[Bug c++/70847] [6/7 Regression] exponential time in cp_fold for chained virtual function calls

["ppalka at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70847

Patrick Palka <ppalka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ppalka at gcc dot gnu.org

--- Comment #8 from Patrick Palka <ppalka at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #7)
> Created attachment 38636 [details]
> gcc7-pr70847.patch
> 
> Untested fix.

Wouldn't that cause us to fully fold only the first occurrence of a shared tree
and miss folding subsequent occurrences?

If we have a CALL_EXPR whose 1st and 2nd arguments share the same tree

  f (x * 0, x * 0);

then IIUC with the patch only the 1st argument would get folded, giving

  f (0, x * 0);

Not sure if this is an actual issue in practice, but if it is, it may be
workable by maintaining the pset ourselves in cp_fold_r and doing something
like

  if (pset->add (*stmt_p))
    *walk_subtrees = 0;

That way *stmt_p will still be set by cp_fold() even when it's a tree that's
alraedy been visited.