Tag Archives: D

C is not magically fast, part 2

I wrote a blog post before about how C is not magically fast, but the sentiment that C has properties lacking in other languages that make it so is still widespread. It was with no surprise at all then that a colleague mentioned something resembling that recently at lunch break, and I attempted to tell him why it wasn’t (at least always) true.

He asked for an example where C++ would be faster, and I resorted to the old sort classic: C++ sort is faster than C’s qsort because of templates and inlining. He then asked me if I’d ever measured it myself, and since I hadn’t, I did just that after lunch. I included D as well because, well, it’s my favourite language. Taking the minimum time after ten runs each to sort a random array of 10M simple structs on my laptop yielded the results below:

  • D: 1.147s
  • C++: 1.723s
  • C: 1.789s

I expected  C++ to be faster than C, I didn’t expect the difference to be so small. I expected D to be the same speed as C++, but for some reason it’s faster. I haven’t investigated the reason why for lack of interest, but maybe because of how strings are handled?

I used the same compiler backend for all 3 so that wouldn’t be an influence: LLVM. I also seeded all of them with the same number and used the same random number generator: the awful srand from C’s standard library. It’s terrible, but it’s the only easy way to do it in standard C and the same function is available from the other two languages. I also only timed the sort, not counting init code.

The code for all 3 implementations:

// sort.c
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>
#include <sys/resource.h>

typedef struct {
    int i;
    char* s;
} Foo;

double get_time() {
    struct timeval t;
    struct timezone tzp;
    gettimeofday(&t, &tzp);
    return t.tv_sec + t.tv_usec*1e-6;

int comp(const void* lhs_, const void* rhs_) {
    const Foo *lhs = (const Foo*)lhs_;
    const Foo *rhs = (const Foo*)rhs_;
    if(lhs->i < rhs->i) return -1;
    if(lhs->i > rhs->i) return 1;
    return strcmp(lhs->s, rhs->s);

int main(int argc, char* argv[]) {
    if(argc < 2) {
        fprintf(stderr, "Must pass in number of elements\n");
        return 1;

    const int size = atoi(argv[1]);
    Foo* foos = malloc(size * sizeof(Foo));
    for(int i = 0; i < size; ++i) {
        foos[i].i = rand() % size;
        foos[i].s = malloc(100);
        sprintf(foos[i].s, "foo%dfoo", foos[i].i);

    const double start = get_time();
    qsort(foos, size, sizeof(Foo), comp);
    const double end = get_time();
    printf("Sort time: %lf\n", end - start);

    return 0;

// sort.cpp
#include <iostream>
#include <algorithm>
#include <string>
#include <vector>
#include <chrono>
#include <cstring>

using namespace std;
using namespace chrono;

struct Foo {
    int i;
    string s;

    bool operator<(const Foo& other) const noexcept {
        if(i < other.i) return true;
        if(i > other.i) return false;
        return s < other.s;


template<typename CLOCK, typename START>
static double getElapsedSeconds(CLOCK clock, const START start) {
    //cast to ms first to get fractional amount of seconds
    return duration_cast<milliseconds>(clock.now() - start).count() / 1000.0;

#include <type_traits>
int main(int argc, char* argv[]) {
    if(argc < 2) {
        cerr << "Must pass in number of elements" << endl;
        return 1;

    const int size = stoi(argv[1]);
    vector<Foo> foos(size);
    for(auto& foo: foos) {
        foo.i = rand() % size;
        foo.s = "foo"s + to_string(foo.i) + "foo"s;

    high_resolution_clock clock;
    const auto start = clock.now();
    sort(foos.begin(), foos.end());
    cout << "Sort time: " << getElapsedSeconds(clock, start) << endl;

// sort.d
import std.stdio;
import std.exception;
import std.datetime;
import std.algorithm;
import std.conv;
import core.stdc.stdlib;

struct Foo {
    int i;
    string s;

    int opCmp(ref Foo other) const @safe pure nothrow {
        if(i < other.i) return -1;
        if(i > other.i) return 1;
        return s < other.s
            ? -1
            : (s > other.s ? 1 : 0);

void main(string[] args) {
    enforce(args.length > 1, "Must pass in number of elements");
    immutable size = args[1].to!int;
    auto foos = new Foo[size];
    foreach(ref foo; foos) {
        foo.i = rand % size;
        foo.s = "foo" ~ foo.i.to!string ~ "foo";

    auto sw = StopWatch();
    writeln("Elapsed: ", cast(Duration)sw.peek);

Tagged , ,

main is just another function

Last week I talked about code that isn’t unit-testable, at least not by my definition of what a unit test is. In keeping with that, this blog post will talk about testing code that has side-effects.

Recently I’d come to accept a defeatist attitude where I couldn’t think of any other way to test that passing certain command-line options to a console binary had a certain effect. I mean, the whole point is to test that running the app differently will have different consequences. As a result I ended up only ever doing end-to-end testing. And… that’s simply not where I want to be.

Then it dawned on me: main is just another function. Granted, it has a special status that makes it so you can’t call it directly from a test, but nearly all my main functions lately have looked like this:

int main(string[] args) {
    try {
        return 0;
    } catch(Exception ex) {
        return 1;

It should be easy enough to translate this to the equivalent C++ in your head. With main so conveniently delegating to a function that does real work, I can now easily write integration tests. After all, is there really any difference between:

doStuff(["myapp", "--option", "arg1", "arg2"]);
// assert stuff happened

And (in, say, a shell script):

./myapp --option arg1 arg2
# assert stuff happened

I’d say no. This way I have one end-to-end test for sanity’s sake, and everything else being tested from the same binary by calling the “real” main function directly.

If your main doesn’t look like the one above, and you happen to be writing C or C++, there’s another technique: use the preprocessor to rename main to something else and call it from your integration/component test. And then, as they say, Bob’s your uncle.

Happy testing!

Tagged , , ,

unit-threaded: now an executable library

It’s one of those ideas that seem obvious in retrospect, but somehow only ocurred to me last week. Let me explain.

I wrote a unit testing library in D called unit-threaded. It uses D’s compile-time reflection capabilities so that no test registration is required. You write your tests, they get found automatically and everything is good and nice. Except… you have to list the files you want to reflect on, explicitly. D’s compiler can’t go reading the filesystem for you while it compiles, so a pre-build step of generating the file list was needed. I wrote a program to do it, but for several reasons it wasn’t ideal.

Now, as someone who actually wants people to use my library (and also to make it easier for myself), I had to find a way so that it would be easy to opt-in to unit-threaded. This is especially important since D has built-in unit tests, so the barrier for entry is low (which is a good thing!). While working on a far crazier idea to make it a no-brainer to use unit-threaded, I stumbled across my current solution: run the library as an executable binary.

The secret sauce that makes this work is dub, D’s package manager. It can download dependencies to compile and even run them with “dub run”. That way, a user need not even have to download it. The other dub feature that makes this feasible is that it supports “configurations” in which a package is built differently. And using those, I can have a regular library configuration and an alternative executable one. Since dub run can take a configuration as an argument, unit-threaded can now be run as a program with “dub run unit-threaded -c gen_ut_main”. And when it is, it generates the file that’s needed to make it all work.

So now all a user need to is add a declaration to their project’s dub.json file and “dub test” works as intended, using unit-threaded underneath, with named unit tests and all of them running in threads by default. Happy days.

Tagged , , , , ,

Rust impressions from a C++/D programmer, part 1

Discussion on programming reddit

Discussion on Rust reddit

C++ and D aren’t the only languages I know, I labeled myself that way in the title because as far as learning Rust is concerned, I figured they would be the most relevant in terms of the audience knowing where I’m coming from.

Since two years ago, my go-to task for learning a new programming language is to implement an MQTT broker in it. It was actually my 3rd project in D, but my first in Haskell and now that I have some time on my hands, it’s what I’m using to learn Rust. I started last week and have worked on it for about 3 days. As expected, writing an MQTT broker is a great source of insight into how a language really is. You know, the post-lovey-dovey phase. It’s like moving in together straight away instead of the first-date-like “here’s how you write a Scheme interpreter”.

I haven’t finished the project yet, I’m probably somewhere around the 75% mark, which isn’t too shabby for 3 days of work. Here are my impressions so far:

The good

The borrow checker. Not surprising since this is basically the whole point of the language. It’s interesting how much insight it gave me in how broken the code I’m writing elsewhere might be.  This will be something I can use when I write in other systems languages, like how learning Haskell makes you wary of doing IO.

Cargo. Setting up, getting started, using someone’s code and unit testing what you write as you go along is painless and just works. Tests in parallel by default? Yes, please. I wonder where I’ve seen that before…

Traits. Is there another language other than D and Rust that make it this easy to use compile-time polymorphism? If there is, please let me know which one. Rust has an advantage here: as in Dylan (or so I read), the same trait can be used for runtime polymorphism.

Warnings. On by default, and I only had to install flycheck-rust in Emacs for syntax highlighting to just work. Good stuff.

Productivity. This was surprising, given the borrow checker’s infamy. It _does_ take a while to get my code to compile, but overall I’ve been able to get a good amound done with not that much time, given these are the first lines of Rust I’ve ever written.

Algebraic types and pattern matching. Even though I didn’t use the former.

Slices. Non-allocating views into data? Yes, please. Made the D programmer in me feel right at home.

Immutable by default. Need I say more?

Debugging. rust-gdb makes printing out values easy. I couldn’t figure out how to break on certain functions though, so I had to use the source file and line number instead.

No need to close a socket due to RAII. This was nice and even caught a bug for me. The reason being that I expected my socket to close because it was dropped, but my test failed. When I looked into it, the reference count was larger than 1 because I’d forgotten to remove the client’s subscriptions. The ref count was 0, the socket was dropped and closed, and the test passed. Nice.

No parens for match, if, for, …

The bad

The syntax. How many times can one write an ampersand in one’s source code? You’ll break new records. Speaking of which…

Explicit borrows. I really dislike the fact that I have to tell the compiler that I’m the function I’m calling is borrowing a parameter when the function signature itself only takes borrows. It won’t compile otherwise (which is good), but… since I can’t get it wrong what’s the point of having to express intent? In C++:

void fun(Widget& w);
auto w = Widget();
fun(w); //NOT fun(&w) as in Rust

In Rust:

fn fun(w: &mut Widget);
let w = Widget::new();
fun(&mut w); //fun(w) doesn't compile but I still need to spell out &mut. Sigh.

Display vs Debug. Printing out integers and strings with {} is fine, but try and do that with a Vec or HashMap and you have to use the weird {:?}. I kept getting the order of the two symbols wrong as well. It’s silly. Even the documentation for HashMap loops over each entry and prints them out individually. Ugh.

Having to rethink my code. More than once I had to find a different way to do the thing I wanted to do. 100% of the time it was because of the borrow checker. Maybe I couldn’t figure out the magical incantation that would get my code to compile, but in one case I went from “return a reference to an internal object, then call methods on it” to “find object and call method here right now”. Why? So I wouldn’t have to borrow it mutably twice. Because the compiler won’t let me. My code isn’t any safer and it was just annoying.

Rc<RefCell<T>> and Arc<Mutex<T>>. Besides the obvious “‘Nuff said”, why do I have to explicitly call .clone on Rc? It’s harder to use than std::shared_ptr.

Slices. Writing functions that slices and passing them vectors works well enough. I got tired of writing &var[..] though. Maybe I’m doing something wrong. Coming from D I wanted to avoid vectors and just slice arrays instead. Maybe that’s not Rusty. What about appending together some values to pass into a test? No Add impl for Vecs, so it’s massive pain. Sigh.

Statements vs Expressions. I haven’t yet made the mistake of forgetting/adding a semicolon, but I can see it happening.

No function overloading.

Serialization. There’s no way to do it well without reflection, and Rust is lacking here. I just did everything by hand, which was incredibly annoying. I’m spoiled though, in D I wrote what I think is a really good serialization library. Good in the lazy sense, I pretty much never have to write custom serialization code.

The ugly

Hashmaps. The language has operator overloading, but HashMap doesn’t use it. So it’s a very Java-like map.insert(key, value). If you want to create a HashMap with a literal… you can’t. There’s no equivalent macro to vec. You could write your own, but come on, this is a basic type from the standard library that will get used a lot. Even C++ does better!

Networking / concurrent IO. So I took a look at what my options were, and as far as my googling took me, it was to use native threads or a library called mio. mio’s API was… not the easiest to use so I punted and did what is the Rust standard library way of writing a server and used threads instead. I was sure I’d have performance problems down the road but it was something to worry about later. I went on writing my code, TDDed an implementation of a broker that wasn’t connected to the outside world and everything. At one point I realised that holding on to a mutable reference for subscribers wasn’t going to work so I used Rc<RefCell<Subscriber>> instead. It compiled, my tests passed, and all was good in the world. Then I tried actually using the broker from my threaded server. Since it’s not safe to use Rc<RefCell<>> in threads, this failed to compile. “Good!”, I thought, I changed Rc to Arc and RefCell to Mutex. Compile, run, …. deadlock. Oops. I had to learn mio after all. It wasn’t as bad as boost::asio but it wasn’t too far away either.

Comparing objects for identity. I just wanted to compare pointers. It was not fun. I had to write this:

fn is_same<T>(lhs: &T, rhs: &T) -> bool {
    lhs as *const T == rhs as *const T;
fn is_same_subscriber<T: Subscriber>(lhs: Rc<RefCell<T>>, rhs: Rc<RefCell<T>>) -> bool {
    is_same(&*lhs.borrow, &*rhs.borrow());



I thought I’d like Rust more than I actually do at this point. I’m glad I’m taking the time to learn it, but I’m not sure how likely I’ll choose to use it for any future project. Currently the only real advantage it has for me over D is that it has no runtime and could more easily be used on bare metal projects. But I’m unlikely to do any of those anytime soon.

I never thought I’d say this a few years back but…I like being able to fall back on a mark-and-sweep GC. I don’t have to use it in D either, so if it ever becomes a performance or latency problem I know how to deal with it. It seems to me to be far easier than getting the borrow checker to agree with me or having to change how I want to write my code.

We’ll see, I guess. Optimising the Rust implementation to be competitive with the D and Java ones is likely to be interesting.

Tagged , , , ,

My first D Improvement Proposal

After over a year of thinking about this (I remember bringing this up at DConf 2014), I finally wrote a DIP about introducing what I call “static inheritance” to D.

The principle is similar to C++ concepts: D already has a way of requiring a certain compile-time interface for templates to be instantiated, the most common being isInputRange. What is lacking right now in my opinion is helpful compiler error messages for when a type was intended to be, e.g. an input range but isn’t due to programmer error.

I tried a library solution first, assuming this would be easier to get accepted than a language change. Since there was nearly 0 interest, I had to write a DIP. Here’s hoping it’s more succesful.

Tagged ,

The craziest code I ever wrote

A few years ago at work my buddy Jeff was as usual trying to do something in Go. I can’t remember why, but he wanted to arrange text strings in memory so that they were all contiguous. I said something about C++ and he remarked that the only thing C++11 could do that Go couldn’t would be perhaps to do this work at compile-time. I hadn’t learned D yet (which would have made the task trivial), so I spent the rest of the day writing the monstrosity below for “teh lulz”. It ended up causing my first ever question on Stackoverflow. “Enjoy” the code:

//Arrange strings contiguously in memory at compile-time from string literals.
//All free functions prefixed with "my" to faciliate grepping the symbol tree
//(none of them should show up).

#include <iostream>

using std::size_t;

//wrapper for const char* to "allocate" space for it at compile-time
template<size_t N>
struct String {
    //C arrays can only be initialised with a comma-delimited list
    //of values in curly braces. Good thing the compiler expands
    //parameter packs into comma-delimited lists. Now we just have
    //to get a parameter pack of char into the constructor.
    template<typename... Args>
    constexpr String(Args... args):_str{ args... } { }
    const char _str[N];

//takes variadic number of chars, creates String object from it.
//i.e. myMakeStringFromChars('f', 'o', 'o', '') -> String<4>::_str = "foo"
template<typename... Args>
constexpr auto myMakeStringFromChars(Args... args) -> String<sizeof...(Args)> {
    return String<sizeof...(args)>(args...);

//This struct is here just because the iteration is going up instead of
//down. The solution was to mix traditional template metaprogramming
//with constexpr to be able to terminate the recursion since the template
//parameter N is needed in order to return the right-sized String<N>.
//This class exists only to dispatch on the recursion being finished or not.
//The default below continues recursion.
template<bool TERMINATE>
struct RecurseOrStop {
    template<size_t N, size_t I, typename... Args>
    static constexpr String<N> recurseOrStop(const char* str, Args... args);

//Specialisation to terminate recursion when all characters have been
//stripped from the string and converted to a variadic template parameter pack.
struct RecurseOrStop<true> {
    template<size_t N, size_t I, typename... Args>
    static constexpr String<N> recurseOrStop(const char* str, Args... args);

//Actual function to recurse over the string and turn it into a variadic
//parameter list of characters.
//Named differently to avoid infinite recursion.
template<size_t N, size_t I = 0, typename... Args>
constexpr String<N> myRecurseOrStop(const char* str, Args... args) {
    //template needed after :: since the compiler needs to distinguish
    //between recurseOrStop being a function template with 2 paramaters
    //or an enum being compared to N (recurseOrStop < N)
    return RecurseOrStop<I == N>::template recurseOrStop<N, I>(str, args...);

//implementation of the declaration above
//add a character to the end of the parameter pack and recurse to next character.
template<bool TERMINATE>
template<size_t N, size_t I, typename... Args>
constexpr String<N> RecurseOrStop<TERMINATE>::recurseOrStop(const char* str,
                                                            Args... args) {
    return myRecurseOrStop<N, I + 1>(str, args..., str[I]);

//implementation of the declaration above
//terminate recursion and construct string from full list of characters.
template<size_t N, size_t I, typename... Args>
constexpr String<N> RecurseOrStop<true>::recurseOrStop(const char* str,
                                                       Args... args) {
    return myMakeStringFromChars(args...);

//takes a compile-time static string literal and returns String<N> from it
//this happens by transforming the string literal into a variadic paramater
//pack of char.
//i.e. myMakeString("foo") -> calls myMakeStringFromChars('f', 'o', 'o', '');
template<size_t N>
constexpr String<N> myMakeString(const char (&str)[N]) {
    return myRecurseOrStop<N>(str);

//Simple tuple implementation. The only reason std::tuple isn't being used
//is because its only constexpr constructor is the default constructor.
//We need a constexpr constructor to be able to do compile-time shenanigans,
//and it's easier to roll our own tuple than to edit the standard library code.

//use MyTupleLeaf to construct MyTuple and make sure the order in memory
//is the same as the order of the variadic parameter pack passed to MyTuple.
template<typename T>
struct MyTupleLeaf {
    constexpr MyTupleLeaf(T value):_value(value) { }
    T _value;

//Use MyTupleLeaf implementation to define MyTuple.
//Won't work if used with 2 String<> objects of the same size but this
//is just a toy implementation anyway. Multiple inheritance guarantees
//data in the same order in memory as the variadic parameters.
template<typename... Args>
struct MyTuple: public MyTupleLeaf<Args>... {
    constexpr MyTuple(Args... args):MyTupleLeaf<Args>(args)... { }

//Helper function akin to std::make_tuple. Needed since functions can deduce
//types from parameter values, but classes can't.
template<typename... Args>
constexpr MyTuple<Args...> myMakeTuple(Args... args) {
    return MyTuple<Args...>(args...);

//Takes a variadic list of string literals and returns a tuple of String<> objects.
//These will be contiguous in memory. Trailing '' adds 1 to the size of each string.
//i.e. ("foo", "foobar") -> (const char (&arg1)[4], const char (&arg2)[7]) params ->
//                       ->  MyTuple<String<4>, String<7>> return value
template<size_t... Sizes>
constexpr auto myMakeStrings(const char (&...args)[Sizes]) -> MyTuple<String<Sizes>...> {
    //expands into myMakeTuple(myMakeString(arg1), myMakeString(arg2), ...)
    return myMakeTuple(myMakeString(args)...);

//Prints tuple of strings
template<typename T> //just to avoid typing the tuple type of the strings param
void printStrings(const T& strings) {
    //No std::get or any other helpers for MyTuple, so intead just cast it to
    //const char* to explore its layout in memory. We could add iterators to
    //myTuple and do "for(auto data: strings)" for ease of use, but the whole
    //point of this exercise is the memory layout and nothing makes that clearer
    //than the ugly cast below.
    const char* const chars = reinterpret_cast<const char*>(&strings);
    std::cout << "Printing strings of total size " << sizeof(strings);
    std::cout << " bytes:\n";
    std::cout << "-------------------------------\n";

    for(size_t i = 0; i < sizeof(strings); ++i) {
        chars[i] == '' ? std::cout << "\n" : std::cout << chars[i];

    std::cout << "-------------------------------\n";
    std::cout << "\n\n";

int main() {
        constexpr auto strings = myMakeStrings("foo", "foobar",
                                               "strings at compile time");

        constexpr auto strings = myMakeStrings("Some more strings",
                                               "just to show Jeff to not try",
                                               "to challenge C++11 again :P",
                                               "with more",
                                               "to show this is variadic");

    std::cout << "Running 'objdump -t |grep my' should show that none of the\n";
    std::cout << "functions defined in this file (except printStrings()) are in\n";
    std::cout << "the executable. All computations are done by the compiler at\n";
    std::cout << "compile-time. printStrings() executes at run-time.\n";
Tagged , , , , , , ,

Computer languages: ordering my favourites

This isn’t even remotely supposed to be based on facts, evidence, benchmarks or anything like that. You could even disagree with what are “scripting languages” or not. All of the below just reflect my personal preferences. In any case, here’s my list of favourite computer languages, divided into two categories: scripting and, err… I guess “not scripting”.


My favourite scripting languages, in order:

  1. Python
  2. Ruby
  3. Emacs Lisp
  4. Lua
  5. Powershell
  6. Perl
  7. bash/zsh
  8. m4
  9. Microsoft batch files
  10. Tcl


I haven’t written enough Ruby yet to really know. I suspect I’d like it more than Python but at the moment I just don’t have enough experience with it to know its warts. Even I’m surprised there’s something below Perl here but Tcl really is that bad. If you’re wondering where PHP is, well I don’t know because I’ve never written any but from what I’ve seen and heard I’d expect it to be (in my opinion of course) better than Tcl and worse than Perl. I’m surprised how high Perl is given my extreme dislike for it. When I started thinking about it I realised there’s far far worse.


My favourite non-scripting languages, in order:

  1. D
  2. C++
  3. Haskell
  4. Common Lisp
  5. Rust
  6. Java
  7. Go
  8. Objective C
  9. C
  10. Pascal
  11. Fortran
  12. Basic / Visual Basic

I’ve never used Scheme, if that explains where Common Lisp is. I’m still learning Haskell so not too sure there. As for Rust, I’ve never written a line of code in it and yet I think I can confidently place it in the list, especially with respect to Go. It might place higher than C++ but I don’t know yet.


Tagged , , , , , , , , , , , , ,

To learn BDD with Cucumber, you must first learn BDD with Cucumber.

So I read about Cucumber a while back and was intrigued, but never had time to properly play with it. While writing my MQTT broker, however, I kept getting annoyed at breaking functionality that wasn’t caught by unit tests. The reason being that the internals were fine, the problems I was creating had to do with the actual business of sending packets. But I was busy so I just dealt with it.

A few weeks ago I read a book about BDD with Cucumber and RSpec but for me it was a bit confusing. The reason being that since the step definitions, unit tests and implementation were all written in Ruby, it was hard for me to distinguish which part was what in the whole BDD/TDD concentric cycles. Even then, I went back to that MQTT project and wrote two Cucumber features (it needs a lot more but since it works I stopped there). These were easy enough to get going: essentially the step definitions run the broker in another process, connect to it over TCP and send packets to it, evaluating if the response was the expected one or not. Pretty cool stuff, and it works! It’s what I should have been doing all along.

So then I started thinking about learning BDD (after all, I wrote the features for MQTT afterwards) by using it on a D project. So I investigated how I could call D code from my step definitions. After spending the better part of an afternoon playing with Thrift and binding Ruby to D, I decided that the best way to go about this was to implement the Cucumber wire protocol. That way a server would listen to JSON requests from Cucumber, call D functions and everything would work. Brilliant.

I was in for a surprise though, me who’s used to implementing protocols after reading an RFC or two. Instead of a usual protocol definition all I had to go on was… Cucumber features! How meta. So I’d use Cucumber to know how to implement my Cucumber server. A word to anyone wanting to do this in another language: there’s hardly any documentation on how to implement the wire protocol. Whenever I got lost and/or confused I just looked at the C++ implementation for guidance. It was there that I found a git submodule with all of Cucumber’s features. Basically, you need to implement all of the “core” features first (therefore ensuring that step definitions actually work), and only then do you get to implement the protocol server itself.

So I wanted to be able to write Cucumber step definitions in D so I could learn and apply BDD to my next project. As it turned out, I learned BDD implementing the wire protocol itself. It took a while to get the hang of transitioning from writing a step definition to unit testing but I think I’m there now. There might be a lot more Cucumber in my future. I might also implement the entirety of Cucumber’s features in D as well, I’m not sure yet.

My implementation is here.

Tagged , , , , , , , , , ,

Adding Java and C++ to the MQTT benchmarks or: How I Learned to Stop Worrying and Love the Garbage Collector

This is a followup to my first post, where I compared different MQTT broker implementations written in D, C, Erlang and Go. Then my colleague who wrote the Erlang version decided to write a Java version too, and I felt compelled to do a C+11 implementation. This was only supposed to simply add the results of those two to the benchmarks but unfortunately had problems with the C++ version, which led to the title of this blog post. More on that later. Suffice it to say for now that the C++ results should be taken with a large lump of salt. Results:

loadtest (throughput - bigger is better)
Connections:         500            750            1k
D + vibe.d:          166.9 +/- 1.5  171.1 +/- 3.3  167.9 +/- 1.3
C (Mosquitto):       122.4 +/- 0.4   95.2 +/- 1.3   74.7 +/- 0.4
Erlang:              124.2 +/- 5.9  117.6 +/- 4.6  117.7 +/- 3.2
Go:                  100.1 +/- 0.1   99.3 +/- 0.2   98.8 +/- 0.3
Java:                105.1 +/- 0.5  105.8 +/- 0.3  105.8 +/- 0.5
C++11 + boost::asio: 109.6 +/- 2.0  107.8 +/- 1.1  108.5 +/- 2.6

pingtest (throughput constrained by latency - bigger is better)
parameters:          400p 20w       200p 200w      100p 400w
D + vibe.d:          50.9 +/- 0.3   38.3 +/- 0.2   20.1 +/- 0.1
C (Mosquitto):       65.4 +/- 4.4   45.2 +/- 0.2   20.0 +/- 0.0
Erlang:              49.1 +/- 0.8   30.9 +/- 0.3   15.6 +/- 0.1
Go:                  45.2 +/- 0.2   27.5 +/- 0.1   16.0 +/- 0.1
Java:                63.9 +/- 0.8   45.7 +/- 0.9   23.9 +/- 0.5
C++11 + boost::asio: 50.8 +/- 0.9   44.2 +/- 0.2   21.5 +/- 0.4

In loadtest the C++ and Java implementations turned out to be in the middle of the pack with comparable performance between the two. Both of them are slightly worse than Erlang and D is still a good distance ahead. In pingtest it gets more interesting: Java mostly matches the previous winner (the C version) and beats it in the last benchmark, so it’s now the clear winner. The C++ version matches both of those in the middle benchmark, does well in the last one but only performs as well as the D version in the first one. A win for Java.

Now about my C++ woes: I brought it on myself a little bit, but the way I approached it was by trying to minimise the amount of work I had to do. After all, writing C++ takes a long while at the best of times so I went and ported it from my D version by translating it by hand. I gleaned a few insights from doing so:

  • Using C++11 made my life a lot easier since it closes the gap with D considerably.  const and immutable became const auto, auto remained the same except when used as a return value, etc.
  • Having also written both C++ and D versions of the serialisation libraries I used as well as the unit-testing ones made things a lot easier, since I used the same concepts and names.
  • I’m glad I took the time to port the unit tests as well. I ended up introducing several bugs in the manual translation.
  • A lot of those bugs were initialisation errors that simply don’t exist in D. Or Java. Or Go. Sigh.
  • I hate headers with a burning passion. Modules should be the top C++17 priority IMHO since there’s zero chance of them making into C++14.
  • I missed slices. A lot. std::vector and std::deque are poor substitutes.
  • Trying to port code written in a garbage collected language and trying to simply introduce std::unique_ptr and std::shared_ptr where appropriate was a massive PITA. I’m not even sure I got it right, more on that below.

The C++ implementation is incomplete and will continue to be like that, since I’m now bored of it, tired, and just want to move on. It’s also buggy. All of the loadtest benchmarks were done with only 1000 messages instead of the values at the top since it crashes if left to run for long enough. I’m not going to debug it because it’s not going to be any fun and nobody is paying me to do it.

It’s not optimised either. I never even bothered to run a profiler. I was going to do it as soon as I fixed all the bugs but I gave up long before that. I know it’s doing excessive copying because copying vectors of bytes around was the easiest way I could get it to compile after copying the D code using slices. It was on my TODO list to remove and replace with iterators, but, as I mentioned, it’s not going to happen.

I reckon a complete version would probably do as well as Java at pingtest but have a hunch that D would probably still win loadtest. This is, of course, pure speculation. So why did I bother to include the C++ results? I thought it would still be interesting and give a rough idea of how it would compare. I wish I had the energy to finish it, but I just wasn’t having fun anymore and I don’t see the point. Writing it from scratch in C++ would have been a better idea, but it definitely would have taken a longer amount of time. It would’ve looked similar to what I have now anyway (I’d still be the author), but I have the feeling it would have fewer bugs. Thinking about memory management from the start is very different from trying to apply smart pointers to an already existing design that depended on a garbage collector.

My conclusion from all of this is that I really don’t want to write C++ again unless I have to. And that for all the misgivings I had about a garbage collector, it saves me time that I would’ve used tracking down memory leaks, double frees and all of those other “fun” activities. And, at least for this exercise, it doesn’t even seem to make a dent in performance. Java was the pingtest winner after all, but its GC is a lot better than D’s. To add insult to C++’s injury, that Java implementation took Patrick a morning to write from scratch, and an afternoon to profile and optimise. It took me days to port an existing working implementation from the closest language there is to C++ and ended up with a crashing binary. It just wasn’t worth the time and effort, but at least now I know that.

Tagged , , , , , , , , , , , , , , , , ,

Go vs D vs Erlang vs C in real life: MQTT broker implementation shootout.

At work we recently started using the MQTT protocol, which uses a publish / subscribe model. It’s simple in the good way and well thought out. We went with an open source implementation named Mosquitto. A few weeks ago on the way back from lunch break my colleague Jeff told me he was writing an MQTT broker in Go, his new favourite language. We’re using MQTT at work, I guess he was looking for a new project to write in Go and voilà. It should be a good fit, after all, this is the type of application that Go was made for. But hubris caught up to him when he uttered “And, of course, it’ll be super fast. It won’t even be fair to other languages”. I’m paraphrasing, but that’s how I remember it. You can read Jeff’s account here.

I’m not a fan of Go at all. I wasn’t particularly impressed when I first read about it but given how much I keep hearing about it on proggit and from Jeff himself, I gave it a go a few months back writing a genetic algorithm framework in it. I came out of that experience liking it even less. It’s just not for me. Go is an opinionated language, which would be fine if I agreed with its creators’ opinions. The way my brain works is that I’m on the opposite side of nearly all of them. It does have a few things I like. The absence of semicolons and parentheses, for instance. Goroutines and channels are a huge win. I can live without exceptions, even though I’d rather not, but generics? They can pry them away from my cold dead hands.

D, on the other hand… now we’re talking. Everything I like about C++ and more, with none of the warts. Of course, it has its own warts too, but nothing’s perfect. So, as a D fan and not so much of a Go one, I took Jeff’s statement as a gauntlet to the face. I learned of vibe.d watching the dconf2013 videos and really liked its idea of a synchronous API on top of asynchronous IO. I was convinced I could at least match a Go implementation’s performance, if not exceed it. So I wrote enough of an MQTT broker implementation to be able to run Jeff’s Go benchmark and compare performances. I reached a version that was faster than his after about 2 days. He came up with a second benchmark and my implementation performed poorly, so I went back to optimising. Around this time another colleague wanted in on the competition and used it as an excuse to learn Erlang, and wrote his own implementation. A few rounds of optimising later, and the results were in, which I’ve included below. Explanations on methodology follow.

loadtest (throughput - bigger is better)
Connections:   100            500            750            1k
D + vibe.d:    121.7 +/- 1.5  166.9 +/- 1.5  171.1 +/- 3.3  167.9 +/- 1.3
C (Mosquitto): 106.1 +/- 0.8  122.4 +/- 0.4   95.2 +/- 1.3   74.7 +/- 0.4
Erlang:        104.1 +/- 2.2  124.2 +/- 5.9  117.6 +/- 4.6  117.7 +/- 3.2
Go:             90.9 +/- 11   100.1 +/- 0.1   99.3 +/- 0.2   98.8 +/- 0.3

pingtest (latency - bigger is better)
parameters:    400p 20w       200p 200w      100p 400w
D + vibe.d:    50.9 +/- 0.3   38.3 +/- 0.2   20.1 +/- 0.1
C (Mosquitto): 65.4 +/- 4.4   45.2 +/- 0.2   20.0 +/- 0.0
Erlang:        49.1 +/- 0.8   30.9 +/- 0.3   15.6 +/- 0.1
Go:            45.2 +/- 0.2   27.5 +/- 0.1   16.0 +/- 0.1

All of the numbers are thousands of messages received by the client application per second. All measurements were done on my laptop, a Lenovo W530 running Arch Linux so all of the TCP connections were on localhost. Each number is the mean of several measurements, and I used the standard deviation as an estimate of the systematic error. All of the MQTT broker implementations run in one system thread. Using multiple threads resulted in no performance benefits for latency and worse performance for throughput.

Mosquitto was compiled with gcc 4.8.2, the Go implementation was executed with go run, the D implementation was compiled with dmd and the Erlang version I’m not sure. I installed the Arch Linux erlang package and used my colleague’s Makefile without looking at it.

The two benchmarks are loadtest and pingtest. The former measures throughput whereas the latter measures latency. In loadtest a few hundred connections are set up to the broker. Half of these subscribe to a topic and the other half publishes to that topic as fast as possible. The benchmark ends when all of the subscribers have received a certain number of messages, determined by a command-line argument. I varied the number of connections to see how that would affect each broker. There was no contest here, the D implementation was by far the fastest. With a 100 connections I think there wasn’t enough work to do so that all implementations ended up waiting on IO. Except for Mosquitto, they all scaled rather nicely. I had problems measuring Jeff’s implementation due to a bug. He knows about the bug but just can’t be bothered fixing it. The numbers were taken from Go 1.1 (the pingtest numbers are Go 1.2). When his implementation works, Go 1.2 produces a binary that performs on the order of 10%-15% faster than the numbers above, which might mean equivalent performance to the Erlang implementation. I even think the bug shows up more often in Go 1.2 exactly because the resulting binary is more performant.

In pingtest Jeff tried to write a better benchmark and it measures latency. The two main command-line arguments are the number of connection pairs and the number of wildcard subscribers. For each pair, one of the connections subscribes to a request topic unique to that pair and the partner connection subscribes to a reply topic. One partner publishes a request and waits for the other connection to publish a reply. The number of messages sent per second now depends on the round-trip time between these two. Additionally, the wildcard subscribers receive both the request and reply messages from the first connection pair. The number before the ‘p’ is the number of connection pairs, and the number before the ‘w’ is the number of wildcard subscriber connections. Here Mosquitto is the fastest, but the performance difference diminishes with more wildcards, being on par with the D implementation in the last column. I’m not sure why it’s the fastest. I think there’s a possibility that vibe.d might be switching to the “wrong” fiber but that’s pure speculation on my part.

What about readability and ease of writing? I can’t read Erlang so I can’t comment on that. Despite my preference for D I think the D and Go implementations are equally readable. Since the Erlang unit tests are in the same files as the implementation, it’s hard to know exactly how many lines long it is. It gets worse since it implements most of MQTT, the D implementation essentially only implements what’s necessary to run the benchmarks. With those caveats (and the fact that dependencies aren’t counted) the 3 implementations clock in at somewhere between 800 and 1000 lines, without filtering out blank lines and comments.

Could they be optimised further? Probably. In the end the choice of algorithm and data structures matter more than the programming language so my personal advice is to choose the language that makes you productive. None of them magically made the implementations performant; we all had to profile, analyse, optimise, try ideas and measure. I loved writing it in D, but then again I’m a convert. I particularly enjoyed using the serialisation library I wrote for it, Cerealed. Much of the typical bit twiddling boilerplate in networking code disappeared, and that was only made possible by D’s compile-time reflection and user-defined attributes.


D: https://github.com/atilaneves/mqtt
C: https://bitbucket.org/oojah/mosquitto/
Go: https://github.com/jeffallen/mqtt
Erlang: https://bitbucket.org/pvalsecc/erlangmqtt
Tagged , , , , , , , , , , ,