merge bitcoin#23522: Improve fs::PathToString documentation

doc/developer-notes.md

@@ -1256,6 +1256,12 @@ A few guidelines for introducing and reviewing new RPC interfaces:
 
   - *Rationale*: User-facing consistency.
 
+- Use `fs::path::u8string()` and `fs::u8path()` functions when converting path
+  to JSON strings, not `fs::PathToString` and `fs::PathFromString`
+
+  - *Rationale*: JSON strings are Unicode strings, not byte strings, and
+    RFC8259 requires JSON to be encoded as UTF-8.
+
 Internal interface guidelines
 -----------------------------
 

src/dbwrapper.cpp

@@ -137,6 +137,10 @@ CDBWrapper::CDBWrapper(const fs::path& path, size_t nCacheSize, bool fMemory, bo
         TryCreateDirectories(path);
         LogPrintf("Opening LevelDB in %s\n", fs::PathToString(path));
     }
+    // PathToString() return value is safe to pass to leveldb open function,
+    // because on POSIX leveldb passes the byte string directly to ::open(), and
+    // on Windows it converts from UTF-8 to UTF-16 before calling ::CreateFileW
+    // (see env_posix.cc and env_windows.cc).
     leveldb::Status status = leveldb::DB::Open(options, fs::PathToString(path), &pdb);
     dbwrapper_private::HandleError(status);
     LogPrintf("Opened LevelDB successfully\n");

src/fs.h

@@ -94,31 +94,34 @@ static inline path operator+(path p1, path p2)
 
 /**
  * Convert path object to byte string. On POSIX, paths natively are byte
- * strings so this is trivial. On Windows, paths natively are Unicode, so an
+ * strings, so this is trivial. On Windows, paths natively are Unicode, so an
- * encoding step is necessary.
+ * encoding step is necessary. The inverse of \ref PathToString is \ref
+ * PathFromString. The strings returned and parsed by these functions can be
+ * used to call POSIX APIs, and for roundtrip conversion, logging, and
+ * debugging.
  *
- * The inverse of \ref PathToString is \ref PathFromString. The strings
+ * Because \ref PathToString and \ref PathFromString functions don't specify an
- * returned and parsed by these functions can be used to call POSIX APIs, and
+ * encoding, they are meant to be used internally, not externally. They are not
- * for roundtrip conversion, logging, and debugging. But they are not
+ * appropriate to use in applications requiring UTF-8, where
- * guaranteed to be valid UTF-8, and are generally meant to be used internally,
+ * fs::path::u8string() and fs::u8path() methods should be used instead. Other
- * not externally. When communicating with external programs and libraries that
+ * applications could require still different encodings. For example, JSON, XML,
- * require UTF-8, fs::path::u8string() and fs::u8path() methods can be used.
+ * or URI applications might prefer to use higher level escapes (\uXXXX or
- * For other applications, if support for non UTF-8 paths is required, or if
+ * &XXXX; or %XX) instead of multibyte encoding. Rust, Python, Java applications
- * higher-level JSON or XML or URI or C-style escapes are preferred, it may be
+ * may require encoding paths with their respective UTF-8 derivatives WTF-8,
- * also be appropriate to use different path encoding functions.
+ * PEP-383, and CESU-8 (see https://en.wikipedia.org/wiki/UTF-8#Derivatives).
- *
- * Implementation note: On Windows, the std::filesystem::path(string)
- * constructor and std::filesystem::path::string() method are not safe to use
- * here, because these methods encode the path using C++'s narrow multibyte
- * encoding, which on Windows corresponds to the current "code page", which is
- * unpredictable and typically not able to represent all valid paths. So
- * std::filesystem::path::u8string() and std::filesystem::u8path() functions
- * are used instead on Windows. On POSIX, u8string/u8path functions are not
- * safe to use because paths are not always valid UTF-8, so plain string
- * methods which do not transform the path there are used.
  */
 static inline std::string PathToString(const path& path)
 {
+    // Implementation note: On Windows, the std::filesystem::path(string)
+    // constructor and std::filesystem::path::string() method are not safe to
+    // use here, because these methods encode the path using C++'s narrow
+    // multibyte encoding, which on Windows corresponds to the current "code
+    // page", which is unpredictable and typically not able to represent all
+    // valid paths. So std::filesystem::path::u8string() and
+    // std::filesystem::u8path() functions are used instead on Windows. On
+    // POSIX, u8string/u8path functions are not safe to use because paths are
+    // not always valid UTF-8, so plain string methods which do not transform
+    // the path there are used.
 #ifdef WIN32
     return path.u8string();
 #else