Re: Deallocations upon parse errors?

[Hans Aberg]

Not going into the stuff you are doing (as I use a C++ semantic type,  
doing the cleanup via the language :-)), the %destructor is not  
output language related, but merely invokes some code when the parser  
stack unwinds during error recovery. So if you use a pointer type, I  
gather %destructor should be as usable both when using the C and C++  
parsers. The C-parser does not support compiled under C (though in  
earlier version of Bison, it may have compiled that way), so better  
use the C++ parser when using C++.

  Hans Aberg


On 17 Jan 2007, at 05:10, David Fang wrote:

> > Pick down some later version of Bison (2.3+); then, if you use the C
> > parser, there is a %destructor command, that can be used to handle
> > clean-up during error recovery. Another way is to use a C++ parser,
> > and a semantic type that does cleanup via the C++ language, but one
> > cannot then use %union.
> >
> >    Hans Aberg
>
> Hi,
>         One of these days I'll sit down an use the new %destructor feature
> in bison, but FWIW, it IS *possible* to to clean-up of the parser  
> stack in
> C++, using only the C skeleton, and while using a union of various  
> pointer
> types.  It involves deducing the union type based on the state  
> numbers and
> transitions on the parser stack.  I described the recipe for doing  
> this in
> an old thread.
>
> http://lists.gnu.org/archive/html/help-bison/2006-04/msg00023.html
>
> Yeah, it might be disgusting, but it was educational.  :)
>
> I verified that no symbol objects were leaked in the event of parse  
> errors
> by defining class-overloaded operator new/delete (for all pointee  
> types in
> the union) to use pool-allocators and asserting that the pools were
> 'clean' upon global static destruction.  This method works with
> traditional yacc, byacc, and all released versions of bison from  
> 1.35 to
> 2.3 (present).
>
> Fang