libCat VS EA Standard Template Library

That's awesome! I'm working on something that sounds similar. https://github.com/cons-cat/libcat

I'd love to see your work if you're willing to share it here!

This runtime size bothers me a lot. So much that I've been working on a new runtime for C++ that breaks POSIX compatibility to keep binaries as small as they can be. The hello world with LTO is 330ish bytes right now, and I think that can get smaller. https://github.com/Cons-Cat/libCat

I'm working on a library that replaces both C++ and C/POSIX standard libraries (https://github.com/Cons-Cat/libCat), but even then I need to define a few std:: namespace symbols for some features. In the case of std::initializer_list, my answer is just don't use that feature, because you don't really need it.

It's worse in the standard library than it has to be. When I refactored my traits to minimize template instantiations and lean on concepts as much as possible, I measured over 30% improvement to clean build compile times. It's not possible for the standard to do this, because it would subtly change the API. For instance, you can't instantiate or take the address of a concept, but you can for a type-trait class. No reason you'd want to do that, but you can, so they can't "break" the standard library by optimizing this.

This doesn't have to be true. Over the past year I've made progress towards demonstrating how even non-freestanding C++ can be written without any C or C++ standard library headers or DLLs (with large benefits). There are a few names which the compilers require to be in the std:: namespace, though, but they're very special features like source_location and construct_at with semantics that can't be expressed otherwise.

This bit cast has no overhead in debug mode, and is a little bit more generally useful than std::bit_cast(), but cannot be constant evaluated. https://github.com/Cons-Cat/libCat/blob/main/src/libraries/utility/implementations/bit_cast.tpp

It isn't that hard to put a predicate into a type. We have lambdas in an unevaluated context, CTAD, and templated type aliases. https://github.com/Cons-Cat/libCat/blob/main/src/libraries/scaredy/cat/scaredy https://github.com/Cons-Cat/libCat/blob/main/src/global_includes.hpp#L70 https://github.com/Cons-Cat/libCat/blob/main/src/libraries/linux/cat/linux#L289 You do it like this.

But suppose that you have code with no standard library calls at all. Would it still make sense to choose this naming convention? This is actually possible, with a few special exceptions. GCC requires that an implementation of std::source_location has very particular class member names, GCC assigns special semantics to a few function names including std::construct_at and std::move (people seem to know it's inlined, but did you know std::move is required for move-related warnings?), and most intrusively of all, a promise_type must be snake_case. Other names can be worked around by using them into a different namespace with a different letter-case, but promise_type seems unavoidable.

I use C++ for a low-level Linux runtime. Other people are using it for operating systems like SerenityOS and Zircon/Fuschia. People also use C++ for making more compilers like GCC and LLVM.

> you can still use it with smart pointers provided by any other library

Is the point of having a kitchen-sink library like this not that you dont have to reach for a 3rdparty library for things that you need 'all the time'?

Certainly, not everyone needs it.

...but, not everyone needs threads either. Not everyone needs an http server; and yet, if you have an application framework that provides them, when you do need them, it saves you reaching for yet-another-dependency.

Was that no the point from the beginning?

unique_ptr is a fundamental primitive for many, as you see from some other frameworks (1), and implementation is not always either a) trivial, or b) as simple as 'just use std::unique_ptr'.

This does seem like a very opinionated decision with reasonably unclear justification.

[1] - eg. https://github.com/EpicGames/UnrealEngine/blob/release/Engin..., https://github.com/electronicarts/EASTL/blob/master/include/...

Can I introduce you to EASTL

For games focused stuff have a look at EASTL https://github.com/electronicarts/EASTL also perhaps some of the Data Oriented Design stuff (see Mike Acton's CPP Con Talks). This also have loads of good stuff https://www.dataorienteddesign.com/dodbook/

C++, but generally treat it as C. STL is pretty slow while debugging so we avoid it and write our own replacements. If you don't want to drive that deep use something like EASTL: https://github.com/electronicarts/EASTL

In my experience game devs don't eschew the STL because of compile times, but because (a) its memory management is not sufficiently tunable and (b) it uses exceptions heavily which many game engines disable entirely due to perceived performance problems. For this reason, some of the engines I've worked on have used their own forks of the STL in the spirit of EASTL to rectify these issues. Others like my current project just don't use the STL at all (outside of some third-party library code) and use custom libraries for everything.

In my opinion, you should also benchmark it against something like boost::container::flat_set or eastl::vector_set and you should expect the same performance as with your ordered functionality. Another interesting idea for organization of such flat and sorted container can be found here.

finally, if you're worried about performance (a lot of aspiring game programmers seem to be, even though for anything but high-end engines the STL is likely fine if you know what you're doing..) you could look into the EA STL or other STL replacements..

check it out: https://github.com/electronicarts/EASTL

I just took a look at their atomic.h and wow, "well-documented" is an understatement !