Which projects? Which features? What exactly was the impact on performance and reliability, how and why? How did critical projects adopt the stream of features, considering nobody sane touches nothing in a new C++ standard for a decade, waiting for DRs to settle and for codegen stopping to suck?

Reflection has been solved for years with custom codegen, including dimensions which std::meta cannot even touch such as stable cross-platform, cross-compiler ABI.

Speaking seriously, I agree there's definitely a lot of bloat in the new C++ standards. E.g. I'm not a fan of the C++26 linalg stuff. But most performance-focused trading firms still use the latest standard with the latest compiler. Just a small example of new C++ features that are used every day in those firms:

Smart pointers (C++11), Constexpr and consteval (all improvements since C++11), Concepts (C++20), Spans (C++20), Optional (C++17), String views (C++17)

I don't agree at all. For most, linear algebra is the primary reason they pick up C++. Up until now, the best option C++ newbies had was to go through arcane processes to onboard a high performance BLAS implementation which then requires even more arcane steps such as tuning.

With C++26, anyone can simply jump into implementing algorithms.

If anything, BLAS support was conspicuously missing from C++ (and also C).

This blend of comments is more perplexing given that a frequent criticism of C++ is its spartan standard lib, and how the selling point of some commercial software projects such as Matlab is that, unlike C++, linear algebra work is trivial.

Just like the std::random mess, most people are in for a surprise when they attempt to write portable numeric code with it.

I think this criticism is silly. You're complaining about the C++ standard not enforcing something that is virtually impossible and no implementation under the sun even suggests it would conceivably support. And we should just assume C++ would be able to force it across implementations, target platforms, and even architectures? Ridiculous.

> Just like the std::random mess (...)

The comments I've seen regarding std::random mainly refer to gotchas, such as std::rand returning values in [0, RAND_MAX] and the latter being a platform-specific constant.

There is a reason after all why you only throw vague complains of "mess" instead of pointing out specific concerns or grievances. You need to complain, regardless of merit.

Overall, this blend of criticism is no different than the cliche criticism targeting the STL. Sure, some niche applications can and do have better ways to scratch their niche itches. In the meantime the STL perfectly serves the need of 99.9% of all common usages. Is this not the point? Doesn't linalg and rand achieve this?

Of course vapid nitpickers can still pull out the last resort card of complaining that the implementation is too complicated, a grievance also directed at the STL. But that's just the extent where this silliness goes.

I would have voted SA on this matter, if I had a life that would allow me to go around voting at WG21 meetings.

We have vcpkg and conan now, the standard library cannot be a distribution vehicle for organisations that refuse to adopt C++ package managers.

The frequency doesn't make the criticism more valid and those repeating it would be better served to let go of their fear of third-party libraries.

The criticism is not valid. It's specious reasoning at best, fueled by a hefty dose of gatekeeing.

The only rationale that is relevant is whether the standard library provides a way to do a very basic and fundamental task instead of having to force anyone to onboard and manage third party dependencies. That's the whole point of a standard library, isn't it?

"Onboarding" a third party library isn't this herculean task that you make it out to be but is in fact a very basic part of software development that almost any project will have to deal with anyway unless you are into reinventing the wheel - even excessively bloated standard libraries don't manage to cover everything.

Out of hundreds of hundreds of projects I've interacted with, maybe less than 1% have used linear algebra in any non-basic capacity (e.g. more than multiplying two 4x4 matrices) and had to use Eigen or BLAS

Are you really trying to argue that if you ignore the bulk of applications that have linear algebra then only a few have linear algebra?

> and had to use Eigen or BLAS

What if I told you that if C++ provided basic support for linalg, your average C++ developer wouldn't ever have to hear about Eigen or blas?

It's perfectly fine if you want to use the likes of Eigen. It's also perfectly fine if any developer opts to ditch the STL and use any performance-oriented container.

But is it ok to force everyone to onboard a third party library just to be able to do very basic things like use a stack or do a dense matrix-vector multiplication? I don't think that leads to a good developer experience.

Anecdata: A year or so ago I have been in discussion if beta features of C++20 on platforms are good to be used on large scale. It makes it not a sum but an intersection of partial implementations. Anyway it looked positive until we needed a pilot project to try. One of the projects came back with 'just flipping C++20 switch with no changes causes significant regression on build times'. After confirming it that it is indeed not an error on our side it was kinda obvious. Proportional increase of remote compilation cloud costs for few minor features is a 'no'. After a year the beta support is no longer beta but still partial on platforms and no improvements on build times in community. YMMV of course because gamedev mostly supports closed source platforms with closed set of build tools.

Given that C++20 introduced modules, which are intended to make builds faster, I think just flipping C++20 switch with no changes and checking build times should not be the end of checking whether C++20 is worth it for your setup.

Turning on modules effectively requires that all of your project dependencies themselves have turned on modules. Fail to do so, and a lot of the benefits start to become hindrances (Clang is currently debating going to 64-bit source locations because modularizing in this manner tends to exhaust the current 32-bit source locations).

I think this just proves that your team is highly inexperienced in C++ projects, which you implicitly attest by admitting this was your first C++ upgrade you had to go through.

Let me be very clear: there is never an upgrade of the C++ version targeted by a project that does not require full regression tests and a few bugs to squash. Why? Because even if the C++ side of things is perfectly fine, libraries often introduce all sorts of unexpected issues.

For example, once I had to migrate a legacy project to C++14 and flipping the compiler flag to c++14 caused a wall of compiler errors. It turned out the C++ was perfectly fine, but a single library behaved very poorly with a constexpr constructor they enabled conditionally with C++14.

You should understand that upgrades to the core language and standard libraries are exceptionally stable, and a clear focus of the standardization committee. But they only have a say in how the core language and standard libs should be. The bulk of the code any relatively complex project consumes is not core lang+ stdlib, but third-party libraries and frameworks. These often are riddled with flags to toggle whole components only in specific versions of the C++ language, mainly for backwards compatibility. Once you target a new version of C++, often that means you replace whole components of upstream dependencies. This often requires fixing your code. This happens very frequently, even with the likes of Boost.

So, what you're complaining about is not C++ but your inexperience in software engineering in general. I mean, what is the rule of thumb about major version upgrades?

How high are those compilation costs compared the developer time that might be saved with even minor features?

Good, but I think what happens is there are people on the bleeding edge of C++, usually writing libraries that ship with new code. Each new feature is a godsend for them -- it's the reason why the features are proposed in the first place. It allows you to write libraries more simply, more generally, more safely, and more efficiently.

The rest of us are dealing with old code that is a hodgepodge of older standards and toolchains, that has to run in multiple environments, mostly old ones. It's like yeah, this C++26 feature will come in handy for me someday, but if that day comes then it will be in 2036, and I might not be writing C++ by then.

Things seem to be catching up. I had the same view up until recently, but now I'm able to use most of the C++23 features in an embedded platform (granted, some are still missing (limited to GCC 11.2).