[clang-format] Improve detection of ObjC for-in statements

[benhamilton]

Summary

Previously, clang-format would detect the following as an
Objective-C for-in statement:

for (int x = in.value(); ...) {}

because the logic only decided a for-loop was definitely *not*
an Objective-C for-in loop after it saw a semicolon or a colon.

To fix this, I delayed the decision of whether this was a for-in
statement until after we found the matching right-paren, at which
point we know if we've seen a semicolon or not.

Depends On D43904

Test Plan: New tests added. Ran tests with:

make -j12 FormatTests && ./tools/clang/unittests/Format/FormatTests

[benhamilton]

Whoops, sent out twice.

lib/Format/TokenAnnotator.cpp

while (CurrentToken) {
 
         if (!HasMultipleLines)
           Left->PackingKind = PPK_Inconclusive;
         else if (HasMultipleParametersOnALine)
           Left->PackingKind = PPK_BinPacked;
         else
           Left->PackingKind = PPK_OnePerLine;
 
+        if (MightBeObjCForRangeLoop) {
+          FormatToken *ForInToken = Left;
+          while (ForInToken && ForInToken != CurrentToken) {
+            if (ForInToken->is(Keywords.kw_in)) {
+              ForInToken->Type = TT_ObjCForIn;
+              break;
+            }
+            ForInToken = ForInToken->Next;
+          }
+        }
         next();
         return true;
       }
       if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
         return false;
 
       if (CurrentToken->is(tok::l_brace))
         Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen
       if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
           !CurrentToken->Next->HasUnescapedNewline &&
           !CurrentToken->Next->isTrailingComment())
         HasMultipleParametersOnALine = true;
       if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) ||
            CurrentToken->Previous->isSimpleTypeSpecifier()) &&
           !CurrentToken->is(tok::l_brace))
         Contexts.back().IsExpression = false;
       if (CurrentToken->isOneOf(tok::semi, tok::colon))
         MightBeObjCForRangeLoop = false;
-      if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in))
-        CurrentToken->Type = TT_ObjCForIn;
       // When we discover a 'new', we set CanBeExpression to 'false' in order to
       // parse the type correctly. Reset that after a comma.
       if (CurrentToken->is(tok::comma))
         Contexts.back().CanBeExpression = true;
 
       FormatToken *Tok = CurrentToken;
       if (!consumeToken())
         return false;

unittests/Format/FormatTest.cpp

EXPECT_EQ(FormatStyle::LK_Cpp,
             guessLanguage("foo.h", "[[using gsl: suppress(\"type\")]];"));
   EXPECT_EQ(
       FormatStyle::LK_Cpp,
       guessLanguage("foo.h",
                     "[[clang::callable_when(\"unconsumed\", \"unknown\")]]"));
   EXPECT_EQ(FormatStyle::LK_Cpp, guessLanguage("foo.h", "[[foo::bar, ...]]"));
 }
 
+TEST_F(FormatTest, GuessLanguageWithForIn) {
+  EXPECT_EQ(FormatStyle::LK_Cpp,
+            guessLanguage("foo.h", "for (Foo *x = 0; x != in; x++) {}"));
+  EXPECT_EQ(FormatStyle::LK_ObjC,
+            guessLanguage("foo.h", "for (Foo *x in bar) {}"));
+  EXPECT_EQ(FormatStyle::LK_ObjC,
+            guessLanguage("foo.h", "for (Foo *x in [bar baz]) {}"));
+  EXPECT_EQ(FormatStyle::LK_ObjC,
+            guessLanguage("foo.h", "for (Foo *x in [bar baz:blech]) {}"));
+  EXPECT_EQ(
+      FormatStyle::LK_ObjC,
+      guessLanguage("foo.h", "for (Foo *x in [bar baz:blech, 1, 2, 3, 0]) {}"));
+  EXPECT_EQ(FormatStyle::LK_ObjC,
+            guessLanguage("foo.h", "for (Foo *x in [bar baz:^{[uh oh];}]) {}"));
+  EXPECT_EQ(FormatStyle::LK_Cpp,
+            guessLanguage("foo.h", "Foo *x; for (x = 0; x != in; x++) {}"));
+  EXPECT_EQ(FormatStyle::LK_ObjC,
+            guessLanguage("foo.h", "Foo *x; for (x in y) {}"));
+  EXPECT_EQ(
+      FormatStyle::LK_Cpp,
+      guessLanguage(
+          "foo.h",
+          "for (const Foo<Bar>& baz = in.value(); !baz.at_end(); ++baz) {}"));
+}
+
 } // end namespace
 } // end namespace format
 } // end namespace clang