Re: RFC: Freeing discarded objects

[Cris Bailiff]

> Subject: RFC: Freeing discarded objects
> From: Akim Demaille <address@hidden>
> Date: 17 Jun 2002 10:43:16 +0200
>
> Nevertheless, I would really appreciate comments on this.  In
> particular, I tried to stick to the Yacc traditional order =E0 la %type:
> the important thing is at the beginning, then a list of symbols
> behind.  Does it sound right to you?

Akim,

I scanned your patch (esp. to bison.simple), and it's looking useful, but I 
think you missed a case (I missed the same spot at first):

Just before here:
>  /* YY_DECL_VARIABLES -- depending whether we use a pure parser,
>     variables are global, or local to YYPARSE.  */
> @@ -1123,6 +1126,7 @@
>       YYABORT;
>        YYDPRINTF ((stderr, "Discarding token %d (%s).\n",
>                 yychar, yytname[yychar1]));
> +      yydestructor (yychar1, yylval);
>        yychar =3D YYEMPTY;
>      }

(where the discard is performed) is the case for exiting at EOF. If you're 
half way through a production when eof comes, and your not otherwise caught 
by an error  handler (e.g. "statement:  error ';' " doesn't match at eof!), 
then bison will return at YYABORT here with leaked data on the stack.

I've attached a small patch which should fix this case - I'm not so informed 
about all the internals of the run-time engine, so I hope I picked the right 
parameters for yydestructor here...

Cris

bison.simple.free-at-eof.patch
Description: fix EOF case