On 6/13/06, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
"Garrett Rooney" <[EMAIL PROTECTED]> wrote:
> On 6/13/06, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
> > "Garrett Rooney" <[EMAIL PROTECTED]> wrote:
> > > I've been using lemon as the parser generator for ETL
> > > and we've been running into some problems with recent
> > > versions. The first problem is in revision 1.17 of lempar.c. You
> > > removed an if( yymajor==0 ) return; from Parse,
> >
> > The EOF token is necessary. But I think Lemon is correct as
> > stands - as far as I can tell. Can you please tell me more about
> > the circumstances of your segfault - specifically where it is
> > happening.
>
> Well, it doesn't manifest itself in all my test cases, but with the
> ones that do crash it's because they end up in a %syntax_error block
> with the %extra_argument set to NULL. With the if( yymajor==0 ) check
> they parse just fine, never falling into the syntax error block at
> all.
Your grammer does not generate an empty string. That means
if you call the parser with an EOF token first, without any
prior tokens, you are going to get a syntax error. The
"yymajor==0" test was removed in order to get this to
work correctly. Geert Janssen at IBM found this bug back
in November - his grammar does not generate an empty string
but lemon was not giving him a syntax error like it should
if he passed in an empty string. If I add the yymajor==0
test back, it will reintroduce the bug.
Hmm. I don't believe that's happening, but I'll look into adding
something to deal with that case.
Perhaps you are calling Parse() with EOF twice in a row
when you should only be calling it once? That would trigger
the syntax error. Or perhaps you should adjust your grammar
to accept an empty string?
Nope, I only send EOF in one location, the lexer never produces it,
and I only pass it in after I've run out of tokens from the lexer.
The %extra_argument that gets passed to the %syntax_error procedure
should be the %extra_argument that got passed into with the
call to Parse() that contained the EOF token. If that %extra_argument
was NULL, then the %syntax_error procedure will get a NULL.
Perhaps you can either pass in a non-null parameter to Parse
or check for NULL in %syntax_error and branch accordingly.
The extra argument passed in to the parse function is a pointer to
memory on the stack. It's just not possible for it to be null, unless
something is messing with that value before it gets into the
%syntax_error block...
-garrett
