Re: Question on tree-walking and mutually-recursive types

[Andrew Pinski <pinskia at physics dot uc dot edu>]

> 
> walk_tree walks the type of all variables in a BIND_EXPR.  And walking
> a POINTER_TYPE walks the type it points to.
> 
> But it's possible for a pointer type X to point to a pointer type Y, which
> points back to X.  I don't think you can express this in C, but you can in Ada:
> 
> procedure tpself is
>   type Tp1;
>   type tp2 is access Tp1;
>   type Tp1 is access Tp2;
>   X: Tp1;
> begin
>    null;
> end tpself;
> 
> You may ask why somebody would want to write something like that and it's a
> good question, but the general rule for Ada is that if you *can* write
> something, you can be sure that an ACATS test *will* try it.  In this case
> it's c38102a, part of which is explicitly written just to test this case!
> 
> We go into an infinite loop here (not recursion because of WALK_SUBTREE_TAIL).
> 
> It's tempting to not go into TREE_TYPE of a POINTER_TYPE in the tree walk,
> but I found out the hard way last week that issuing certain error messages
> for C++ templates depends on that walk being done.
> 
> Any suggestions how to deal with this?

Yes walk using a hashtable saying you already walked the tree.
walk_tree_without_duplicates is what you want.

Thanks,
Andrew Pinski