3 files changed, 15 insertions, 7 deletions
diff --git a/clang/lib/Format/Format.cpp b/clang/lib/Format/Format.cpp
index effe0bc2482..4d401a5be9e 100644
--- a/clang/lib/Format/Format.cpp
+++ b/clang/lib/Format/Format.cpp
@@ -483,8 +483,7 @@ private:
 
     unsigned Newlines =
         std::min(Token.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
-    unsigned Offset = SourceMgr.getFileOffset(Token.WhiteSpaceStart);
-    if (Newlines == 0 && Offset != 0)
+    if (Newlines == 0 && !Token.IsFirst)
       Newlines = 1;
     unsigned Indent = Line.Level * 2;
     if ((Token.Tok.is(tok::kw_public) || Token.Tok.is(tok::kw_protected) ||
@@ -685,9 +684,10 @@ public:
       next();
       if (Index >= Tokens.size())
         return;
-      // It is the responsibility of the UnwrappedLineParser to make sure
-      // this sequence is not produced inside an unwrapped line.
-      assert(Tokens[Index].Tok.getIdentifierInfo() != NULL);
+      // Hashes in the middle of a line can lead to any strange token
+      // sequence.
+      if (Tokens[Index].Tok.getIdentifierInfo() == NULL)
+        return;
       switch (Tokens[Index].Tok.getIdentifierInfo()->getPPKeywordID()) {
       case tok::pp_include:
       case tok::pp_import:
@@ -1033,6 +1033,8 @@ public:
     Lex.LexFromRawLexer(FormatTok.Tok);
     StringRef Text = tokenText(FormatTok.Tok);
     FormatTok.WhiteSpaceStart = FormatTok.Tok.getLocation();
+    if (SourceMgr.getFileOffset(FormatTok.WhiteSpaceStart) == 0)
+      FormatTok.IsFirst = true;
 
     // Consume and record whitespace until we find a significant token.
     while (FormatTok.Tok.is(tok::unknown)) {
diff --git a/clang/lib/Format/UnwrappedLineParser.cpp b/clang/lib/Format/UnwrappedLineParser.cpp
index 2f3a6034bcf..72b47503ee8 100644
--- a/clang/lib/Format/UnwrappedLineParser.cpp
+++ b/clang/lib/Format/UnwrappedLineParser.cpp
@@ -470,7 +470,9 @@ void UnwrappedLineParser::nextToken() {
 
 void UnwrappedLineParser::readToken() {
   FormatTok = Tokens->getNextToken();
-  while (!Line.InPPDirective && FormatTok.Tok.is(tok::hash)) {
+  while (!Line.InPPDirective && FormatTok.Tok.is(tok::hash) &&
+         ((FormatTok.NewlinesBefore > 0 && FormatTok.HasUnescapedNewline) ||
+          FormatTok.IsFirst)) {
     // FIXME: This is incorrect - the correct way is to create a
     // data structure that will construct the parts around the preprocessor
     // directive as a structured \c UnwrappedLine.
diff --git a/clang/lib/Format/UnwrappedLineParser.h b/clang/lib/Format/UnwrappedLineParser.h
index a8e5b739bc8..287143dae29 100644
--- a/clang/lib/Format/UnwrappedLineParser.h
+++ b/clang/lib/Format/UnwrappedLineParser.h
@@ -31,7 +31,8 @@ namespace format {
 /// whitespace characters preceeding it.
 struct FormatToken {
   FormatToken()
-      : NewlinesBefore(0), HasUnescapedNewline(false), WhiteSpaceLength(0) {
+      : NewlinesBefore(0), HasUnescapedNewline(false), WhiteSpaceLength(0),
+        IsFirst(false) {
   }
 
   /// \brief The \c Token.
@@ -56,6 +57,9 @@ struct FormatToken {
   /// \brief The length in characters of the whitespace immediately preceeding
   /// the \c Token.
   unsigned WhiteSpaceLength;
+
+  /// \brief Indicates that this is the first token.
+  bool IsFirst;
 };
 
 /// \brief An unwrapped line is a sequence of \c Token, that we would like to