bpo-43667: Fix broken Unicode encoding in non-UTF locales on Solaris (pythonGH-25096)

Author: kulikjak

Diff for: Include/internal/pycore_fileutils.h

@@ -53,6 +53,18 @@ PyAPI_FUNC(void) _Py_closerange(int first, int last);
 PyAPI_FUNC(wchar_t*) _Py_GetLocaleEncoding(void);
 PyAPI_FUNC(PyObject*) _Py_GetLocaleEncodingObject(void);
 
+#ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+extern int _Py_LocaleUsesNonUnicodeWchar(void);
+
+extern wchar_t* _Py_DecodeNonUnicodeWchar(
+    const wchar_t* native,
+    Py_ssize_t size);
+
+extern int _Py_EncodeNonUnicodeWchar_InPlace(
+    wchar_t* unicode,
+    Py_ssize_t size);
+#endif
+
 #ifdef __cplusplus
 }
 #endif

Diff for: Objects/unicodeobject.c

@@ -57,6 +57,10 @@ OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 #include <windows.h>
 #endif
 
+#ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+#include "pycore_fileutils.h"     // _Py_LocaleUsesNonUnicodeWchar()
+#endif
+
 /* Uncomment to display statistics on interned strings at exit
    in _PyUnicode_ClearInterned(). */
 /* #define INTERNED_STATS 1 */
@@ -2217,6 +2221,20 @@ PyUnicode_FromWideChar(const wchar_t *u, Py_ssize_t size)
     if (size == 0)
         _Py_RETURN_UNICODE_EMPTY();
 
+#ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+    /* Oracle Solaris uses non-Unicode internal wchar_t form for
+       non-Unicode locales and hence needs conversion to UCS-4 first. */
+    if (_Py_LocaleUsesNonUnicodeWchar()) {
+        wchar_t* converted = _Py_DecodeNonUnicodeWchar(u, size);
+        if (!converted) {
+            return NULL;
+        }
+        PyObject *unicode = _PyUnicode_FromUCS4(converted, size);
+        PyMem_Free(converted);
+        return unicode;
+    }
+#endif
+
     /* Single character Unicode objects in the Latin-1 range are
        shared when using this constructor */
     if (size == 1 && (Py_UCS4)*u < 256)
@@ -3295,6 +3313,17 @@ PyUnicode_AsWideChar(PyObject *unicode,
         res = size;
     }
     unicode_copy_as_widechar(unicode, w, size);
+
+#if HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+    /* Oracle Solaris uses non-Unicode internal wchar_t form for
+       non-Unicode locales and hence needs conversion first. */
+    if (_Py_LocaleUsesNonUnicodeWchar()) {
+        if (_Py_EncodeNonUnicodeWchar_InPlace(w, size) < 0) {
+            return -1;
+        }
+    }
+#endif
+
     return res;
 }
 
@@ -3321,6 +3350,17 @@ PyUnicode_AsWideCharString(PyObject *unicode,
         return NULL;
     }
     unicode_copy_as_widechar(unicode, buffer, buflen + 1);
+
+#if HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+    /* Oracle Solaris uses non-Unicode internal wchar_t form for
+       non-Unicode locales and hence needs conversion first. */
+    if (_Py_LocaleUsesNonUnicodeWchar()) {
+        if (_Py_EncodeNonUnicodeWchar_InPlace(buffer, (buflen + 1)) < 0) {
+            return NULL;
+        }
+    }
+#endif
+
     if (size != NULL) {
         *size = buflen;
     }

Diff for: Python/fileutils.c

@@ -18,6 +18,10 @@ extern int winerror_to_errno(int);
 #include <sys/ioctl.h>
 #endif
 
+#ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+#include <iconv.h>
+#endif
+
 #ifdef HAVE_FCNTL_H
 #include <fcntl.h>
 #endif /* HAVE_FCNTL_H */
@@ -93,6 +97,12 @@ _Py_device_encoding(int fd)
 static size_t
 is_valid_wide_char(wchar_t ch)
 {
+#ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+    /* Oracle Solaris doesn't use Unicode code points as wchar_t encoding
+       for non-Unicode locales, which makes values higher than MAX_UNICODE
+       possibly valid. */
+    return 1;
+#endif
     if (Py_UNICODE_IS_SURROGATE(ch)) {
         // Reject lone surrogate characters
         return 0;
@@ -922,6 +932,102 @@ _Py_GetLocaleEncodingObject(void)
     return str;
 }
 
+#ifdef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+
+/* Check whether current locale uses Unicode as internal wchar_t form. */
+int
+_Py_LocaleUsesNonUnicodeWchar(void)
+{
+    /* Oracle Solaris uses non-Unicode internal wchar_t form for
+       non-Unicode locales and hence needs conversion to UTF first. */
+    char* codeset = nl_langinfo(CODESET);
+    if (!codeset) {
+        return 0;
+    }
+    /* 646 refers to ISO/IEC 646 standard that corresponds to ASCII encoding */
+    return (strcmp(codeset, "UTF-8") != 0 && strcmp(codeset, "646") != 0);
+}
+
+static wchar_t *
+_Py_ConvertWCharForm(const wchar_t *source, Py_ssize_t size,
+                     const char *tocode, const char *fromcode)
+{
+    Py_BUILD_ASSERT(sizeof(wchar_t) == 4);
+
+    /* Ensure we won't overflow the size. */
+    if (size > (PY_SSIZE_T_MAX / (Py_ssize_t)sizeof(wchar_t))) {
+        PyErr_NoMemory();
+        return NULL;
+    }
+
+    /* the string doesn't have to be NULL terminated */
+    wchar_t* target = PyMem_Malloc(size * sizeof(wchar_t));
+    if (target == NULL) {
+        PyErr_NoMemory();
+        return NULL;
+    }
+
+    iconv_t cd = iconv_open(tocode, fromcode);
+    if (cd == (iconv_t)-1) {
+        PyErr_Format(PyExc_ValueError, "iconv_open() failed");
+        PyMem_Free(target);
+        return NULL;
+    }
+
+    char *inbuf = (char *) source;
+    char *outbuf = (char *) target;
+    size_t inbytesleft = sizeof(wchar_t) * size;
+    size_t outbytesleft = inbytesleft;
+
+    size_t ret = iconv(cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft);
+    if (ret == DECODE_ERROR) {
+        PyErr_Format(PyExc_ValueError, "iconv() failed");
+        PyMem_Free(target);
+        iconv_close(cd);
+        return NULL;
+    }
+
+    iconv_close(cd);
+    return target;
+}
+
+/* Convert a wide character string to the UCS-4 encoded string. This
+   is necessary on systems where internal form of wchar_t are not Unicode
+   code points (e.g. Oracle Solaris).
+
+   Return a pointer to a newly allocated string, use PyMem_Free() to free
+   the memory. Return NULL and raise exception on conversion or memory
+   allocation error. */
+wchar_t *
+_Py_DecodeNonUnicodeWchar(const wchar_t *native, Py_ssize_t size)
+{
+    return _Py_ConvertWCharForm(native, size, "UCS-4-INTERNAL", "wchar_t");
+}
+
+/* Convert a UCS-4 encoded string to native wide character string. This
+   is necessary on systems where internal form of wchar_t are not Unicode
+   code points (e.g. Oracle Solaris).
+
+   The conversion is done in place. This can be done because both wchar_t
+   and UCS-4 use 4-byte encoding, and one wchar_t symbol always correspond
+   to a single UCS-4 symbol and vice versa. (This is true for Oracle Solaris,
+   which is currently the only system using these functions; it doesn't have
+   to be for other systems).
+
+   Return 0 on success. Return -1 and raise exception on conversion
+   or memory allocation error. */
+int
+_Py_EncodeNonUnicodeWchar_InPlace(wchar_t *unicode, Py_ssize_t size)
+{
+    wchar_t* result = _Py_ConvertWCharForm(unicode, size, "wchar_t", "UCS-4-INTERNAL");
+    if (!result) {
+        return -1;
+    }
+    memcpy(unicode, result, size * sizeof(wchar_t));
+    PyMem_Free(result);
+    return 0;
+}
+#endif /* HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION */
 
 #ifdef MS_WINDOWS
 static __int64 secs_between_epochs = 11644473600; /* Seconds between 1.1.1601 and 1.1.1970 */

Diff for: configure

@@ -15264,6 +15264,22 @@ else
 $as_echo "no" >&6; }
 fi
 
+case $ac_sys_system/$ac_sys_release in
+SunOS/*)
+  if test -f /etc/os-release; then
+    OS_NAME=$(awk -F= '/^NAME=/ {print substr($2,2,length($2)-2)}' /etc/os-release)
+    if test "x$OS_NAME" = "xOracle Solaris"; then
+      # bpo-43667: In Oracle Solaris, the internal form of wchar_t in
+      # non-Unicode locales is not Unicode and hence cannot be used directly.
+      # https://docs.oracle.com/cd/E37838_01/html/E61053/gmwke.html
+
+$as_echo "#define HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION 1" >>confdefs.h
+
+    fi
+  fi
+  ;;
+esac
+
 # check for endianness
  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether byte ordering is bigendian" >&5
 $as_echo_n "checking whether byte ordering is bigendian... " >&6; }

Diff for: configure.ac

@@ -4765,6 +4765,22 @@ else
   AC_MSG_RESULT(no)
 fi
 
+case $ac_sys_system/$ac_sys_release in
+SunOS/*)
+  if test -f /etc/os-release; then
+    OS_NAME=$(awk -F= '/^NAME=/ {print substr($2,2,length($2)-2)}' /etc/os-release)
+    if test "x$OS_NAME" = "xOracle Solaris"; then
+      # bpo-43667: In Oracle Solaris, the internal form of wchar_t in
+      # non-Unicode locales is not Unicode and hence cannot be used directly.
+      # https://docs.oracle.com/cd/E37838_01/html/E61053/gmwke.html
+      AC_DEFINE(HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION, 1,
+      [Define if the internal form of wchar_t in non-Unicode locales
+       is not Unicode.])
+    fi
+  fi
+  ;;
+esac
+
 # check for endianness
 AC_C_BIGENDIAN
 

Diff for: pyconfig.h.in

@@ -748,6 +748,10 @@
 /* Define to 1 if you have the `nice' function. */
 #undef HAVE_NICE
 
+/* Define if the internal form of wchar_t in non-Unicode locales is not
+   Unicode. */
+#undef HAVE_NON_UNICODE_WCHAR_T_REPRESENTATION
+
 /* Define to 1 if you have the `openat' function. */
 #undef HAVE_OPENAT