Fix canonicalization of non-ASCII paths on Windows

On Windows, existing_path_length counts UTF-16 code units but is used as
an index into a UTF-8 string. This causes problems when canonicalizing
paths containing non-ASCII characters.

Fix indexing to be consistent with the data's encoding.

Author: strager

src/file-canonical.cpp

@@ -567,11 +567,14 @@ class windows_path_canonicalizer
 
   canonical_path_result result() {
     // HACK(strager): Convert UTF-16 to UTF-8.
-    // TODO(strager): existing_path_length_ is in UTF-16 code units, but it's
-    // interpreted as UTF-8 code units! Fix by storing a std::wstring in
-    // canonical_path.
-    return canonical_path_result(std::filesystem::path(canonical_).string(),
-                                 existing_path_length_);
+    std::string canonical_utf_8 =
+        to_string(std::filesystem::path(canonical_).u8string());
+    std::size_t existing_path_length_utf_8 =
+        count_utf_8_code_units(std::u16string_view(
+            reinterpret_cast<const char16_t *>(canonical_.data()),
+            existing_path_length_));
+    return canonical_path_result(std::move(canonical_utf_8),
+                                 existing_path_length_utf_8);
   }
 
   quick_lint_js::result<void, canonicalizing_path_io_error>

src/temporary-directory.cpp

@@ -8,6 +8,7 @@
 #include <cstring>
 #include <iostream>
 #include <limits.h>
+#include <quick-lint-js/char8.h>
 #include <quick-lint-js/have.h>
 #include <quick-lint-js/temporary-directory.h>
 #include <quick-lint-js/unreachable.h>
@@ -74,7 +75,7 @@ std::string make_temporary_directory() {
 
 void create_directory(const std::string &path) {
 #if QLJS_HAVE_STD_FILESYSTEM
-  std::filesystem::create_directory(path);
+  std::filesystem::create_directory(to_string8(path));
 #else
   if (::mkdir(path.c_str(), 0755) != 0) {
     std::fprintf(stderr, "error: failed to create directory %s: %s\n",

test/test-file-canonical.cpp

@@ -33,6 +33,7 @@
 using ::testing::AnyOf;
 using ::testing::HasSubstr;
 using ::testing::Not;
+using namespace std::literals::string_view_literals;
 
 namespace quick_lint_js {
 namespace {
@@ -185,6 +186,31 @@ TEST_F(test_file_canonical,
   EXPECT_EQ(canonical->path(), temp_dir_canonical->path());
 }
 
+TEST_F(test_file_canonical,
+       canonical_path_to_non_existing_file_with_non_ascii_succeeds) {
+  for (string8_view character8 : {u8"\u00e0"sv, u8"\u0800"sv}) {
+    std::string character = to_string(character8);
+    SCOPED_TRACE(character);
+
+    std::string temp_dir = this->make_temporary_directory();
+    create_directory(temp_dir + "/parent" + character + "dir");
+    result<canonical_path_result, canonicalize_path_io_error>
+        parent_dir_canonical =
+            canonicalize_path(temp_dir + "/parent" + character + "dir");
+    ASSERT_TRUE(parent_dir_canonical.ok())
+        << parent_dir_canonical.error().to_string();
+
+    result<canonical_path_result, canonicalize_path_io_error> canonical =
+        canonicalize_path(temp_dir + "/parent" + character + "dir/does-not-" +
+                          character + "exist/file" + character + "name.txt");
+    ASSERT_TRUE(canonical.ok()) << canonical.error().to_string();
+
+    EXPECT_TRUE(canonical->have_missing_components());
+    canonical->drop_missing_components();
+    EXPECT_EQ(canonical->path(), parent_dir_canonical->path());
+  }
+}
+
 TEST_F(test_file_canonical, canonical_path_with_file_parent_fails) {
   std::string temp_dir = this->make_temporary_directory();
   write_file(temp_dir + "/file", u8"");