Control Hooks

July 4, 2016 · View on GitHub

Beyond the top-level grammar rule, which has to be supplied to a parsing run, and an action class template, which can be supplied to a parsing run, a third customisation point within the PEGTL allows the user to provide a control class (template) to a parsing run.

The control class must implement a set of static hook functions that are called by the PEGTL in order to

customise which exceptions are thrown in case of errors, and
add custom debug logging to follow the details of a parsing run.

Normal Control
Debug Functions
Exception Throwing
Debugging and Tracing
Advanced Control
Changing Control

Normal Control

The normal control class template included with the PEGTL is used by default and shows which hook functions there are.

template< typename Rule >
struct normal
{
   template< typename Input, typename ... States >
   static void start( const Input &, States && ... )
   { }

   template< typename Input, typename ... States >
   static void success( const Input &, States && ... )
   { }

   template< typename Input, typename ... States >
   static void failure( const Input &, States && ... )
   { }

   template< typename Input, typename ... States >
   static void raise( const Input & in, States && ... )
   {
      throw parse_error( "parse error matching " +
                         internal::demangle< Rule >(), in );
   }

   template< apply_mode A,
             template< typename ... > class Action,
             template< typename ... > class Control,
             typename Input,
             typename ... States >
   static bool match( Input & in, States && ... st )
   {
      return internal::rule_match_one< Rule,
                                       A,
                                       Action,
                                       Control >( in, st ... );
   }
};

All control functions are called with the current input and all additional state arguments.

Debug Functions

Before attempting to match a rule R, the PEGTL calls C< R >::start() where C is the current control class template.

Depending on what happens during the attempt to match R, one of the other three functions might be called.

If R succeeds, then C< R >::success() is called; compared to the call to C< R >::start(), the input will have consumed whatever the successful match of R consumed.
If R finishes with a failure, i.e. a return value of false from its match()-function, then C< R >::failure() is called; a failed match is not allowed to consume input.
If R is wrapped in must< R >, a global failure is generated by calling C< R >::raise() to throw some exception as is expected by the PEGTL in this case.
If a sub-rule of R finishes with a global failure, and the exception is not caught by a try_catch or similar combinator, then no other function of C< R > is called after C< R >::start().

Exception Throwing

The raise()-control-hook-function must throw an exception. For most parts of the PEGTL the exception class is irrelevant and any user-defined data type can be thrown by a user-defined control hook.

The try_catch rule only catches exceptions of type pegtl::parse_error!

When custom exception types are used then try_catch_type must be used with the custom exception class that they are supposed to catch as first template argument.

Debugging and Tracing

For debugging a grammar and tracing exactly what happens during a parsing run, the control class methods start(), success() and failure() can be used. A minimal example of how this can be done is included with the PEGTL in the pegtl/trace.hh header file.

template< typename Rule >
struct tracer
      : normal< Rule >
{
   template< typename Input, typename ... States >
   static void start( const Input & in, States && ... )
   {
      std::cerr << position_info( in )
                << "  start  "
                << internal::demangle< Rule >()
                << std::endl;
   }

   template< typename Input, typename ... States >
   static void success( const Input & in, States && ... )
   {
      std::cerr << position_info( in )
                << " success "
                << internal::demangle< Rule >()
                << std::endl;
   }

   template< typename Input, typename ... States >
   static void failure( const Input & in, States && ... st )
   {
      std::cerr << position_info( in )
                << " failure "
                << internal::demangle< Rule >()
                << std::endl;
   }
};

Advanced Control

The control template's match()-function is the first or outside-most function that is called in the flow that eventually leads to calling a rule's match()-function.

The normal implementation simply forwards all template parameters and function arguments unchanged.

For advanced use cases it is however possible to create a custom control class with a custom match()-function that can change "everything" before forwarding as needed.

Changing Control

Just like the action class template, a custom control class template can be used by either

supplying it as explicit template argument to the parse()-functions, or
setting it for a portion of the grammar with the pegtl::control combinator.