Add support for 4-byte UCNs like \U12345678. Warn about UCNs in c90 mode.

llvm-svn: 115743

clang/include/clang/Basic/DiagnosticLexKinds.td

@@ -98,6 +98,10 @@ def warn_hex_escape_too_large : ExtWarn<"hex escape sequence out of range">;
 def ext_string_too_long : Extension<"string literal of length %0 exceeds "
   "maximum length %1 that %select{C90|ISO C99|C++}2 compilers are required to "
   "support">, InGroup<OverlengthStrings>;
+def warn_ucn_escape_too_large : ExtWarn<
+  "character unicode escape sequence too long for its type">;
+def warn_ucn_not_valid_in_c89 : ExtWarn<
+  "unicode escape sequences are only valid in C99 or C++">;
   
 //===----------------------------------------------------------------------===//
 // PTH Diagnostics

clang/lib/Lex/LiteralSupport.cpp

@@ -172,8 +172,8 @@ static void ProcessUCNEscape(const char *&ThisTokBuf, const char *ThisTokEnd,
                              SourceLocation Loc, Preprocessor &PP,
                              bool wide,
                              bool Complain) {
-  // FIXME: Add a warning - UCN's are only valid in C++ & C99.
-  // FIXME: Handle wide strings.
+  if (!PP.getLangOptions().CPlusPlus && !PP.getLangOptions().C99)
+    PP.Diag(Loc, diag::warn_ucn_not_valid_in_c89);
 
   // Save the beginning of the string (for error diagnostics).
   const char *ThisTokBegin = ThisTokBuf;
@@ -218,13 +218,34 @@ static void ProcessUCNEscape(const char *&ThisTokBuf, const char *ThisTokEnd,
   }
   if (wide) {
     (void)UcnLenSave;
-    assert(UcnLenSave == 4 && 
-           "ProcessUCNEscape - only ucn length of 4 supported");
-    // little endian assumed.
-    *ResultBuf++ = (UcnVal & 0x000000FF);
-    *ResultBuf++ = (UcnVal & 0x0000FF00) >> 8;
-    *ResultBuf++ = (UcnVal & 0x00FF0000) >> 16;
-    *ResultBuf++ = (UcnVal & 0xFF000000) >> 24;
+    assert((UcnLenSave == 4 || UcnLenSave == 8) && 
+           "ProcessUCNEscape - only ucn length of 4 or 8 supported");
+
+    if (!PP.getLangOptions().ShortWChar) {
+      // Note: our internal rep of wide char tokens is always little-endian.
+      *ResultBuf++ = (UcnVal & 0x000000FF);
+      *ResultBuf++ = (UcnVal & 0x0000FF00) >> 8;
+      *ResultBuf++ = (UcnVal & 0x00FF0000) >> 16;
+      *ResultBuf++ = (UcnVal & 0xFF000000) >> 24;
+      return;
+    }
+
+    // Convert to UTF16.
+    if (UcnVal < (UTF32)0xFFFF) {
+      *ResultBuf++ = (UcnVal & 0x000000FF);
+      *ResultBuf++ = (UcnVal & 0x0000FF00) >> 8;
+      return;
+    }
+    PP.Diag(Loc, diag::warn_ucn_escape_too_large);
+
+    typedef uint16_t UTF16;
+    UcnVal -= 0x10000;
+    UTF16 surrogate1 = 0xD800 + (UcnVal >> 10);
+    UTF16 surrogate2 = 0xDC00 + (UcnVal & 0x3FF);
+    *ResultBuf++ = (surrogate1 & 0x000000FF);
+    *ResultBuf++ = (surrogate1 & 0x0000FF00) >> 8;
+    *ResultBuf++ = (surrogate2 & 0x000000FF);
+    *ResultBuf++ = (surrogate2 & 0x0000FF00) >> 8;
     return;
   }
   // Now that we've parsed/checked the UCN, we convert from UTF32->UTF8.

clang/test/CodeGen/string-literal-short-wstring.c

@@ -0,0 +1,14 @@
+// RUN: %clang_cc1 -emit-llvm -fshort-wchar %s -o - | FileCheck %s
+
+int main() {
+  // This should convert to utf8.
+  // CHECK: internal constant [10 x i8] c"\E1\84\A0\C8\A0\F4\82\80\B0\00", align 1
+  char b[10] = "\u1120\u0220\U00102030";
+
+  // CHECK: private constant [6 x i8] c"A\00B\00\00\00"
+  void *foo = L"AB";
+
+  // This should convert to utf16.
+  // CHECK: private constant [10 x i8] c" \11 \02\C8\DB0\DC\00\00"
+  void *bar = L"\u1120\u0220\U00102030";
+}

clang/test/CodeGen/string-literal.c

@@ -1,7 +1,16 @@
-// RUN: %clang_cc1 -emit-llvm %s -o -
+// RUN: %clang_cc1 -emit-llvm %s -o - | FileCheck %s
 
 int main() {
+  // CHECK: internal constant [10 x i8] c"abc\00\00\00\00\00\00\00", align 1
   char a[10] = "abc";
 
+  // This should convert to utf8.
+  // CHECK: internal constant [10 x i8] c"\E1\84\A0\C8\A0\F4\82\80\B0\00", align 1
+  char b[10] = "\u1120\u0220\U00102030";
+
+  // CHECK: private constant [12 x i8] c"A\00\00\00B\00\00\00\00\00\00\00"
   void *foo = L"AB";
+
+  // CHECK: private constant [12 x i8] c"4\12\00\00\0B\F0\10\00\00\00\00\00"
+  void *bar = L"\u1234\U0010F00B";
 }

clang/test/Lexer/c90.c

@@ -27,3 +27,7 @@ void test2() {
     "sdjflksdjf lksdjf skldfjsdkljflksdjf kldsjflkdsj fldks jflsdkjfds"
     "sdjflksdjf lksdjf skldfjsdkljflksdjf kldsjflkdsj fldks jflsdkjfds";
 }
+
+void test3() {
+  (void)L"\u1234";  // expected-error {{unicode escape sequences are only valid in C99 or C++}}
+}

clang/test/Lexer/wchar.c

@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -fsyntax-only -fshort-wchar -verify %s
+
+void f() {
+  (void)L"\U00010000";  // expected-warning {{character unicode escape sequence too long for its type}}
+}
+