Automated instance construction in C++

I’m currently mostly switching back and forth between C# and C++ projects. One of the things that I’m missing most when switching to C++ is a nice dependency-injection (DI) library. After checking out what was already available, I finally decided I wanted to try to build my own slim type-indexed variant. I quickly started by registering factories and instances in a map on std::type_index, making it possible to both have the DI retain ownership (with std::unique_ptr) or just make a type available via a bare pointer. So I was able to do things like:

// Register an instance
di.insert_unique(std::make_unique<foo_service>());
// Register a factory
di.insert_unique([] {return std::make_unique<bar_service>());
// Register an existing bare pointer
di.insert_bare(my_bare_thingy);

// ... and retrieve them
auto& foo = di.get<foo_service>();

One of the most powerful aspects of a DI library is the ability to transitively setup dependencies. I like constructor injection the most, so I implemented a very naive way like this:

di.insert_unique([](auto& p) { return std::make_unique<complex_service>(
  p.get<base_service1>(), p.get<base_service2>(), p.get<base_service3>());
});

This is pretty verbose and we basically have to repeat all the constructor parameter types. But it’s easy to implement. We can do a little bit better by using a templated type-conversion operator and using it to call the get:

class service_provider
{
  struct inferred_locator
  {
    service_provider const* provider;
    template <class T> operator T&() const
    {
      return provider->get<std::remove_const_t<T>>();
    }
  };
  
  inferred_locator get() const
  {
    return { .provider = this };
  }
  
  /** typed get implementations here... */
};

Now we can reduce the previous registration to:

di.insert_unique([](auto& p) { 
  return std::make_unique<complex_service>(p.get(), p.get(), p.get());
});

That is basically only the number of constructor parameters in a verbose way. We could write a small template that takes the number, creates an std::index_sequence from it and then unpacks each index into an invokation of service_provider::get. But then we would still have to update registrations when adding (or removing) a new dependency to a services’s constructor. With a litte more work, we can actually get this instead:

di.insert_unique<complex_service>();

This partly inspired by Antony Polukhin’s C++ reflection talks, and combines std::index_sequence based unpacking, SFINEA and the templated type-conversion operator:

template <class T, std::size_t Head, std::size_t... Rest>
constexpr auto make_unique_impl(provider_wrapper const& p,
    std::index_sequence<Head, Rest...>,
    decltype(T{ mimic{ Head }, mimic{ Rest }... }) * = nullptr) -> std::unique_ptr<T>
{
    // This next requirement is so we do not accidentally recurse into the copy/move-ctors
    static_assert(sizeof...(Rest) + 1 > 1, "Can only deduce constructors with two or more parameters.");
    return std::make_unique<T>(p(Head), p(Rest)...);
}

template <class T, std::size_t... Rest>
constexpr auto make_unique_impl(provider_wrapper const& p, std::index_sequence<Rest...>) -> std::unique_ptr<T>
{
    // This next requirement is so we do not accidentally recurse into the copy/move-ctors
    static_assert(sizeof...(Rest) > 1, "Can only deduce constructors with two or more parameters.");
    return make_unique_impl<T>(p, std::make_index_sequence<sizeof...(Rest) - 1>{});
}

template <class T, std::size_t Max = 8> auto make_unique(service_provider const& p)
{
    return make_unique_impl<T>(provider_wrapper{ &p }, std::make_index_sequence<Max>{});
}

This uses two new types: mimic, which is only used for SFINEA, takes std::size_t on construction (for the unpacking from the std::index_sequence) and converts to anything (templated type conversion again) and the provider_wrapper, which is a simple adaptor around service_provider that takes an unused std::size_t argument (again, for unpacking). The first overload of make_unique_impl is slightly more specialized (because it has Head and Rest), so the compiler tries it first. If it works, it returns a new instance of the service we want. Otherwise, it will fail without an error due to SFINEA in the unused and defaulted third parameter. The compiler will then try the second overload, which will recurse to a variant with fewer parameters. The outermost make_unique starts this recursion with 8 parameters, because that should be enough for any sane service. I stop this recursion at one constructor parameter, even though that is a useful configuration. This is because I have not yet found a way to avoid calling the copy or move constructors accidentally. If anyone knows how to do that, I’d be very happy to hear how. My workaround right now is to explicitly register a factory in that case.

Reading a conanfile.txt from a conanfile.py

I am currently working on a project that embeds another library into its own source tree via git submodules. This is currently convenient because the library’s development is very much tied to the host project and having them both in the same CMake project cuts down dramatically on iteration times. Yet, that library already has its own conan dependencies in a conanfile.txt. Because I did not want to duplicate the dependency information from the library, I decided to pull those into my host projects requirements programmatically using a conanfile.py.

Luckily, you can use conan’s own tools for that:

from conans.client.loader import ConanFileTextLoader

def load_library_conan(recipe_folder):
    text = Path(os.path.join(recipe_folder, "libary_folder", "conanfile.txt")).read_text()
    return ConanFileTextLoader(text)

You can then use that in your stage methods, e.g.:

    def config_options(self):
        for line in load_library_conan(self.recipe_folder).options.splitlines():
            (key, value) = line.split("=", 2)
            (library, option) = key.split(":", 2)
            setattr(self.options[library], option, value)

    def requirements(self):
        for x in load_library_conan(self.recipe_folder).requirements:
            self.requires(x)

I realize this is a niche application, but it helped me very much. It would be cool if conan could delegate into subfolders natively, but I did not find a better way to do this.

Metal in C++ with SDL2

Metal, Cupertino’s own graphics API, is sort of a middle-ground in complexity between OpenGL and Vulkan. I’ve wanted to try it for a while, but the somewhat tight integration into Apple’s ecosystem (ObjectiveC/Swift and XCode) has so far prevented that. My graphics projects are usually using C++ and CMake, so I wanted a solution that worked with that. Apple released Metal-cpp last year and newer SDL2 versions (since 2.0.14) can create a window that supports drawing to it with metal. Here’s how to weld that together (with minimal ObjectiveC).

metal-cpp

I get the metal-cpp code from the linked website (the download is at step 1). I add a library in CMake that builds a single source file that compiles the metal-cpp implementation with the ??_PRIVATE_IMPLEMENTATION macros as described on the page (see step 3). That target also exports the includes to be used later.

SDL window and view

Next, I use conan to install SDL2. After SDL_Init, I call SDL_CreateWindow to create my window. I do not specify SDL_WINDOW_OPENGL (or in the SDL_CreateWindow‘s flags, or next step will fail. After that, I use SDL_Metal_CreateView from SDL_metal.h to create a metal view. This is where things get a little bit icky. I create a metal device using MTL::CreateSystemDefaultDevice(); but I still need to assign it to the view I just created. I’m doing that in ObjectiveC++. In a new .mm file I add a small function to do that:

void assign_device(void* layer, MTL::Device* device)
{
  CAMetalLayer* metalLayer = (CAMetalLayer*) layer;
  metalLayer.device = (__bridge id<MTLDevice>)(device);
}

I use a small .h file to expose this function to my C++ code like any other free function. There’s another helper I create in the .mm file:

CA::MetalDrawable* next_drawable(void* layer)
{
  CAMetalLayer* metalLayer = (CAMetalLayer*) layer;
  id <CAMetalDrawable> metalDrawable = [metalLayer nextDrawable];
  CA::MetalDrawable* pMetalCppDrawable = ( __bridge CA::MetalDrawable*) metalDrawable;
  return pMetalCppDrawable;
}

At the beginning of each frame, I use that together with SDL_Metal_GetLayer to get a texture to render to:

auto surface = next_drawable(SDL_Metal_GetLayer(view));

Next I create a render pass descriptor that starts by clearing that drawable with our fancy red:

MTL::ClearColor clear_color(152.0/255.0, 23.0/255.0, 42.0/255.0, 1.0);
auto pass_descriptor = MTL::RenderPassDescriptor::alloc()->init();
auto attachment = pass_descriptor->colorAttachments()->object(0);
attachment->setClearColor(clear_color);
attachment->setLoadAction(MTL::LoadActionClear);
attachment->setTexture(surface->texture());

And fire that off to the GPU using a command buffer and render encoder:

auto buffer = queue->commandBuffer();
auto encoder = buffer->renderCommandEncoder(pass_descriptor);
encoder->endEncoding();
buffer->presentDrawable(surface);
buffer->commit();

There you have it, a minimal running metal application. Still a long ways from the traditional “Hello Triangle”, but most metal examples that show how to do that can easily be translated to the C++ API. Note that you probably have to take some extra steps to compile metal shaders (aka MSL). You can either load them from source or precompile them using the command line tools.

My favorite C++20 feature

As I evolved my programming style away from mutating long-lived “big” objects and structures and towards are more functional and data-oriented style based mainly on pure functions, I also find myself needing a lot more structs. These naturally occur as return types for functions with ‘richer’ output if you do not want to use std::tuple or other ad-hoc types everywhere. If you see a program as a sequence of data-transformations, I guess the structs are the immediate representations encoded in the type system.

Let my first clarify what I mean by structs, as opposed to what the language says: A type that has all public data members, obeys the rule of zero, and is valid in any configuration. A typical struct v3 { float x{},y{},z{};}; 3d vector is a struct, std::vector is not.

These types are great. You can copy them around, use them with structured binding, they correctly propagate constness, and they are a great fit to pass them through layers of functions calls. And, when used as function parameters, they are great for evolving your program over time, because you can just change the single struct, as opposed to every function call that uses this parameter combination. Or you can easily batch, or otherwise ‘delay’, calls by recording the function parameters. Just throw the parameters into a container and execute the code later.

And with C++20, they got even better, because now you can use them with my favorite new feature: designated initializers, which allows you to use the member names at the initialization site and use RAII. E.g., for a struct that symbolizes an http request: struct http_request { http_method method; std::string url; std::vector<header_entry> headers; }; You can now initialize it like this:

auto request = http_request{
  .method = http_method::get,
  .uri = "localhost:7634",
  .headers = { { .name = "Authorization", .value = "Bearer TOKEN" } },
};

You can even use this directly as a parameter without repeating the type name, de facto giving your named parameters for a pair of extra curlys:

run_request({
    .method = http_method::get,
    .uri = "localhost:7634",
    .headers = { { .name = "Authorization", .value = "Bearer TOKEN" } },
});

You can, of course, combine this named-parameter style-struct with other function parameters in your API, but like with lambdas, I think they are most readable as the last parameter. Hence, also like with lambdas, you probably never want to have more than one at each call-site. I’m very happy with this new feature and it’s already making the code using my APIs a lot more readable.

The boy scout rule and git in practice

There’s a dichotomy when applying the boy scout rule to programming: cleaning up code that you happen to come across ‘pollutes’ your merge-/pull-requests, making it harder to review and therefor more unlikely to be accepted.

One way to cope with this is to submit the ‘clean up’ and the feature/task related changes separately, and merge them back into upstream in separate steps. But often times, it is much easier to just fix a small problem right away instead of switching back to your main branch and doing it there. In fact, it might prevent the developer from doing the improvement, which I want to avoid. Quite the opposite, I want to encourage my fellow developers to do improvements.

So one thing that we do about this is to mark the changes that are unrelated (or tangentially related) to the task with their own commit and a special prefix in the commit message like:

BSR: More consistent function signatures

As you might have guessed, BSR stands for boy scout rule. This does not solve the fact that the diffs get larger than necessary, but it makes it possible to ‘filter out’ the pure refactorings. In some cases, these commits can later be cherry-picked onto the main branch before doing the review. Of course, this only works for small refactorings, but this is where the boy scout rule applies.

Composition of C# iterator methods

Iterator methods in C# or one of my favorite features of that language. I do not use it all that often, but it is nice to know it is there. If you are not sure what they are, here’s a little example:

public IEnumerable<int> Iota(int from, int count)
{
  for (int offset = 0; offset < count; ++offset)
    yield return from + offset;
}

They allow you to lazily generate any sequence directly in code. For example, I like to use them when generating a list of errors on a complex input (think compiler errors). The presence of the yield contextual keyword transforms the function body into a state machine, allowing you to pause it until you need the next value.

However, this makes it a little more difficult to compose such iterator methods, and in reverse, refactor a complex iterator method into several smaller ones. It was not obvious to me right away how to do it at all in a ‘this always works’ manner, so I am sharing how I do it here. Consider this slightly more complex iterator method:

public IEnumerable<int> IotaAndBack(int from, int count)
{
  for (int offset = 0; offset < count; ++offset)
    yield return from + offset;

  for (int offset = 0; offset < count; ++offset)
    yield return from + count - offset - 1;
}

Now we want to extract both loops into their own functions. My one-size-fits-all solution is this:

public IEnumerable<int> AndBack(int from, int count)
{
  for (int offset = 0; offset < count; ++offset)
    yield return from + count - offset - 1;
}

public IEnumerable<int> IotaAndBack(int from, int count)
{
  foreach (var x in Iota(from, count))
     yield return x;

  foreach (var x in AndBack(from, count))
     yield return x;
}

As you can see, a little ‘foreach harness’ is needed to compose the parts into the outer function. Of course, in a simple case like this, the LINQ version Iota(from, count).Concat(AndBack(from, count)) also works. But that only works when the outer function is sufficiently simple.

WPF Redux Sample Application

A while ago, I wrote about how we are using the redux architexture in our C# applications. I have just pushed an example showing ReduxSimple with WPF and our extensions in a .NET 5 application to our github account. The example itself is just a counter with an increment and a decrement button, but it already shows the whole redux cycle.

The store setup in App.xaml.cs shows how the ReducerBuilder can be used to build a State reducer from the Reducer class via reflection.

I also added a small prime-number factorization to show how to use ‘expensive’ functions in the view part of the application using our SelectorGraph. This makes it possible to properly derive view data from the state, only updating them once when one of their inputs changes. In the example, that is the counter. So the number will only be factorized when the counter changes, while all other future state changes do update the selector.

The example does not use the UIDuplexBinder yet. It allows read/write binding of WPF controls to an IObservable and an action-creator, and is hopefully pretty straight-forward to use. Please enjoy!

Chopping up big tasks

As a programmer, you have probably dealt with a task that seemed simple enough in the beginning but just keeps going and going and going. I have been chewing on such a task for the better part of the last three weeks and finally closed it today*. A really long time, considering I usually complete my tasks in less than a day, up to three days for especially long ones.

I know that many programmers can get lost with big tasks like this. I am certainly no exception. Analysis paralysis and decision fatigue can easily get the best of you, considering the mountain of work still ahead.

But I have a few ways to deal with such situations. Of course, your mileage may vary. But I am sure that without them, this specific issue would have taken me even longer. It boils down to one rule:

Focus on the essentials only.

This obviously relates to yak shaving. Sometimes you need to do something else first before you can complete your task. This is recursive and can quickly take up lots of time. This will ultimately be required, but for the moment it distracts from the original task. While you complete a side task, the main task will not advance, leading to a feeling of getting stuck, technical problems (like merge conflicts in long-running branches) and psychological problems (like decision fatigue).

So what can you do about this? My advice is to rigorously cut-off side tasks, by taking up technical debt temporarily. I annotate my code with HACK, TODO and FIXME to mark all the isolated spots I still need to change for the 100% version. The end feature (= user story) does not have to be completed by the end of my task, but I should be reasonably confident that the main work is done. Anything to that end will work.

Some required changes immediately appear to be too extensive for such a small annotation. In that case, I will usually create a new follow-up issue in our issue-tracker and mark the code with a link to it.

After completing the main work in this way, but before I merge my code or close my original work-item/issue, I make another pass over all the HACK, TODO and FIXMEs I generated. The smaller ones I fix right away. Anything where the way to complete them is not super obvious gets converted into an issue in the issue tracker, and cross-linked from the code. This means I add a comment referencing the issue from the code and I make sure that the issue says that it is marked in the code. E.g., for this specific task, I now have 6 open follow-up issues.

After that, I usually merge the code into the main branch. If it’d break something or be misleading with all the follow up issues not done yet, the feature can sometimes be disabled with a feature toggle. Alternatively, the follow up tasks can be completed in their own branches and merged back onto the main task’s branch, which can be merged once everything is done. This hugely depends on your product cycle, of course.

Do you have any clever methods to handle bigger tasks?

Using a C++ service from C# with delegates and PInvoke

Imagine you want to use a C++ service from contained in a .dll file from a C# host application. I was using a C++ service performing some hardware orchestration from a C# WPF application for the UI. This service pushes back events to the UI in undetermined intervals. Let’s write a small C++ service like that real quick:

#include <thread>
#include <string>

using StringAction = void(__stdcall*)(char const*);

void Report(StringAction onMessage)
{
  for (int i = 0; i < 10; ++i)
  {
    onMessage(std::to_string(i).c_str());
    std::this_thread::sleep_for(std::chrono::seconds(1));
  }
}

static std::thread thread;

extern "C"
{
  __declspec(dllexport) void __stdcall Start(StringAction onMessage)
  {
    thread = std::thread([onMessage] {Report(onMessage);});
  }

  __declspec(dllexport) void __stdcall Join()
  {
    thread.join();
  }
}

Compile & link this as a .dll that we’ll call Library.dll for now. Catchy, no?

Now we write a small helper class in C# to access our nice service:

class LibraryLoader
{
  public delegate void StringAction(string message);

  [DllImport("Library.dll", CallingConvention = CallingConvention.StdCall)]
  private static extern void Start(StringAction onMessage);

  [DllImport("Library.dll", CallingConvention = CallingConvention.StdCall)]
  public static extern void Join();

  public static void StartWithAction(Action<string> action)
  {
    Start(x => action(x));
  }
}

Now we can use our service from C#:

LibraryLoader.StartWithAction(x => Console.WriteLine(x));
// Do other things while we wait for the service to do its thing...
LibraryLoader.Join();

If this does not work for you, make sure the C# application can find the C++ Library.dll, as VS does not help you with this. The easiest way to do this, is to copy the dll into the same folder as the C# application files. When you’re starting from VS 2019, that is likely something like bin\Debug\net5.0. You could also adapt the PATH environment variable to include the target directory of your Library.dll.

If you’re getting a BadImageFormatException, make sure the C# application is compiled for the same Platform target as the C++ application. By default, VS builds C++ for “x86”, while it builds C# projects for “Any CPU”. You can change this to x86 in the project settings under Build/Platform target.

Now if this is all you’re doing, the application will probably work fine and report its mysterious number sequence flawlessly. But if you do other things, e.g. something that triggers garbage collection, like this:

LibraryLoader.StartWithAction(x => Console.WriteLine(x));
Thread.Sleep(2000);
GC.Collect();
LibraryLoader.Join();

The application will crash with a very ominous ExecutionEngineException after 2 seconds. In a more realistic environment, e.g. my WPF application, this happened seemingly at random.

Now why is this? The Action<string> we registered to print to the console gets garbage collected, because there is nothing in the managed environment keeping it alive. It exists only as a dependency to the function pointer in C++ land. When C++ wants to message something, it calls into nirvana. Not good. So let’s just store it, to keep it alive:

static StringAction messageDelegate;
public static void StartWithAction(Action<string> action)
{
  messageDelegate = x => action(x);
  Start(messageDelegate);
}

Now the delegate is kept alive in the static variable, thereby matching the lifetime of the C++ equivalent, and the crash is gone. And there you have it, long-lasting callbacks from C++ to C#.

A very strange bug

A week ago, one of our junior programmers encountered a strange bug in his WPF application. This particular application has a main window with pages, i.e. views, that can be switched between, e.g. via the main menu. The first page, however, is the login page. And while on the login page, the main menu should be disabled, so users cannot go where they are not authorized to go.

And this worked fine. A simple boolean in the main window’s view-model was used to disable the menu when in on login page, and enable it otherwise. We have a couple of applications that behave this way, and there were enough examples to get this to work.

Now the programmer introduced a new feature: when the application is started for the first time, there should be a configuration page right after the login page. During the configuration, the main menu should still be disabled. When the user hits the save button on the configuration page, the configuration should be stored and they should get to the dashboard with an enabled main menu.

New Feature, new Bug

Of course, this required changing the condition for when the main menu is disabled: When on either of the two pages, keep it disabled. But now the very strange bug appeared. When going to the dashboard from the configuration page, the main menu was correctly enabled, but all of its menu entries were still disabled. And this only happened when opening the main menu for the first time. When closing and opening it again, all menu entries were correctly enabled.

Now a lot of hands-on debugging ensued. The junior developer used all of the tools at his disposal: web searching, debug output, consulting other senior developers. The leads were plenty, too. Could it be a broken INotifyPropertyChanged implementation? Was ICommand.CanExecute not returning the correct value? Can we attach our own CanExecute handlers to the associated CommandBindings to at least get around the issue? Do we manually have to trigger a refresh of the enabled state?

Nothing worked, and no new information was gained. Even after fiddling around with the problem for a few days, there was no solution, no new insight to be found, not even a workaround. All our code seemed to be working alright.

From good to bad

One of my debugging mantras, that always helped me with the nastiest of bugs, is:

Work from a good, bug-free scenario to the bad, buggy scenario. Use small increments and bisection to find the step that breaks it.

In this situation, we were lucky. We had a good, working scenario in the same application. Starting the application without the “first time configuration” was working nicely. So what was the difference? From the login page the user also hit a button to change to the dashboard page.

The only difference was: the configuration was not stored in between. So we commented that out. Finally! Progress! We could not believe it. Commenting out the “store the configuration” code made our menu items work. Time to dig deeper: The store-the-configuration code was using a helper dialog called TaskDialog that awaits a given Task while showing an “in progress” animation. Our industrious junior developer thought that might be a good idea for storing the configuration data using File.WriteAllTextAsync. Further bisection revealed that it was not actually the “save” Task that was causing the problem, but our TaskDialog: Removing the await from the TaskDialog, our MainWindow‘s main menu was still broken.

This was surprising since the TaskDialog had been in-production, seemingly working alright for quite some time. Yet all our clues hinted at it being the culprit. In its implementation, it runs the given Task directly in its async “Loaded” event handler. Once it is done, it sets the DialogResult to true.

So we hypothesized that it is probably not a good idea to close the dialog while it is currently in the process of opening. The configuration saving task was probably very fast and never yielding, so only that was showing the strange behavior, while all our previous use cases were “slow enough” and yielded at least once.

Hence we tried a small modification: We delayed the execution of our Task and the subsequent DialogResult = true; slightly to the next “event frame” using Application.Current.Dispatcher.InvokeAsync. And that did the trick! The main menu items were finally correctly enabled after leaving the configuration page.

And this is how we solved this very weird bug, where the trigger does not appear to relate to the symptom at all. There is probably still a bug causing this weird behavior somewhere in WPF, but at least we are not longer triggering it with our TaskDialog. Remember, start from the good case, iterate and bisect!