P0067R5: Elementary string conversions

[cjdb]

Link: https://wg21.link/p0067r5

BEGIN-RST-NOTES
For integer types, ``std::(to|from)_chars`` has been available since v7; for ``float`` and ``double``, ``std::to_chars`` since v14 and ``std::from_chars`` since v20. Support is complete except for ``long double``.
END-RST-NOTES

[llvmbot]

@llvm/issue-subscribers-c-17

Author: Christopher Di Bella (cjdb)

[h-vetinari]

I'm glad there are finally issues for these topics; that makes it much easier to collect publicly available information about them, rather have them scattered in the heads of core libcxx contributors.

• [libc++] std::to_chars precision loss for 80-bit long double #56261
• Unexpected error when using libc++ (call to deleted function 'from_chars') #59374
• Floating point std::to_chars has a pretty large code size impact #64180
• Any plan for libc++ to support std::from_chars for floating points? #92711

Also some context I had kept from long ago

@h-vetinari: Stumbled over this while looking for the status of P0067R5; took me a while to unearth some old comments on this topic in https://reviews.llvm.org/D70631:

@h-vetinari: I was under the impression that 87c0160 did most of the work for from_chars?

@mordante: Thanks for both links I wasn't aware of this effort!
At the moment I'm too busy with <format> to look at from_chars, but this is useful when somebody starts working on from_chars. (I might look at it when I finished the larger projects I'm working on.)

[lemire]

The from_chars issue could be largely solved with https://github.com/fastfloat/fast_float

It is what GCC/glibc++ does.

I also left a comment in #92711

I'd be happy to help.

[h-vetinari]

The implementation of the remaining bits landed in #91651, which completes this paper except for long-double support (needs #113237). Feature test macro wasn't set yet.

[frederick-vs-ja]

Looks like that we're now very close to setting FTM on long double-eq-double platforms.

[cor3ntin]

@frederick-vs-ja I'm trying to figure out what the status of that is since you closed your pr. Is it actually complete ? Thanks!

[philnik777]

@cor3ntin No, it's not complete. There is no fundamental hurdle to adding the long double versions (except on some funky IBM system), but someone has to write the tests for it.

[mordante]

but someone has to write the tests for it.

There is some implementation work to do too, especially for to_chars. Just to add a bit more information, adding the tests is no small effort. (I expect it to take over a week of work.)