Most vexing parse

From HandWiki
Short description: Syntactic ambiguity in C++

The most vexing parse is a counterintuitive form of syntactic ambiguity resolution in the C++ programming language. In certain situations, the C++ grammar cannot distinguish between the creation of an object parameter and specification of a function's type. In those situations, the compiler is required to interpret the line as a function type specification.

Occurrence

The term "most vexing parse" was first used by Scott Meyers in his 2001 book Effective STL.[1] While unusual in C, the phenomenon was quite common in C++ until the introduction of uniform initialization in C++11.[2]

Examples

C-style casts

A simple example appears when a functional cast is intended to convert an expression for initializing a variable:

void f(double my_dbl) {
  int i(int(my_dbl));
}

Line 2 above is ambiguous. One possible interpretation is to declare a variable i with initial value produced by converting my_dbl to an int. However, C allows superfluous parentheses around function parameter declarations; in this case, the declaration of i is instead a function declaration equivalent to the following:

// A function named i takes an integer and returns an integer.
int i(int my_dbl);

Unnamed temporary

A more elaborate example is:

struct Timer {};

struct TimeKeeper {
  explicit TimeKeeper(Timer t);
  int get_time();
};

int main() {
  TimeKeeper time_keeper(Timer());
  return time_keeper.get_time();
}

The line

TimeKeeper time_keeper(Timer());

is ambiguous, since it could be interpreted either as

  1. a variable definition for variable time_keeper of class TimeKeeper, initialized with an anonymous instance of class Timer or
  2. a function declaration for a function time_keeper that returns an object of type TimeKeeper and has a single (unnamed) parameter, whose type is a (pointer to a) function[Note 1] taking no input and returning Timer objects.

The C++ standard requires the second interpretation, which is inconsistent with line 9 above. For example, Clang++ warns that the most vexing parse has been applied on line 9 and errors on the following line:[3]

$ clang++ time_keeper.cc
timekeeper.cc:9:25: warning: parentheses were disambiguated as a function declaration
      [-Wvexing-parse]
  TimeKeeper time_keeper(Timer());
                        ^~~~~~~~~
timekeeper.cc:9:26: note: add a pair of parentheses to declare a variable
  TimeKeeper time_keeper(Timer());
                         ^
                         (      )
timekeeper.cc:10:21: error: member reference base type 'TimeKeeper (Timer (*)())' is not a
      structure or union
  return time_keeper.get_time();
         ~~~~~~~~~~~^~~~~~~~~

Solutions

The required interpretation of these ambiguous declarations is rarely the intended one.[4][5] Function types in C++ are usually hidden behind typedefs and typically have an explicit reference or pointer qualifier. To force the alternate interpretation, the typical technique is a different object creation or conversion syntax.

In the type conversion example, there are two alternate syntaxes available for casts: the "C-style cast"

// declares a variable of type int
int i((int)my_dbl);

or a named cast:

int i(static_cast<int>(my_dbl));

In the variable declaration example, the preferred method (since C++11) is uniform (brace) initialization.[6] This also allows limited omission of the type name entirely:

//Any of the following work:

TimeKeeper time_keeper(Timer{}); TimeKeeper time_keeper{Timer()}; TimeKeeper time_keeper{Timer{}}; TimeKeeper time_keeper( {});

TimeKeeper time_keeper{ {}};

Prior to C++11, the common techniques to force the intended interpretation were use of an extra parenthesis or copy-initialization:[5]

TimeKeeper time_keeper( /*Avoid MVP*/ (Timer()) );
TimeKeeper time_keeper = TimeKeeper(Timer());

In the latter syntax, the copy-initialization is likely to be optimized out by the compiler.[7] Since C++17, this optimization is guaranteed.[8]

Notes

  1. According to C++ type decay rules, a function object declared as a parameter is equivalent to a pointer to a function of that type. See Function object.

References

  1. Meyers, Scott (2001). Effective STL: 50 Specific Ways to Improve Your Use of the Standard Template Library. Addison-Wesley. ISBN 0-201-74962-9. 
  2. Coffin, Jerry (29 December 2012). "c++ - What is the purpose of the Most Vexing Parse?". https://stackoverflow.com/questions/14077608/what-is-the-purpose-of-the-most-vexing-parse. 
  3. Lattner, Chris (5 April 2010). "Amazing Feats of Clang Error Recovery". The Most Vexing Parse. https://blog.llvm.org/posts/2010-04-05-amazing-feats-of-clang-error-recovery/. 
  4. DrPizza; Prototyped; wb; euzeka; Simpson, Homer J (October 2002). "C++'s "most vexing parse"". https://arstechnica.com/civis/viewtopic.php?f=20&t=767929. 
  5. 5.0 5.1 Boccara, Jonathan (2018-01-30). "The Most Vexing Parse: How to Spot It and Fix It Quickly" (in en-US). https://www.fluentcpp.com/2018/01/30/most-vexing-parse/. 
  6. Stroustrup, Bjarne (19 August 2016). "C++11 FAQ" (in en). Uniform initialization syntax and semantics. https://www.stroustrup.com/C++11FAQ.html#uniform-init. 
  7. "Myths and urban legends about C++". What is copy elision? What is RVO?. https://isocpp.org/wiki/faq/myths#copy-elision. 
  8. Devlieghere, Jonas (2016-11-21). "Guaranteed Copy Elision" (in en). https://jonasdevlieghere.com/guaranteed-copy-elision/.  Note, however, the caveats covered in Brand, C++ (2018-12-11). "Guaranteed Copy Elision Does Not Elide Copies" (in en-US). https://devblogs.microsoft.com/cppblog/guaranteed-copy-elision-does-not-elide-copies/. 

External links