Re: Visibility and error messages (in named refs).

[Joel E. Denny]

Hi Alex.

On Sat, 18 Jul 2009, Alex Rozenman wrote:

> I think we all agreed now that the "misleading reference" will be a warning.

Ok.

> Now, my biggest (and sole) concern is the following: would we like to give
> warnings in the following cases:
> 
> stmt[result] : IF expr THEN stmt;
> { $result = create_if_stmt($expr, $stmt); }
> 
> if_stmt : IF expr THEN stmt[then] ELSE stmt;
> { $$ = create_if_stmt($expr, $then, $stmt); }
> 
> Currently, my intuitions says not to issue any warnings (compile smoothly)
> in the first case, and give a "misleading reference" warning in the second
> case. This probably because I started with not including LHS at all and my
> brain just refuses to change the view. I just afraid that this single issue
> will cause to multiple complains from users and we will be forced to change
> it anyway.
> Akim, could you please give you opinion ?

I feel this would be a subtle inconsistency for the LHS and thus would 
make Bison's behavior more complex and confusing to the user.  I don't see 
a benefit.  Of course, I'd be happy to hear Akim's opinion too.

> I think it's confusing and not helpful to the user if Bison makes it an
> > error only in
> > some subcases.

I'm not sure my wording was clear here.  By "subcases", I meant subcases 
of the misleading reference case.  In your previous email, you wanted an 
error for the subcase where a "possibly meant" variant involved a 
bracketing change, and so I was arguing against that.  Above, I think 
you're now agreeing that all misleading reference subcases including this 
one should be warnings.

> Sub cases are our implementation detail. I'm not sure users will be ever
> aware of a similarity between these situations. We are free to decide about
> the behavior in each specific case.

I'm not sure I understand this comment.  Are you again suggesting that 
some subcases of the misleading reference case should be errors?  I think 
this inconsistency would make the error subcases even more confusing than 
if we made all subcases errors.

Again, I think Bison's behavior will be most intuitive if a reference is 
never an error when it is valid and unambiguous.  Instead, Bison just 
warns that it's misleading.