[ragel-users] for 6.0: going back to 3.X semantics for >, %, err

[Erich Ocean]

That's great! +1 for the change.

On Oct 9, 2007, at 3:48 PM, Adrian Thurston wrote:

> Hi,
>
> In version 3 of Ragel the semantics of >, % and the error actions were
> different.
>
> 1. The leaving operator '%' would also embed the action as an EOF  
> action
> if the final states of the machine it was applied to remained final in
> the compiled machine. This meant that as long as the machine didn't
> error, the leaving action would always be executed even if it came  
> right
> at the end of the input.
>
> 2. The starting transition operator '>' would embed a leaving action
> into the start state if the start state was final. This meant that  
> if a
> machine accepted the zero-length string and control passed through the
> zero-length string then the starting transition action would still get
> executed.
>
> 3. All global and local error actions would be executed when the input
> ended in a non-final state (and the error action was embedded into  
> that
> state).
>
> This semantics meant a single action block could be executed on a
> regular character transition and on an EOF event. The problem was that
> regular actions were generated in the main processing loop using  
> 'write
> exec' and EOF actions were generated using 'write eof'. This often led
> to compile errors because the context was not the same. You could fix
> the compile errors by providing dummy variables to the EOF code, but
> there would be no effect in the EOF code.
>
> This inconsistency seemed to drive up the complexity of these  
> operators
> and in the 4.0 line I decided to simplify the operators by eliminating
> the above action embedding behaviour.
>
> In the 6.0 release I will be bringing this behaviour back. Now that  
> EOF
> actions are written in the main processing loop, they have  
> available to
> them everything that regular actions have and the original problem  
> is gone.
>
> -Adrian
>