clang-format: Pointer `*` is seen as multiplication in C-style casts

[mxbOctasic]

Summary

The * was treated as multiplication operator in complex pointer type casts.

e.g.
(type *const)bar
(type *restrict)bar

Patch by cameron314

[djasper]

lib/Format/TokenAnnotator.cpp

1272

I think it would be better to not have such cast detection heuristics in several places. Could we change the type of the "*" to TT_PointerOrReference after the fact, when rParenEndsCast returns true?

[mxbOctasic]

Moved cast detection logic to rParenEndsCast.

lib/Format/TokenAnnotator.cpp

bool ParensAreType =
         !Tok.Previous ||
         Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) ||
         Tok.Previous->isSimpleTypeSpecifier();
     bool ParensCouldEndDecl =
         Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
     if (ParensAreType && !ParensCouldEndDecl)
       return true;
 
+    // Some more obscure casts can leak through the above heuristic, like
+    // (type* const)bar or (type* MACRO)bar
+    if (Tok.Next->is(tok::identifier) && Tok.Previous &&
+        Tok.Previous->isOneOf(tok::kw_const, tok::identifier) &&
+        Tok.Previous->Previous &&
+        Tok.Previous->Previous->is(TT_PointerOrReference))
+      return true;
+
     // At this point, we heuristically assume that there are no casts at the
     // start of the line. We assume that we have found most cases where there
     // are by the logic above, e.g. "(void)x;".
     if (!LeftOfParens)
       return false;
 
     // If the following token is an identifier or 'this', this is a cast. All
     // cases where this can be something else are handled above.
if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
       return TT_BinaryOperator;
 
     // This catches some cases where evaluation order is used as control flow:
     //   aaa && aaa->f();
     const FormatToken *NextNextToken = NextToken->getNextNonComment();
     if (NextNextToken && NextNextToken->is(tok::arrow))
       return TT_BinaryOperator;
 
+    // Casts are never binary operators, regardless of IsExpression
+    if (NextToken->isOneOf(tok::r_paren, tok::greater))
+      return TT_PointerOrReference;
+    if (NextToken->isOneOf(tok::identifier, tok::kw_const) && NextNextToken &&
+        NextNextToken->is(tok::r_paren) && NextNextToken->Next &&
+        NextNextToken->Next->isOneOf(tok::identifier, tok::l_paren) &&
+        (!PrevToken->Previous || !PrevToken->Previous->is(tok::l_paren) ||
+         !PrevToken->Previous->Previous ||
+         !PrevToken->Previous->Previous->is(tok::kw_decltype)))
+      return TT_PointerOrReference;
+
     // It is very unlikely that we are going to find a pointer or reference type
     // definition on the RHS of an assignment.
     if (IsExpression && !Contexts.back().CaretFound)
       return TT_BinaryOperator;
 
     return TT_PointerOrReference;
   }
 

unittests/Format/FormatTest.cpp

TEST_F(FormatTest, UnderstandsUsesOfStarAndAmp) {
 
   // Member function reference qualifiers aren't binary operators.
   verifyFormat("string // break\n"
                "operator()() & {}");
   verifyFormat("string // break\n"
                "operator()() && {}");
   verifyGoogleFormat("template <typename T>\n"
                      "auto x() & -> int {}");
+
+  // Some casts look like they might be multiplications, but they are definitely
+  // not!
+  // Here, foo could be a macro containing e.g. `const`
+  verifyFormat("(type *foo)bar");
+  verifyFormat("(type *const)bar");
+  verifyFormat("type *x = (type *foo)bar");
+  verifyFormat("type *const x = (type *const)bar");
+  verifyFormat("type *x = (type *foo)(bar)");
 }
 
 TEST_F(FormatTest, UnderstandsAttributes) {
   verifyFormat("SomeType s __attribute__((unused)) (InitValue);");
   verifyFormat("aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa __attribute__((unused))\n"
                "aaaaaaaaaaaaaaaaaaaaaaa(int i);");
   FormatStyle AfterType = getLLVMStyle();
   AfterType.AlwaysBreakAfterReturnType = FormatStyle::RTBS_AllDefinitions;