[libcxx-commits] [PATCH] D70631: Microsoft's floating-point to_chars powered by Ryu and Ryu Printf

[David Blaikie via Phabricator via libcxx-commits]

dblaikie added inline comments.


================
Comment at: libcxx/src/include/to_chars_floating_point.h:119
+
+    const _Uint_type _Uint_value    = _VSTD::bit_cast<_Uint_type>(_Value);
+    const _Uint_type _Ieee_mantissa = _Uint_value & _Traits::_Denormal_mantissa_mask;
----------------
STL_MSFT wrote:
> floam wrote:
> > dblaikie wrote:
> > > STL_MSFT wrote:
> > > > dblaikie wrote:
> > > > > dblaikie wrote:
> > > > > > This bit_cast means the to_chars functionality (intended to be available in C++17 or above) fails when used in a C++17 context, since bit_cast is only available in C++20 onwards - at least in an internal Google build this seems to be the case (possible there's something esoteric there).
> > > > > > 
> > > > > > Does that seem plausible/a problem that needs fixing upstream here?
> > > > > Looking back at some of the review discussion, it looks like this was deemed OK because bit_cast would be available in the dylib even when compiling in an earlier language mode? But the declaration of the function is still gated on C++20 or above here: https://github.com/llvm/llvm-project/blob/572d1ecccc473ba4ddb46dd04759dc2e336f0e1c/libcxx/include/__bit/bit_cast.h#L22 so this usage fails to compile before ever worrying about whether it links correctly.
> > > > Good question about upstream - it's unaffected as we used an internal helper `_Bit_cast` precisely so that we could have access to it in C++17 mode. This is https://github.com/microsoft/STL/blob/3c2fd04d441d46ec9d914d9cbb621a3bac96c3a5/stl/inc/charconv#L2081 and https://github.com/microsoft/STL/blob/3c2fd04d441d46ec9d914d9cbb621a3bac96c3a5/stl/inc/xutility#L62-L74 .
> > > The code that was committed doesn't appear to use that `_Bit_cast` internal helper.
> > > The helper and its use seemed to have been removed in this iteration of the proposed patch: https://reviews.llvm.org/D70631?vs=380156&id=380203#toc
> > > The final commited code ( https://reviews.llvm.org/rGa8025e06fc0f2fe1bbee9e1a6f15c336bfbdcb05 ) have no mention of `_Bit_cast` and use `_VSTD::bit_cast` in several places, including this one (line 119 here that I'm commenting on).
> > > 
> > > Am I misunderstanding something? 
> > @dblaikie They are saying that is how Microsoft solved that quandary in //their product//.
> I think you're completely correct - I was merely clarifying that there is no issue that needs to be fixed upstream here, it was introduced during the later changes here.
Ah, sorry, I meant "upstream" as in "LLVM" here (as contrasted to "Downstream" being "inside google" - ie: something we need to fix on our end internally).

Sounds like we're on the same page here. Sorry for the confusion.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70631/new/

https://reviews.llvm.org/D70631