diff --git a/src/locale.cpp b/src/locale.cpp
index 4efdc63..219575b 100644
--- a/src/locale.cpp
+++ b/src/locale.cpp
@@ -38,8 +38,8 @@
 #  include <langinfo.h>
 #endif
 
-#include "include/atomic_support.h"
 #include "include/sso_allocator.h"
+#include <mutex>
 
 // On Linux, wint_t and wchar_t have different signed-ness, and this causes
 // lots of noise in the build log, but no bugs that I know of.
@@ -593,8 +593,17 @@ void locale::facet::__on_zero_shared() noexcept { delete this; }
 constinit int32_t locale::id::__next_id = 0;
 
 long locale::id::__get() {
-  call_once(__flag_, [&] { __id_ = __libcpp_atomic_add(&__next_id, 1); });
-  return __id_ - 1;
+  int32_t result = __id_.load(std::memory_order_acquire);
+  if (result == 0) {
+    static std::mutex m;
+    std::lock_guard<std::mutex> guard(m);
+    result = __id_.load(std::memory_order_acquire);
+    if (result == 0) {
+      result = ++__next_id;
+      __id_.store(result, std::memory_order_release);
+    }
+  }
+  return result - 1;
 }
 
 // template <> class collate_byname<char>
@@ -895,6 +904,11 @@ extern "C" const int** __ctype_tolower_loc();
 extern "C" const int** __ctype_toupper_loc();
 #endif
 
+#if defined(__ANDROID__)
+// See src/support/android/android_locale.cpp
+extern "C" const unsigned short* const _ctype_android;
+#endif
+
 #ifdef _LIBCPP_PROVIDES_DEFAULT_RUNE_TABLE
 const ctype<char>::mask* ctype<char>::classic_table() noexcept {
   // clang-format off
@@ -990,6 +1004,8 @@ const ctype<char>::mask* ctype<char>::classic_table() noexcept {
 #  elif defined(_NEWLIB_VERSION)
   // Newlib has a 257-entry table in ctype_.c, where (char)0 starts at [1].
   return _ctype_ + 1;
+#  elif defined(__ANDROID__)
+  return _ctype_android;
 #  elif defined(_AIX)
   return (const unsigned int*)__lc_ctype_ptr->obj->mask;
 #  elif defined(__MVS__)
@@ -3644,15 +3660,25 @@ __codecvt_utf16<char32_t, true>::result __codecvt_utf16<char32_t, true>::do_out(
     extern_type* to,
     extern_type* to_end,
     extern_type*& to_nxt) const {
+#if defined(_LIBCPP_SHORT_WCHAR)
+  const uint16_t* _frm     = reinterpret_cast<const uint16_t*>(frm);
+  const uint16_t* _frm_end = reinterpret_cast<const uint16_t*>(frm_end);
+  const uint16_t* _frm_nxt = _frm;
+#else
   const uint32_t* _frm     = reinterpret_cast<const uint32_t*>(frm);
   const uint32_t* _frm_end = reinterpret_cast<const uint32_t*>(frm_end);
   const uint32_t* _frm_nxt = _frm;
-  uint8_t* _to             = reinterpret_cast<uint8_t*>(to);
-  uint8_t* _to_end         = reinterpret_cast<uint8_t*>(to_end);
-  uint8_t* _to_nxt         = _to;
-  result r                 = ucs4_to_utf16le(_frm, _frm_end, _frm_nxt, _to, _to_end, _to_nxt, __maxcode_, __mode_);
-  frm_nxt                  = frm + (_frm_nxt - _frm);
-  to_nxt                   = to + (_to_nxt - _to);
+#endif
+  uint8_t* _to     = reinterpret_cast<uint8_t*>(to);
+  uint8_t* _to_end = reinterpret_cast<uint8_t*>(to_end);
+  uint8_t* _to_nxt = _to;
+#if defined(_LIBCPP_SHORT_WCHAR)
+  result r = ucs2_to_utf16le(_frm, _frm_end, _frm_nxt, _to, _to_end, _to_nxt, __maxcode_, __mode_);
+#else
+  result r = ucs4_to_utf16le(_frm, _frm_end, _frm_nxt, _to, _to_end, _to_nxt, __maxcode_, __mode_);
+#endif
+  frm_nxt = frm + (_frm_nxt - _frm);
+  to_nxt  = to + (_to_nxt - _to);
   return r;
 }
 
@@ -3712,7 +3738,7 @@ __codecvt_utf8_utf16<wchar_t>::result __codecvt_utf8_utf16<wchar_t>::do_out(
     extern_type* to,
     extern_type* to_end,
     extern_type*& to_nxt) const {
-#  if defined(_LIBCPP_SHORT_WCHAR)
+#  ifdef _WIN32
   const uint16_t* _frm     = reinterpret_cast<const uint16_t*>(frm);
   const uint16_t* _frm_end = reinterpret_cast<const uint16_t*>(frm_end);
   const uint16_t* _frm_nxt = _frm;
@@ -3741,7 +3767,7 @@ __codecvt_utf8_utf16<wchar_t>::result __codecvt_utf8_utf16<wchar_t>::do_in(
   const uint8_t* _frm     = reinterpret_cast<const uint8_t*>(frm);
   const uint8_t* _frm_end = reinterpret_cast<const uint8_t*>(frm_end);
   const uint8_t* _frm_nxt = _frm;
-#  if defined(_LIBCPP_SHORT_WCHAR)
+#  ifdef _WIN32
   uint16_t* _to     = reinterpret_cast<uint16_t*>(to);
   uint16_t* _to_end = reinterpret_cast<uint16_t*>(to_end);
   uint16_t* _to_nxt = _to;
@@ -3880,12 +3906,18 @@ __codecvt_utf8_utf16<char32_t>::result __codecvt_utf8_utf16<char32_t>::do_in(
   const uint8_t* _frm     = reinterpret_cast<const uint8_t*>(frm);
   const uint8_t* _frm_end = reinterpret_cast<const uint8_t*>(frm_end);
   const uint8_t* _frm_nxt = _frm;
-  uint32_t* _to           = reinterpret_cast<uint32_t*>(to);
-  uint32_t* _to_end       = reinterpret_cast<uint32_t*>(to_end);
-  uint32_t* _to_nxt       = _to;
-  result r                = utf8_to_utf16(_frm, _frm_end, _frm_nxt, _to, _to_end, _to_nxt, __maxcode_, __mode_);
-  frm_nxt                 = frm + (_frm_nxt - _frm);
-  to_nxt                  = to + (_to_nxt - _to);
+#if defined(_LIBCPP_SHORT_WCHAR)
+  uint16_t* _to     = reinterpret_cast<uint16_t*>(to);
+  uint16_t* _to_end = reinterpret_cast<uint16_t*>(to_end);
+  uint16_t* _to_nxt = _to;
+#else
+  uint32_t* _to     = reinterpret_cast<uint32_t*>(to);
+  uint32_t* _to_end = reinterpret_cast<uint32_t*>(to_end);
+  uint32_t* _to_nxt = _to;
+#endif
+  result r = utf8_to_utf16(_frm, _frm_end, _frm_nxt, _to, _to_end, _to_nxt, __maxcode_, __mode_);
+  frm_nxt  = frm + (_frm_nxt - _frm);
+  to_nxt   = to + (_to_nxt - _to);
   return r;
 }
 
@@ -5123,7 +5155,7 @@ static void __init_pat(
           // We insert the space into the symbol instead of
           // setting pat.field[2]=space so that when
           // showbase is not set, the space goes away too.
-          __curr_symbol_.insert(0, 1, space_char);
+          __curr_symbol_.insert((size_t)0, 1, space_char);
         }
         return;
       default:
@@ -5145,7 +5177,7 @@ static void __init_pat(
           // We insert the space into the symbol instead of
           // setting pat.field[2]=space so that when
           // showbase is not set, the space goes away too.
-          __curr_symbol_.insert(0, 1, space_char);
+          __curr_symbol_.insert((size_t)0, 1, space_char);
         }
         return;
       case 2: // Space between sign and currency or value.
@@ -5174,7 +5206,7 @@ static void __init_pat(
           // We insert the space into the symbol instead of
           // setting pat.field[1]=space so that when
           // showbase is not set, the space goes away too.
-          __curr_symbol_.insert(0, 1, space_char);
+          __curr_symbol_.insert((size_t)0, 1, space_char);
         }
         pat.field[1] = none;
         pat.field[2] = symbol;
@@ -5216,7 +5248,7 @@ static void __init_pat(
           // We insert the space into the symbol instead of
           // setting pat.field[2]=space so that when
           // showbase is not set, the space goes away too.
-          __curr_symbol_.insert(0, 1, space_char);
+          __curr_symbol_.insert((size_t)0, 1, space_char);
         }
         return;
       default:
@@ -5238,7 +5270,7 @@ static void __init_pat(
           // We insert the space into the symbol instead of
           // setting pat.field[1]=space so that when
           // showbase is not set, the space goes away too.
-          __curr_symbol_.insert(0, 1, space_char);
+          __curr_symbol_.insert((size_t)0, 1, space_char);
         }
         return;
       case 2: // Space between sign and currency or value.
@@ -5277,7 +5309,7 @@ static void __init_pat(
           // We insert the space into the symbol instead of
           // setting pat.field[2]=space so that when
           // showbase is not set, the space goes away too.
-          __curr_symbol_.insert(0, 1, space_char);
+          __curr_symbol_.insert((size_t)0, 1, space_char);
         }
         return;
       default: