blocks
==7744==      possibly lost: 0 bytes in 0 blocks
==7744==    still reachable: 672 bytes in 8 blocks
==7744==         suppressed: 0 bytes in 0 blocks
==7744== Rerun with --leak-check=full to see details of leaked memory
==7744== 
==7744== For lists of detected and suppressed errors, rerun with: -s
==7744== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
Enter fullscreen mode Exit fullscreen mode

Hold up, heap in use at exit? Ah, of course - I'd written my destructors, but never actually call delete on the top-level instance anywhere! After a quick edit:

$ valgrind build/volley 
==8122== Memcheck, a memory error detector
==8122== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==8122== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
==8122== Command: build/volley
==8122== 

            Battleship!!!

// ... etc - play a game

Game over!
==8122== 
==8122== HEAP SUMMARY:
==8122==     in use at exit: 0 bytes in 0 blocks
==8122==   total heap usage: 3,993 allocs, 3,993 frees, 178,686 bytes allocated
==8122== 
==8122== All heap blocks were freed -- no leaks are possible
==8122== 
==8122== For lists of detected and suppressed errors, rerun with: -s
==8122== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
Enter fullscreen mode Exit fullscreen mode

That's 672 whole bytes now present and accounted for. Boo-yah. All I needed was the nudge to go double-check from just running it with no options, there's also a lot more this tool can do for you.

Struct Equality

Off the bat one of my problems was std::find(). This is used to locate an element in a vector. Clearly, such a function will be comparing elements for equality. In Rust, you'd derive or hand-implement the PartialEq trait on a struct in order to enable that behavior. C++ doesn't have that, but you still need to be able to define equality for structs.

Structs are basically equivalent to classes, but their members are public by default. This is something I knew from textbook-reading, but had never needed to use.

Without providing a definition, you get this somewhat opaque error from clang:

usr/lib/gcc/x86_64-pc-linux-gnu/9.1.0/include/g++-v9/bits/predefined_ops.h:241:17: error: invalid operands to binary expression ('Cell' and 'const Cell')
        { return *__it == _M_value; }
Enter fullscreen mode Exit fullscreen mode

This happens because std::find() tried to use == on two structs, but we hadn't defined how to do that. I think the problem is that it was expecting it to be passed by reference, and instead it got passed by value.

You can allow equality checks to work on structs you define by overloading the == operator and specifically passing a const reference:

// A single cell on the board
typedef struct Cell
{
    int row;
    char col;
    bool operator==(const Cell &other) const
    {
        return row == other.col && col == other.col;
    }
} Cell;
Enter fullscreen mode Exit fullscreen mode

This looks a lot like a handwritten impl PartialEq block (from the docs), which also uses what's essentially the Rust-y const &:

struct Book {
    isbn: i32,
    format: BookFormat,
}

impl PartialEq for Book {
    fn eq(&self, other: &Self) -> bool {
        self.isbn == other.isbn
    }
}

Enter fullscreen mode Exit fullscreen mode

Constantly Const

This leads into the next point - sprinkle const everywhere. This is something Rust has actually prepared me well for. I essentially use it like the opposite of mut. Here's one of my class headers:

class Board
{
    int dimension;
    std::vector<Cell> receivedShots;
    std::vector<Ship> ships;

public:
    Board(int boardSize = BOARD_SIZE);
    bool doesFit(ShipPlacement sp) const;
    char getCharAt(Cell c, bool showShips) const;
    Cell getRandomCell() const;
    Cell promptCell(const std::string &promptStr) const;
    void pushShip(Ship s);
    std::vector<Cell> getAllShots() const;
    bool receiveFire(Cell target);
    int size() const;
    lines toLineStrings(bool showShips) const;
};
Enter fullscreen mode Exit fullscreen mode

This thing is so full of const it's ridiculous. When I started coding I didn't realize quite how much it would be applicable and was hesitant to use it for fear of not understanding it. Now my rule of thumb is to add it by default to any method, and only take it away if I'm sure I cannot have it.

My God, It's Full Of Streams

C++ leans hard into the stream abstraction. I ran into this relatively quickly when I wanted to pretty-print some data. In Rust I'd reach for impl Display, in something more OOP i'd override toString() or something.

In C++, you actually overload the << stream insertion operator. For a simple example:

enum Direction
{
    Left,
    Down
};

std::ostream &operator<<(std::ostream &stream, const Direction &d)
{
    if (d == Direction::Left)
        return stream << "Left";
    else
        return stream << "Down";
}
Enter fullscreen mode Exit fullscreen mode

Now you can pop it right in a stream, no need to call anything:

std::cout << "Direction: " << direction << "!" << std::endl;
Enter fullscreen mode Exit fullscreen mode

This pattern was not obvious to me at first but feels a lot more natural after a few days.

Overloaded Constructors

I've never worked with a language that does this, so it's still novel and neat to me. In Rust, you use traits and it's a little more unwieldy. In C++ I can just literally define three constructors:

class ShipClass
{
  // ..
public:
    ShipClass();
    ShipClass(char c);
    ShipClass(ShipClassType sc);
 // ..
}
Enter fullscreen mode Exit fullscreen mode

That's a pretty easy way to get yourself a flexible API.

Frustrations

Not everything has been happy. I've generally expected everything C++ has had to throw at me given prior knowledge, but there are a few outstanding things I'm still not sure how to learn to like.

Build tooling / Package Management

C++ is, to me, the wild friggin' west.

I haven't gotten started with things like CMake and Autotools, but the very fact these tools exist says a lot. It's really hard to just use external libraries, so often projects will simply just not. There's a lot of reinventing the wheel because package management is such a complete mess. That's not a healthy ecosystem to the untrained eye, but the language itself is powerful enough that maybe it makes up for it. It also is a gigantic ecosystem despite this shortcoming, so I want to be able to explore and use it, but if it's so complicated I won't bother.

Then there's things like boost which are just their own beasts in and of themselves. I think it will quite literally be years until I'm able to make a reasonable and informed statement about the power and quality of the C++ ecosystem. Until then, it's a newbie turn-off.

I've already written a post about make, which I won't recreate here. The second example I walk through in that post is the exact Makefile I'm using to build this project.

This is the very first C++ course in the curriculum, and I imagine it will be covered later on, but for this one the professor basically said "I do not care how you build your code, just make sure I can recreate it, if you don't know what to do, here's a link to download Visual Studio".

I should probably learn Visual Studio sometime, but I think it's easier to learn one thing at a time, so I just stuck with my usual text editor and compilation via CLI. I already knew how to use make from years of tinkering with Linux. I don't know what the best way to go about this is. It seems like in a professional setting CI/CD would run all the compilers anyway.

Exceptions

This wasn't difficult to pick up, being not dissimilar from exceptions in JS or Python:

try
{
    row = stoi(originStr.substr(1, originStr.size() - 1));
}
catch (const std::invalid_argument &ia)
{
    std::cerr << "Please enter a number as your second term." << endl;
    // ..
}
Enter fullscreen mode Exit fullscreen mode

I guess I'm just spoiled with type-level stuff. I don't like this, it seems like it seems to spaghetti-type code and lots of needless verbosity to catch errors for a whole app. I also haven't really started mucking about with templates much, so it's likely this is a familiarity issue. It does seem to work more or less the same as I expect from Python or JS.

Class vs Struct

This isn't a problem, but still doesn't really make sense to me. In Rust, everything is a struct, and you can provide an impl Struct block to define a constructor and/or methods if needed, or just not for plain data.

C++ has structs and classes, but they're almost identical. The only difference is the default visibility: structs are public, classes are private, but through explicit annotations they're functionally equivalent otherwise. I try to use structs for plain data and classes for anything else, but the line is blurry. If I have a struct that just wraps an enum but has a bunch of different getters, like a char and a string and an int, is that a class or a struct? Right now I have a struct which just holds a row and a column, also has some constructors and an equality method defined. That's not that different from a class. I don't know which, if either, is correct, or if it matters at all. I'm just kind making a gut decision when I define a new one and not thinking about it again, it doesn't seem to make a difference.

Follow-up: is that the kind of thing you'd use a union for? I still don't quite know when I'd want one of those unless I'm specifically space constrained.

Conclusion

I'm glad I'm finally ripping the band-aid and using C++ for something a little more substantial, but never before has the vastness of the mountain been so apparent at the outset.

Photo by Matthew Henry on Unsplash

Top comments (29)

Collapse
 
dwd profile image
Dave Cridland

As a long-time C++ developer, this is absolutely spot on. One change you might wish to make:

If having std::find is forcing you into defining an operator==() that you didn't otherwise want, you might want to look at std::find_if, which takes a UnaryPredicate - basically a functor accepting a single value. Conveniently, this can be a lambda.

So you'd do:

std::find_if(vec.begin(), vec.end(), [&t](Cell const & other) {
    return t.row == other.col && t.col == other.col;
}

It achieves the same thing (and, likely, with the same CPU instructions), but you've kept the required definition of equality for this purpose next to the point of use, instead of defining it globally.

Collapse
 
deciduously profile image
Ben Lovy

Ah, perfect! That does make sense, I'll probably refactor.

Collapse
 
banzyme2 profile image
ENDEESA

Typo: t.row == other.col

Collapse
 
vimmer9 profile image
Damir Franusic

Great article. I spent the last 8 years doing only c++ network programming. Then after thinking things through, I thought to myself that this is overly complex for what I'm doing and OOP is something I can live without. Also, I can easily replicate it with function pointers which I also use in C++ anyway.

This was just a quick intro, you should stick with C++ if you like it; you wrote a great post about it anyway. Now, here's the reason I'm writing this comment; there's an alternative to Valgrind which you could consider using once you get more acquainted with C++. Valgrind incurrs 20x slowdon or was 10x I don't rememeber honestly, but it can get quite slow and eat up a huge amount of memory. The other alternative that doesn't cause any noticeable slowdown are gcc/clang fsanitize methods

gcc -ggdb -o a.out a.cpp \
-fsanitize=address \
-fno-omit-frame-pointer

When I first started coding in C++ I also usw Valgrind; not even sure if fsanitize methods were available back then.

Happy programming and congrats on this great post.

DF

Collapse
 
deciduously profile image
Ben Lovy

Ah, thank you for the tip! I didn't know about that flag.

Collapse
 
vimmer9 profile image
Damir Franusic

It's a series of flags, check out the man pages of gcc/clang or look it up online.

Good luck with segfaults 😉

Collapse
 
codemouse92 profile image
Jason C. McDonald

Great article! Definitely saving this for newcomers to the language.

One other distinction that is easily forgotten is enum vs. enum class. The former is more of the C way of doing things, while the latter provides all the type safety you'd want out of an enumeration.

Collapse
 
deciduously profile image
Ben Lovy • Edited

Whoa! Thanks a bunch, I missed this detail entirely. We didn't go over this in this course, they only showed us the old-style enum.

This exactly what I want, and I'm glad I've got time to refactor before I submit this project!

Collapse
 
pgradot profile image
Pierre Gradot • Edited

I also wanted to point this out: try to use enum classes only. There are a lot of articles out there explaining why ;)

Also, take a look at Magic Enum to easily print enum values. Include it and then simply:

std::cout << magic_enum::name(value);

This change my life this year :D

Collapse
 
amorganpd profile image
amorganPD

Thanks for sharing your learnings!

So it looks like structs and classes do only differ in their default access and nothing else:

justsoftwaresolutions.co.uk/cplusp...

Looks like really only for C compatibility, which makes sense. I would default to class when writing for just C++ unless it's a small data structure.

Collapse
 
lucpattyn profile image
Mukit, Ataul • Edited

struct and classes looks like the same thing, but using the keyword class instead of struct gives an indication to the reader (of your code) that you are trying to abstract a real world concept into your class. For example, it can be an abstraction/conceptualisation of the animal kingdom,
i.e, a human inherits its traits from a mammal which inherits from animal. Yes, you can use the keyword struct too, but usually struct should be used to conceptualise a data structure to pass around for use, for example a leaf node inside a tree. Yes, classes and structures are interchangeable but following a convention shows professionalism and intention behind the code.

Collapse
 
deciduously profile image
Ben Lovy

This is a really great response, and aligns pretty well with the intuition I'd sort of already been building. Thanks for putting it so clearly, I will keep this in mind.

Collapse
 
mg_verycoolthx profile image
mg, thx

Thank you for this! I am new to C++ and to this entire field itself, and I found your perspective to be very useful in terms of grounding my love and appreciation for this language. As an aside, I've heard through the grapevine that the ecosystem surrounding C++ is on the mend and its popularity on the rise. It's hard not to be excited!

Collapse
 
deciduously profile image
Ben Lovy • Edited

I agree! I still plan to get my head around C++17 first, but C++20 does propose at least part of the way forward here.

I really like looking into all the new cute hipster languages with new cool features, and I've seen some new industry trends that might buck long-held standards, but I have never yet seen anything that could possibly replace C++.

Collapse
 
mg_verycoolthx profile image
mg, thx

Heyy, that's good for me, then! C++ is close enough to the hardware and far enough from it at the same time that it's really in an ideal position to be irreplaceable I think, barring a total revolution in computing altogether. Here's to a long life for our friend!

Collapse
 
adam_cyclones profile image
Adam Crockett 🌀 • Edited

I want to learn Rust because I actually like how hard c++ is, and that is disturbing. But I keep telling myself I have come this far. It's like you said, it can be hard to know where stuff is coming from, that's my biggest bugbear with not the language but with my inexpensive. In JavaScript I can do some crazy things, but in C++ I feel like I wound back the clock several years and I'm a newbie. That excites me and terrifies me!

Collapse
 
spqrbob profile image
Bob McCann

The Visual Studio IDE (visualstudio.microsoft.com/) is a heavy lift, but does have quite good support for C++. With all the good things everyone has to say about Visual Studio Code (code.visualstudio.com/), I wouldn't be surprised if there are some excellent C++ plugins for that platform as well. I have also heard and read good things about Jetbrains' CLion (jetbrains.com/clion/).

All of that being said (and linked), there are many folks who neither want nor need a memory-hogging GUI IDE getting in between them and their code. I have a close personal friend who is quite the Vim evangelist, and another who just uses notepad or whatever other simple text editor is at hand.

It sounds like you have a good start into the joy and frustration that is C++. Your article gave me flashbacks to my own college coursework in the language, working with threads and smart pointers and overloaded constructor/destructor/deep copy classes (lions and tigers and bears, oh my!). Good luck and have fun!

Collapse
 
deciduously profile image
Ben Lovy • Edited

Thanks for the tip! I'm currently using VS Code, and have spent a little time setting up a workflow in Emacs as well - for now that's more my speed. I also keep hearing good things about CLion, and have generally had good experiences with other JetBrains products, so that's probably where I'll look if VS Community isn't my style.

I've only ever mucked with threads from within the safe compiler-enforced confines of Rust, so I'm a little terrified to do it manually myself! Should be fun :)

Collapse
 
siy profile image
Sergiy Yevtushenko

Yeah, CLion is quite good. You might also take a look at NetBeans.

Collapse
 
siy profile image
Sergiy Yevtushenko

Union sometimes useful in cases, when different types of data can be stored within same data structure. For example, sometimes it is convenient to have different "view" of same data, like integers as series of bytes and vice versa. In practice it's rarely necessary, quite low level and often error prone. There is an type safe alternative to unions - std:variant.

Collapse
 
deciduously profile image
Ben Lovy

Ah, that makes sense - I've been using std::variant. Seems like unions are most useful for interior with C that uses them.

Collapse
 
pgradot profile image
Pierre Gradot • Edited

As of C++17, you should use std::variant instead of plain old unions :)

Collapse
 
pgradot profile image
Pierre Gradot

Hey!

A few (other) comments that may help :)

"C++ has structs and classes, but they're almost identical. The only difference is the default visibility: structs are public, classes are private, but through explicit annotations they're functionally equivalent otherwise."

Before talking why we have both, I wanted to point about a second important difference (this is not clearly said in your sentence): default inheritance is not the same. See here:

"If access-specifier is omitted, it defaults to public for classes declared with class-key struct and to private for classes declared with class-key class."

Try to modify the solution code to replace struct with class and see when it compiles and when it doesn't:

struct Foo {};

struct Bar : Foo {};

void doSomething(const Foo&);

int main() {
    Bar bar;
    doSomething(bar);
}

So "why struct AND class?". Well... C++ came from C and was supposed to be compatible. So we have struct and class for this reason only. Sorry.

From my experience, there is no real reason to use one why over the other. A quite common way is indeed to use struct for basic data type, where all members are public (and most of the time with no member function). Example:

struct Point {
  int x
  int y;
};

Class is used when you are modeling an object with behavior:

class NetworkConnection {
public:
   // ctor(s)
   void open();
   void close();
   void setTimeout(int);
private:
   // we don't care this is private
};

This is just a convention, there is no technical reason. Do what you want, simply be coherent.

By the way:

typedef struct Cell
{
    // ...
} Cell;

This is C-style code! Get rid of this typedef :)

struct Cell
{
    // ...
};

Finally:

"The biggest takeaway for me was that if you just use a pointer to a specific object, you don't need to actually include it, you can (and probably should) just forward-declare it."

First: try to use references at much as possible. If not possible, try to use smart pointers as much as possible.

Second: be prudent with this technique. Remember that including your class header should be enough for client code to compile. If you forward-declare a type in your HPP that is only needed in your CPP for the class implementation, no problem. If class users must also include the good headers, this can be painful.

Example with Foo.hpp:

#pragma once

class Foo;

struct Bar {
   Foo* foo;
};

If I include Foo.hpp in mySource.cpp, it won't compile. And I will have to wonder what other files I should include.

This may sound obvious but I saw code like this in real life...

Good luck with C++ ^^