Re: C23 thoughts and opinions

On 5/24/2024 7:50 AM, David Brown wrote:

On 24/05/2024 01:05, Chris M. Thomasson wrote:

On 5/23/2024 6:35 AM, David Brown wrote:

On 22/05/2024 23:24, Chris M. Thomasson wrote:

On 5/22/2024 9:55 AM, David Brown wrote:

In an attempt to bring some topicality to the group, has anyone started using, or considering, C23 ?  There's quite a lot of change in it, especially compared to the minor changes in C17.

>
Love the way std::vectors respect alignas... C++20, iirc?
>
[...]

>
I have no idea what you are talking about.

>
std::vector actually respects alignas, on MSVC at least. I did not know this worked until I tried it. Iirc, Bonita was the one that sparked my test. It aligned itself on the proper boundaries. Very nice.
>

But did you notice that this is c.l.c, not c.l.c++, and the topic is C23, not C++23 ?  Discussing comparisons or compatibility with C++ is fair enough, but talking about pure C++ matters (such as std::vector<>) is unlikely to be helpful.

>
C has it as well... Very useful!

 I know C has alignas (now as a keyword in C23, instead of just _Alignas from C11).
 I know C++ has alignas (from C++11 onwards).
 What I don't understand is why you think std::vector<> "respects alignas" in C++20 - alignment for std::vector<> works like alignment for any other class in C++, and always has done.
 And what I /really/ don't understand is why you think it is remotely relevant here?  Even "alignas" in C is not particular relevant to this thread, except that it has become a keyword in C23 instead of a macro defined to _Alignas in <stdalign.h>.

alignas is very nice because it can help me make a 100% portable version of some of my old exotic lock-free memory allocators that use rounding to get down to a header. Any point in the region can be rounded down to get at the header for the block. It involves aligning the main region on a large boundary, say 8192 bytes. This is a little trick for high performance lock-free allocators.
Iirc, I can make std::vector align its elements to say, L2 cachelines, and I can make std::vector align itself on a large boundary say 8192 bytes. All in std C++! That is nice.
Any point in the block can round itself down to the nearest 8192 block, then we have direct access to its header. Also, then I make a rule that each memory block in a region is at least the size of a pointer. Then I can use them in a lock-free stack, or something. This is damn near zero overhead.
I have posted about some of these allocators in the past.

Perhaps I should just be grateful for the small mercy of there being no random youtube link in your post.

No shit. ;^)