[clang] [Clang] Warn on backslash-newline-EOF (PR #97585)

[Mital Ashok via cfe-commits]

================
@@ -3165,7 +3165,17 @@ bool Lexer::LexEndOfFile(Token &Result, const char *CurPtr) {
 
   // C99 5.1.1.2p2: If the file is non-empty and didn't end in a newline, issue
   // a pedwarn.
-  if (CurPtr != BufferStart && (CurPtr[-1] != '\n' && CurPtr[-1] != '\r')) {
+  if (CurPtr != BufferStart) {
+    StringRef LastNewline;
+    if (CurPtr[-1] == '\r' || CurPtr[-1] == '\n') {
+      LastNewline = StringRef(CurPtr - 1, 1);
+      if (CurPtr - 1 != BufferStart && CurPtr[-2] != CurPtr[-1] &&
+          (CurPtr[-2] == '\r' || CurPtr[-2] == '\n')) {
+        // \r\n or \n\r is one newline
----------------
MitalAshok wrote:

https://github.com/llvm/llvm-project/blob/e46468407a7bb7f8b2fe13675a5a1c32b85f8cad/clang/lib/Lex/Lexer.cpp#L1288

https://github.com/llvm/llvm-project/blob/e46468407a7bb7f8b2fe13675a5a1c32b85f8cad/clang/lib/Lex/Lexer.cpp#L2775

And with the current clang:

```c++
int main() { // this line ends with \n\r\
  return 1;
}
```

(That is, the file is generated with `python3 -c 'open("test.cpp", "wb").write(b"int main() { // this line ends with \\n\\r\\\n\r  return 1;\n}\n")'`)

Exits 0. It seems like Clang has always done this, gcc exits with code 1 here.

The reason given is efficiency:

https://github.com/llvm/llvm-project/blob/e46468407a7bb7f8b2fe13675a5a1c32b85f8cad/clang/lib/Lex/Lexer.cpp#L2711-L2715

Maybe those other places are a mistake since it does affect "correctness", and it should treat \n\r as a Unix style newline followed by an old Mac style newline, but it is currently treated as a single newline so thats what this patch treats it as.

https://github.com/llvm/llvm-project/pull/97585