Any plan for libc++ to support std::from_chars for floating points?

[Extra-Creativity]

It may be frustrating to know c++17 feature isn't supported in the latest libc++ yet.

GCC 12 has partially adopted https://github.com/fastfloat/fast_float as its library implementation, is it possibly proper to adopt it in libc++?

[mordante]

I understand it's frustrating. However this feature is a lot of work even adapting an existing algorithm requires quite a bit of effort. Thusfar nobody has volunteered to do the work.

There is this WIP PR #91651 to use LLVM's libc implementation. I don't know which algorithm they use. IIRC it's one Lemire worked on so it might be fast float.

Since we don't use bugs to track missing parts of our implementation I'm closing the issue.

[Extra-Creativity]

I really appreciate volunteers' efforts. The mentioned PR seems not complete yet (not having error handling), but I believe this will be solved in the near future. Great thanks to you all.

[mordante]

I really appreciate volunteers' efforts.

Thanks!

The mentioned PR seems not complete yet (not having error handling), but I believe this will be solved in the near future. Great thanks to you all.

That PR indeed not complete and it may take some time. It is the first time libc++ and libc are sharing code. So we're also looking how to do that.

[michaelrj-google]

Hi! I'm still working on the proof of concept PR (I want to make sure all the tests pass) but I am hopeful that it's going to lead to a finished implementation before the end of the year. As for the algorithm it uses, it is primarily the one that Lemire worked on, I gave a talk on exactly how it works at the 2022 dev conference: https://www.youtube.com/watch?v=s-UjbTV8p6g

[lemire]

@michaelrj-google @ldionne

GCC (since GCC12) uses https://github.com/fastfloat/fast_float for from_chars. It is fully spec compliant and can just be dropped in with maybe a thin wrapper. It is an open community project. To my knowledge, it is effectively bug free at this point.

[philnik777]

@lemire Thanks for the suggestion, but we've decided to share the code with the LLVM libc. See #91651 for the status.

[lemire]

@philnik777 Looking forward to the implementation. We'll be able to benchmark!!! :-)

[ldionne]

@michaelrj-google @ldionne

GCC (since GCC12) uses https://github.com/fastfloat/fast_float for from_chars. It is fully spec compliant and can just be dropped in with maybe a thin wrapper. It is an open community project. To my knowledge, it is effectively bug free at this point.

Thanks for the ping. I actually wanted to reach out to you for your implementation, and then @michaelrj-google suggested using the LLVM libc implementation so we went for that. I'll be eager to see how the implementations compare performance-wise.

[lemire]

then @michaelrj-google suggested using the LLVM libc implementation so we went for that

I think that's 100% a good recommendation as far as I can tell. I'll be happier if we end up with competing implementations. :-)