[libcxx][libc] Hand in Hand PoC with from_chars (#91651)

Implements std::from_chars for float and double.

The implementation uses LLVM-libc to do the real parsing. Since this is
the first time libc++
uses LLVM-libc there is a bit of additional infrastructure code. The
patch is based on the
[RFC] Project Hand In Hand (LLVM-libc/libc++ code sharing)

https://discourse.llvm.org/t/rfc-project-hand-in-hand-llvm-libc-libc-code-sharing/77701

libc/shared/fp_bits.h

@@ -0,0 +1,22 @@
+//===-- Floating point number utils -----------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIBC_SHARED_FP_BITS_H
+#define LLVM_LIBC_SHARED_FP_BITS_H
+
+#include "src/__support/FPUtil/FPBits.h"
+
+namespace LIBC_NAMESPACE_DECL {
+namespace shared {
+
+using fputil::FPBits;
+
+} // namespace shared
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LLVM_LIBC_SHARED_FP_BITS_H

libc/shared/str_to_float.h

@@ -0,0 +1,27 @@
+//===-- String to float conversion utils ------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIBC_SHARED_STR_TO_FLOAT_H
+#define LLVM_LIBC_SHARED_STR_TO_FLOAT_H
+
+#include "src/__support/str_to_float.h"
+
+namespace LIBC_NAMESPACE_DECL {
+namespace shared {
+
+using internal::ExpandedFloat;
+using internal::FloatConvertReturn;
+using internal::RoundDirection;
+
+using internal::binary_exp_to_float;
+using internal::decimal_exp_to_float;
+
+} // namespace shared
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LLVM_LIBC_SHARED_STR_TO_FLOAT_H

libc/shared/str_to_integer.h

@@ -0,0 +1,24 @@
+//===-- String to int conversion utils --------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIBC_SHARED_STR_TO_INTEGER_H
+#define LLVM_LIBC_SHARED_STR_TO_INTEGER_H
+
+#include "src/__support/str_to_integer.h"
+
+namespace LIBC_NAMESPACE_DECL {
+namespace shared {
+
+using LIBC_NAMESPACE::StrToNumResult;
+
+using internal::strtointeger;
+
+} // namespace shared
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LLVM_LIBC_SHARED_STR_TO_INTEGER_H

libc/src/__support/FPUtil/FPBits.h

@@ -6,6 +6,12 @@
 //
 //===----------------------------------------------------------------------===//
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This file is shared with libc++. You should also be careful when adding
+// dependencies to this file, since it needs to build for all libc++ targets.
+// -----------------------------------------------------------------------------
+
 #ifndef LLVM_LIBC_SRC___SUPPORT_FPUTIL_FPBITS_H
 #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_FPBITS_H
 
@@ -795,6 +801,12 @@ template <typename T> LIBC_INLINE static constexpr FPType get_fp_type() {
     static_assert(cpp::always_false<UnqualT>, "Unsupported type");
 }
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++. You should also be careful when adding
+// dependencies to this file, since it needs to build for all libc++ targets.
+// -----------------------------------------------------------------------------
 // A generic class to manipulate C++ floating point formats.
 // It derives its functionality to FPRepImpl above.
 template <typename T>

libc/src/__support/high_precision_decimal.h

@@ -6,6 +6,12 @@
 //
 //===----------------------------------------------------------------------===//
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This file is shared with libc++. You should also be careful when adding
+// dependencies to this file, since it needs to build for all libc++ targets.
+// -----------------------------------------------------------------------------
+
 #ifndef LLVM_LIBC_SRC___SUPPORT_HIGH_PRECISION_DECIMAL_H
 #define LLVM_LIBC_SRC___SUPPORT_HIGH_PRECISION_DECIMAL_H
 
@@ -23,6 +29,11 @@ struct LShiftTableEntry {
   char const *power_of_five;
 };
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++.
+// -----------------------------------------------------------------------------
 // This is used in both this file and in the main str_to_float.h.
 // TODO: Figure out where to put this.
 enum class RoundDirection { Up, Down, Nearest };

libc/src/__support/str_to_float.h

@@ -6,6 +6,12 @@
 //
 //===----------------------------------------------------------------------===//
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This file is shared with libc++. You should also be careful when adding
+// dependencies to this file, since it needs to build for all libc++ targets.
+// -----------------------------------------------------------------------------
+
 #ifndef LLVM_LIBC_SRC___SUPPORT_STR_TO_FLOAT_H
 #define LLVM_LIBC_SRC___SUPPORT_STR_TO_FLOAT_H
 
@@ -32,11 +38,21 @@
 namespace LIBC_NAMESPACE_DECL {
 namespace internal {
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++.
+// -----------------------------------------------------------------------------
 template <class T> struct ExpandedFloat {
   typename fputil::FPBits<T>::StorageType mantissa;
   int32_t exponent;
 };
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++.
+// -----------------------------------------------------------------------------
 template <class T> struct FloatConvertReturn {
   ExpandedFloat<T> num = {0, 0};
   int error = 0;
@@ -637,6 +653,11 @@ template <> LIBC_INLINE constexpr int32_t get_lower_bound<double>() {
   return -(309 + 15 + 20);
 }
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++.
+// -----------------------------------------------------------------------------
 // Takes a mantissa and base 10 exponent and converts it into its closest
 // floating point type T equivalient. First we try the Eisel-Lemire algorithm,
 // then if that fails then we fall back to a more accurate algorithm for
@@ -716,6 +737,11 @@ LIBC_INLINE FloatConvertReturn<T> decimal_exp_to_float(
   return output;
 }
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++.
+// -----------------------------------------------------------------------------
 // Takes a mantissa and base 2 exponent and converts it into its closest
 // floating point type T equivalient. Since the exponent is already in the right
 // form, this is mostly just shifting and rounding. This is used for hexadecimal

libc/src/__support/str_to_integer.h

@@ -6,6 +6,12 @@
 //
 //===----------------------------------------------------------------------===//
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This file is shared with libc++. You should also be careful when adding
+// dependencies to this file, since it needs to build for all libc++ targets.
+// -----------------------------------------------------------------------------
+
 #ifndef LLVM_LIBC_SRC___SUPPORT_STR_TO_INTEGER_H
 #define LLVM_LIBC_SRC___SUPPORT_STR_TO_INTEGER_H
 
@@ -73,6 +79,11 @@ LIBC_INLINE int infer_base(const char *__restrict src, size_t src_len) {
   return 10;
 }
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++.
+// -----------------------------------------------------------------------------
 // Takes a pointer to a string and the base to convert to. This function is used
 // as the backend for all of the string to int functions.
 template <class T>

libc/src/__support/str_to_num_result.h

@@ -6,6 +6,12 @@
 //
 //===----------------------------------------------------------------------===//
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This file is shared with libc++. You should also be careful when adding
+// dependencies to this file, since it needs to build for all libc++ targets.
+// -----------------------------------------------------------------------------
+
 #ifndef LLVM_LIBC_SRC___SUPPORT_STR_TO_NUM_RESULT_H
 #define LLVM_LIBC_SRC___SUPPORT_STR_TO_NUM_RESULT_H
 
@@ -16,6 +22,11 @@
 
 namespace LIBC_NAMESPACE_DECL {
 
+// -----------------------------------------------------------------------------
+//                               **** WARNING ****
+// This interface is shared with libc++, if you change this interface you need
+// to update it in both libc and libc++.
+// -----------------------------------------------------------------------------
 template <typename T> struct StrToNumResult {
   T value;
   int error;

libcxx/docs/Status/Cxx17Papers.csv

@@ -71,7 +71,7 @@
 "`P0394R4 <https://wg21.link/P0394R4>`__","Hotel Parallelifornia: terminate() for Parallel Algorithms Exception Handling","2016-06 (Oulu)","|Complete|","17.0",""
 "","","","","",""
 "`P0003R5 <https://wg21.link/P0003R5>`__","Removing Deprecated Exception Specifications from C++17","2016-11 (Issaquah)","|Complete|","5.0",""
-"`P0067R5 <https://wg21.link/P0067R5>`__","Elementary string conversions, revision 5","2016-11 (Issaquah)","|Partial|","","``std::(to|from)_chars`` for integrals has been available since version 7.0. ``std::to_chars`` for ``float`` and ``double`` since version 14.0 ``std::to_chars`` for ``long double`` uses the implementation for ``double``."
+"`P0067R5 <https://wg21.link/P0067R5>`__","Elementary string conversions, revision 5","2016-11 (Issaquah)","|Partial|","","``std::(to|from)_chars`` for integrals has been available since version 7.0. ``std::to_chars`` for ``float`` and ``double`` since version 14.0 ``std::to_chars`` for ``long double`` uses the implementation for ``double``. ``std::from_chars`` for ``float`` and ``double`` since version 20.0."
 "`P0403R1 <https://wg21.link/P0403R1>`__","Literal suffixes for ``basic_string_view``\ ","2016-11 (Issaquah)","|Complete|","4.0",""
 "`P0414R2 <https://wg21.link/P0414R2>`__","Merging shared_ptr changes from Library Fundamentals to C++17","2016-11 (Issaquah)","|Complete|","11.0",""
 "`P0418R2 <https://wg21.link/P0418R2>`__","Fail or succeed: there is no atomic lattice","2016-11 (Issaquah)","","",""

libcxx/docs/Status/Cxx2cIssues.csv

@@ -78,4 +78,5 @@
 "","","","","",""
 "`LWG3343 <https://wg21.link/LWG3343>`__","Ordering of calls to ``unlock()`` and ``notify_all()`` in Effects element of ``notify_all_at_thread_exit()`` should be reversed","Not Adopted Yet","|Complete|","16.0",""
 "`LWG4139 <https://wg21.link/LWG4139>`__","§[time.zone.leap] recursive constraint in <=>","Not Adopted Yet","|Complete|","20.0",""
+"`LWG3456 <https://wg21.link/LWG3456>`__","Pattern used by std::from_chars is underspecified (option B)",,"Not Yet Adopted","|Complete|","20.0",""
 "","","","","",""

libcxx/include/CMakeLists.txt

@@ -235,6 +235,7 @@ set(files
   __bit/rotate.h
   __bit_reference
   __charconv/chars_format.h
+  __charconv/from_chars_floating_point.h
   __charconv/from_chars_integral.h
   __charconv/from_chars_result.h
   __charconv/tables.h

libcxx/include/__charconv/from_chars_floating_point.h

@@ -0,0 +1,73 @@
+// -*- C++ -*-
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___CHARCONV_FROM_CHARS_FLOATING_POINT_H
+#define _LIBCPP___CHARCONV_FROM_CHARS_FLOATING_POINT_H
+
+#include <__assert>
+#include <__charconv/chars_format.h>
+#include <__charconv/from_chars_result.h>
+#include <__config>
+#include <__system_error/errc.h>
+#include <cstddef>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+#if _LIBCPP_STD_VER >= 17
+
+template <class _Fp>
+struct __from_chars_result {
+  _Fp __value;
+  ptrdiff_t __n;
+  errc __ec;
+};
+
+template <class _Fp>
+_LIBCPP_EXPORTED_FROM_ABI __from_chars_result<_Fp> __from_chars_floating_point(
+    [[clang::noescape]] const char* __first, [[clang::noescape]] const char* __last, chars_format __fmt);
+
+extern template __from_chars_result<float> __from_chars_floating_point(
+    [[clang::noescape]] const char* __first, [[clang::noescape]] const char* __last, chars_format __fmt);
+
+extern template __from_chars_result<double> __from_chars_floating_point(
+    [[clang::noescape]] const char* __first, [[clang::noescape]] const char* __last, chars_format __fmt);
+
+template <class _Fp>
+_LIBCPP_HIDE_FROM_ABI from_chars_result
+__from_chars(const char* __first, const char* __last, _Fp& __value, chars_format __fmt) {
+  __from_chars_result<_Fp> __r = std::__from_chars_floating_point<_Fp>(__first, __last, __fmt);
+  if (__r.__ec != errc::invalid_argument)
+    __value = __r.__value;
+  return {__first + __r.__n, __r.__ec};
+}
+
+_LIBCPP_AVAILABILITY_FROM_CHARS_FLOATING_POINT _LIBCPP_HIDE_FROM_ABI inline from_chars_result
+from_chars(const char* __first, const char* __last, float& __value, chars_format __fmt = chars_format::general) {
+  return std::__from_chars<float>(__first, __last, __value, __fmt);
+}
+
+_LIBCPP_AVAILABILITY_FROM_CHARS_FLOATING_POINT _LIBCPP_HIDE_FROM_ABI inline from_chars_result
+from_chars(const char* __first, const char* __last, double& __value, chars_format __fmt = chars_format::general) {
+  return std::__from_chars<double>(__first, __last, __value, __fmt);
+}
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_END_NAMESPACE_STD
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___CHARCONV_FROM_CHARS_FLOATING_POINT_H

libcxx/include/__configuration/availability.h

@@ -87,6 +87,9 @@
 // in all versions of the library are available.
 #if defined(_LIBCPP_HAS_NO_VENDOR_AVAILABILITY_ANNOTATIONS)
 
+#  define _LIBCPP_INTRODUCED_IN_LLVM_20 1
+#  define _LIBCPP_INTRODUCED_IN_LLVM_20_ATTRIBUTE /* nothing */
+
 #  define _LIBCPP_INTRODUCED_IN_LLVM_19 1
 #  define _LIBCPP_INTRODUCED_IN_LLVM_19_ATTRIBUTE /* nothing */
 
@@ -132,6 +135,11 @@
 
 // clang-format off
 
+// LLVM 20
+// TODO: Fill this in
+#  define _LIBCPP_INTRODUCED_IN_LLVM_20 0
+#  define _LIBCPP_INTRODUCED_IN_LLVM_20_ATTRIBUTE __attribute__((unavailable))
+
 // LLVM 19
 // TODO: Fill this in
 #  define _LIBCPP_INTRODUCED_IN_LLVM_19 0
@@ -409,6 +417,11 @@
 #define _LIBCPP_AVAILABILITY_HAS_BAD_EXPECTED_ACCESS_KEY_FUNCTION _LIBCPP_INTRODUCED_IN_LLVM_19
 #define _LIBCPP_AVAILABILITY_BAD_EXPECTED_ACCESS_KEY_FUNCTION _LIBCPP_INTRODUCED_IN_LLVM_19_ATTRIBUTE
 
+// This controls the availability of floating-point std::from_chars functions.
+// These overloads were added later than the integer overloads.
+#define _LIBCPP_AVAILABILITY_HAS_FROM_CHARS_FLOATING_POINT _LIBCPP_INTRODUCED_IN_LLVM_20
+#define _LIBCPP_AVAILABILITY_FROM_CHARS_FLOATING_POINT _LIBCPP_INTRODUCED_IN_LLVM_20_ATTRIBUTE
+
 // Define availability attributes that depend on _LIBCPP_HAS_EXCEPTIONS.
 // Those are defined in terms of the availability attributes above, and
 // should not be vendor-specific.

libcxx/include/charconv

@@ -65,6 +65,12 @@ namespace std {
   constexpr from_chars_result from_chars(const char* first, const char* last,
                                see below& value, int base = 10);                         // constexpr since C++23
 
+  from_chars_result from_chars(const char* first, const char* last,
+                               float& value, chars_format fmt);
+
+  from_chars_result from_chars(const char* first, const char* last,
+                               double& value, chars_format fmt);
+
 } // namespace std
 
 */
@@ -73,6 +79,7 @@ namespace std {
 
 #if _LIBCPP_STD_VER >= 17
 #  include <__charconv/chars_format.h>
+#  include <__charconv/from_chars_floating_point.h>
 #  include <__charconv/from_chars_integral.h>
 #  include <__charconv/from_chars_result.h>
 #  include <__charconv/tables.h>

libcxx/include/module.modulemap

@@ -898,6 +898,7 @@ module std [system] {
 
   module charconv {
     module chars_format               { header "__charconv/chars_format.h" }
+    module from_chars_floating_point  { header "__charconv/from_chars_floating_point.h" }
     module from_chars_integral        { header "__charconv/from_chars_integral.h" }
     module from_chars_result          { header "__charconv/from_chars_result.h" }
     module tables                     { header "__charconv/tables.h" }