<format> assumes strings are encoded in the active code page

stl/CMakeLists.txt

@@ -249,6 +249,7 @@ endforeach()
 # Objs that exist in both libcpmt[d][01].lib and msvcprt[d].lib.
 set(IMPLIB_SOURCES
     ${CMAKE_CURRENT_LIST_DIR}/src/filesystem.cpp
+    ${CMAKE_CURRENT_LIST_DIR}/src/format.cpp
     ${CMAKE_CURRENT_LIST_DIR}/src/locale0_implib.cpp
     ${CMAKE_CURRENT_LIST_DIR}/src/nothrow.cpp
     ${CMAKE_CURRENT_LIST_DIR}/src/sharedmutex.cpp

stl/inc/xfilesystem_abi.h

@@ -205,7 +205,7 @@ _BITMASK_OPS(__std_fs_file_flags)
 
 enum class __std_fs_file_handle : intptr_t { _Invalid = -1 };
 
-enum class __std_code_page : unsigned int { _Utf8 = 65001 };
+enum class __std_code_page : unsigned int { _Acp = 0, _Utf8 = 65001 };
 
 struct __std_fs_convert_result {
     int _Len;

stl/msbuild/stl_base/stl.files.settings.targets

@@ -161,6 +161,7 @@ SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
              (controlled by IncludeInLink and IncludeInImportLib). -->
         <BuildFiles Include="
             $(CrtRoot)\github\stl\src\filesystem.cpp;
+            $(CrtRoot)\github\stl\src\format.cpp;
             $(CrtRoot)\github\stl\src\locale0_implib.cpp;
             $(CrtRoot)\github\stl\src\nothrow.cpp;
             $(CrtRoot)\github\stl\src\sharedmutex.cpp;

stl/src/format.cpp

@@ -0,0 +1,45 @@
+// Copyright (c) Microsoft Corporation.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+// Implements a win32 API wrapper for <format>
+
+// This must be as small as possible, because its contents are
+// injected into the msvcprt.lib and msvcprtd.lib import libraries.
+// Do not include or define anything else here.
+// In particular, basic_string must not be included here.
+
+#include <xfilesystem_abi.h>
+#include <xlocinfo.h>
+
+#include <Windows.h>
+
+static_assert(__std_code_page::_Acp == __std_code_page{CP_ACP});
+
+extern "C" [[nodiscard]] __std_win_error __stdcall __std_get_cvt(
+    const __std_code_page _Codepage, _Cvtvec* const _Pcvt) noexcept {
+    // get conversion info for an arbitrary codepage
+    *_Pcvt = {};
+
+    CPINFOEXW _Info{};
+    const DWORD _Flags = 0; // reserved, must be zero
+    if (GetCPInfoExW(static_cast<UINT>(_Codepage), _Flags, &_Info) == 0) {
+        // NB: the only documented failure mode for GetCPInfoExW is ERROR_INVALID_PARAMETER,
+        // so in practice it should never fail for CP_ACP.
+        return __std_win_error{GetLastError()};
+    }
+
+    _Pcvt->_Page     = _Info.CodePage;
+    _Pcvt->_Mbcurmax = _Info.MaxCharSize;
+
+    for (int _Idx = 0; _Idx < MAX_LEADBYTES; _Idx += 2) {
+        if (_Info.LeadByte[_Idx] == 0 && _Info.LeadByte[_Idx + 1] == 0) {
+            break;
+        }
+
+        for (unsigned char _First = _Info.LeadByte[_Idx], _Last = _Info.LeadByte[_Idx + 1]; _First != _Last; ++_First) {
+            _Pcvt->_Isleadbyte[_First >> 3] |= 1u << (_First & 0b111u);
+        }
+    }
+
+    return __std_win_error::_Success;
+}

tests/std/include/test_format_support.hpp

@@ -117,9 +117,10 @@ void test_parse_helper(const CharT* (*func)(const CharT*, const CharT*, callback
     callback_type&& callbacks                                               = {}) {
     try {
         auto end = func(view.data(), view.data() + view.size(), std::move(callbacks));
-        if (expected_end_position != std::basic_string_view<CharT>::npos) {
-            assert(end == view.data() + expected_end_position);
+        if (expected_end_position == std::basic_string_view<CharT>::npos) {
+            expected_end_position = view.size();
         }
+        assert(end == view.data() + expected_end_position);
         assert(!err_expected);
     } catch (const std::format_error&) {
         assert(err_expected);

tests/std/tests/P0645R10_text_formatting_formatting/test.cpp

@@ -975,31 +975,6 @@ void test_size() {
     test_size_helper<charT>(8, STR("{:8}"), STR("scully"));
 }
 
-void test_multibyte_format_strings() {
-#ifndef MSVC_INTERNAL_TESTING // TRANSITION, the Windows version on Contest VMs doesn't always understand ".UTF-8"
-    {
-        assert(setlocale(LC_ALL, ".UTF-8") != nullptr);
-        // Filling with footballs ("\xf0\x9f\x8f\x88" is U+1F3C8 AMERICAN FOOTBALL)
-        assert(format("{:\xf0\x9f\x8f\x88>4}"sv, 42) == "\xf0\x9f\x8f\x88\xf0\x9f\x8f\x88\x34\x32");
-
-        assert(format("{:\xf0\x9f\x8f\x88<4.2}", "1") == "\x31\xf0\x9f\x8f\x88\xf0\x9f\x8f\x88\xf0\x9f\x8f\x88"sv);
-        assert(format("{:\xf0\x9f\x8f\x88^4.2}", "1") == "\xf0\x9f\x8f\x88\x31\xf0\x9f\x8f\x88\xf0\x9f\x8f\x88"sv);
-        assert(format("{:\xf0\x9f\x8f\x88>4.2}", "1") == "\xf0\x9f\x8f\x88\xf0\x9f\x8f\x88\xf0\x9f\x8f\x88\x31"sv);
-    }
-
-    {
-        assert(setlocale(LC_ALL, ".UTF-8") != nullptr);
-        try {
-            (void) format("{:\x9f\x8f\x88<10}"sv, 42); // Bad fill character encoding: missing lead byte before \x9f
-            assert(false);
-        } catch (const format_error&) {
-        }
-    }
-#endif // MSVC_INTERNAL_TESTING
-
-    assert(setlocale(LC_ALL, "C") != nullptr);
-}
-
 // The libfmt_ tests are derived from tests in
 // libfmt, Copyright (c) 2012 - present, Victor Zverovich
 // See NOTICE.txt for more information.
@@ -1318,8 +1293,6 @@ void test() {
     test_size<char>();
     test_size<wchar_t>();
 
-    test_multibyte_format_strings();
-
     libfmt_formatter_test_escape<char>();
     libfmt_formatter_test_escape<wchar_t>();
 

tests/std/tests/P0645R10_text_formatting_legacy_text_encoding/test.cpp

@@ -1,6 +1,8 @@
 // Copyright (c) Microsoft Corporation.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
+#define _FORMAT_CODEPAGE (__std_code_page{932})
+
 #include <cassert>
 #include <clocale>
 #include <format>
@@ -11,52 +13,41 @@
 using namespace std;
 
 void test_multibyte_format_strings() {
-    {
-        assert(setlocale(LC_ALL, ".932") != nullptr);
-        const auto s =
-            "\x93\xfa\x96{\x92\x6e\x90}"sv; // Note the use of `{` and `}` as continuation bytes (from GH-1576)
-        assert(format(s) == s);
-
-        assert(format("{:.2}", s) == "\x93\xfa"sv);
-        assert(format("{:4.2}", s) == "\x93\xfa  "sv);
+    const auto s = "\x93\xfa\x96{\x92\x6e\x90}"sv; // Note the use of `{` and `}` as continuation bytes (from GH-1576)
+    assert(format(s) == s);
 
-        assert(format("{:<4.2}", s) == "\x93\xfa  "sv);
-        assert(format("{:^4.2}", s) == " \x93\xfa "sv);
-        assert(format("{:>4.2}", s) == "  \x93\xfa"sv);
+    assert(format("{:.2}", s) == "\x93\xfa"sv);
+    assert(format("{:4.2}", s) == "\x93\xfa  "sv);
 
-        assert(format("{:\x90}<4.2}", s) == "\x93\xfa\x90}\x90}"sv);
-        assert(format("{:\x90}^4.2}", s) == "\x90}\x93\xfa\x90}"sv);
-        assert(format("{:\x90}>4.2}", s) == "\x90}\x90}\x93\xfa"sv);
+    assert(format("{:<4.2}", s) == "\x93\xfa  "sv);
+    assert(format("{:^4.2}", s) == " \x93\xfa "sv);
+    assert(format("{:>4.2}", s) == "  \x93\xfa"sv);
 
-        assert(format("{:.3}", s) == "\x93\xfa"sv);
-        assert(format("{:4.3}", s) == "\x93\xfa  "sv);
+    assert(format("{:\x90}<4.2}", s) == "\x93\xfa\x90}\x90}"sv);
+    assert(format("{:\x90}^4.2}", s) == "\x90}\x93\xfa\x90}"sv);
+    assert(format("{:\x90}>4.2}", s) == "\x90}\x90}\x93\xfa"sv);
 
-        assert(format("{:<4.3}", s) == "\x93\xfa  "sv);
-        assert(format("{:^4.3}", s) == " \x93\xfa "sv);
-        assert(format("{:>4.3}", s) == "  \x93\xfa"sv);
+    assert(format("{:.3}", s) == "\x93\xfa"sv);
+    assert(format("{:4.3}", s) == "\x93\xfa  "sv);
 
-        assert(format("{:\x90}<4.3}", s) == "\x93\xfa\x90}\x90}"sv);
-        assert(format("{:\x90}^4.3}", s) == "\x90}\x93\xfa\x90}"sv);
-        assert(format("{:\x90}>4.3}", s) == "\x90}\x90}\x93\xfa"sv);
-    }
+    assert(format("{:<4.3}", s) == "\x93\xfa  "sv);
+    assert(format("{:^4.3}", s) == " \x93\xfa "sv);
+    assert(format("{:>4.3}", s) == "  \x93\xfa"sv);
 
-    assert(setlocale(LC_ALL, "C") != nullptr);
+    assert(format("{:\x90}<4.3}", s) == "\x93\xfa\x90}\x90}"sv);
+    assert(format("{:\x90}^4.3}", s) == "\x90}\x93\xfa\x90}"sv);
+    assert(format("{:\x90}>4.3}", s) == "\x90}\x90}\x93\xfa"sv);
 }
 
 void test_parse_align() {
-    auto parse_align_fn = _Parse_align<char, testing_callbacks<char>>;
-
-    {
-        assert(setlocale(LC_ALL, ".932") != nullptr);
-        test_parse_helper(parse_align_fn, "\x93\xfa<X"sv, false, 3,
-            {.expected_alignment = _Fmt_align::_Left, .expected_fill = "\x93\xfa"sv});
-        test_parse_helper(parse_align_fn, "\x96\x7b>X"sv, false, 3,
-            {.expected_alignment = _Fmt_align::_Right, .expected_fill = "\x96\x7b"sv});
-        test_parse_helper(parse_align_fn, "\x92\x6e^X"sv, false, 3,
-            {.expected_alignment = _Fmt_align::_Center, .expected_fill = "\x92\x6e"sv});
-    }
-
-    assert(setlocale(LC_ALL, "C") != nullptr);
+    const auto parse_align_fn = _Parse_align<char, testing_callbacks<char>>;
+
+    test_parse_helper(parse_align_fn, "\x93\xfa<X"sv, false, 3, //
+        {.expected_alignment = _Fmt_align::_Left, .expected_fill = "\x93\xfa"sv});
+    test_parse_helper(parse_align_fn, "\x96\x7b>X"sv, false, 3,
+        {.expected_alignment = _Fmt_align::_Right, .expected_fill = "\x96\x7b"sv});
+    test_parse_helper(parse_align_fn, "\x92\x6e^X"sv, false, 3,
+        {.expected_alignment = _Fmt_align::_Center, .expected_fill = "\x92\x6e"sv});
 }
 
 int main() {

tests/std/tests/P0645R10_text_formatting_parsing/test.cpp

@@ -36,7 +36,7 @@ bool test_parse_align() {
         // \x343E (which is from CJK unified ideographs extension A) and similar characters to parse as
         // an alignment specifier.
         auto s4 = L"*\x343E"sv;
-        test_parse_helper(parse_align_fn, s4, false, view_typ::npos, {.expected_fill = L"*"sv});
+        test_parse_helper(parse_align_fn, s4, false, 0, {.expected_fill = L"*"sv});
 
         // test multi-code-unit fill characters
         {
@@ -47,22 +47,6 @@ bool test_parse_align() {
             test_parse_helper(parse_align_fn, L"\U0001F3C8^X"sv, false, 3,
                 {.expected_alignment = _Fmt_align::_Center, .expected_fill = L"\U0001F3C8"sv});
         }
-    } else {
-        // test multibyte fill characters
-#ifndef MSVC_INTERNAL_TESTING // TRANSITION, the Windows version on Contest VMs doesn't always understand ".UTF-8"
-        {
-            assert(setlocale(LC_ALL, ".UTF-8") != nullptr);
-            // "\xf0\x9f\x8f\x88" is U+1F3C8 AMERICAN FOOTBALL
-            test_parse_helper(parse_align_fn, "\xf0\x9f\x8f\x88<X"sv, false, 5,
-                {.expected_alignment = _Fmt_align::_Left, .expected_fill = "\xf0\x9f\x8f\x88"sv});
-            test_parse_helper(parse_align_fn, "\xf0\x9f\x8f\x88>X"sv, false, 5,
-                {.expected_alignment = _Fmt_align::_Right, .expected_fill = "\xf0\x9f\x8f\x88"sv});
-            test_parse_helper(parse_align_fn, "\xf0\x9f\x8f\x88^X"sv, false, 5,
-                {.expected_alignment = _Fmt_align::_Center, .expected_fill = "\xf0\x9f\x8f\x88"sv});
-        }
-#endif // MSVC_INTERNAL_TESTING
-
-        assert(setlocale(LC_ALL, "C") != nullptr);
     }
 
     return true;

tests/std/tests/P0645R10_text_formatting_utf8/test.cpp

@@ -1,6 +1,8 @@
 // Copyright (c) Microsoft Corporation.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
+#define _FORMAT_CODEPAGE __std_code_page::_Utf8
+
 #include <cassert>
 #include <clocale>
 #include <format>