Hacker News Reader

Comment by jjmarr | Hacker News Reader

It's very useful in C++, funnily enough. This is because I can have a non-templated interface base class, then a templated impl class.

Then my templated impl header can be very heavy without killing my build times since only the interface base class is #included.

Not sure if this is as common in Java.

View on HN

Using more complex architecture (which requires more human time to understand) to merely make build time shorter is a ridiculous choice.

For a large project this could save hours of developer time.

C++ is a hell of a language.

just buy your devs faster computers to compile on

You can't buy your way out of this, because C++ builds are only parallelizable across multiple translation units[1] (i.e. separate .cpp files). Unless you're willing to build a better single-core CPU, there's not much you can do.

The challenge with modern C++ projects is that every individual TU can take forever to build because it involves parsing massive header files. Oftentimes you can make this faster with "unity builds" that combine multiple C++ files into a single TU since the individual .cpp file's build time is negligible compared to your chonky headers.

The reason the header files are so massive is because using a templated entity (function or class) requires seeing the ENTIRE DEFINITION at the point of use, because otherwise the compiler doesn't know if the substitution will be successful. You can't forward declare a templated entity like you would with normal code.[2]

If you want to avoid including these definitions, you create an abstract interface and inherit from that in your templated implementations, then pass the abstract interface around.

[1] or linking with mold

[2] There used to be a feature that allowed forward declaring templated entities called "export". A single compiler tried to implement it and it was such a failure it was removed from the language. https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2003/n14...

Java uses type erasure which are very cheap in compile time but you cannot do things like

   t = new T(); // T is a template parameter class

C++ uses reified generics which are heavy on compile time but allows the above.

That's why they're called generic parameters, not template parameters; the code is generic over all possible parameters, not templated for every possible parameter.

    > C++ uses reified generics

I was a C++ programmer for many years, but I never heard this claim. I asked Google AI and it disagees.

    > does c++ have reified generics?

    > C++ templates do not provide reified generics in the same sense as languages like C# or Java (to a limited extent). Reified generics mean that the type information of generic parameters is available and accessible at runtime.

Interesting I’d never picked up on this pedantic subtlety. I too thought reified meant what you could do at the call site not what you could do at runtime. Was my understanding wrong, or is Gemini hallucinating.

In any event, you have to use weird (I think “unsafe”) reflection tricks to get the type info back at runtime in Java. To the point where it makes you think it’s not supported by the language design but rather a clever accident that someone figured out how to abuse.

Reply on news.ycombinator.com