[Bug c++/66786] [5/6 Regression] ICE: Segmentation fault

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66786

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
                 CC|                            |ppalka at gcc dot gnu.org
           Assignee|unassigned at gcc dot gnu.org      |ppalka at gcc dot gnu.org

--- Comment #4 from Patrick Palka <ppalka at gcc dot gnu.org> ---
Slightly more simplified test case:

  template <typename... T> auto list = [](T... xs) { [=](auto f) { f(xs...); };
};
  int main() { list<int>; }


The problem ultimately seems to lie in template_class_depth, which returns the
number of enclosing templates in which the given type or decl is embedded.


In particular, given the RECORD_TYPE that corresponds to the outermost lambda
in the above test case, template_class_depth returns 0 instead of 1 since the
TYPE_CONTEXT of this RECORD_TYPE is the file's TRANSLATION_UNIT_DECL and not
the VAR_DECL so template_class_depth never sees it.

That the TYPE_CONTEXT doesn't point to the enclosing VAR_DECL doesn't seem like
a bug because according to the documentation, the TYPE_CONTEXT (and
DECL_CONTEXT) of a tree is never expected to be a VAR_DECL.  It can be a
FUNCTION_DECL however, which explains why the following test case, which just
replaces the enclosing variable template with a function template, compiles
successfully (the TYPE_CONTEXT points to the enclosing FUNCTION_DECL and so
template_class_depth appropriately returns 1):

  template <typename... T>
  auto
  list ()
  {
    return [](T... xs) { [=](auto f) { f(xs...); }; }; 
  }

  int main() { list<int> (); }

So I wonder how to make template_class_depth be aware of the given type's
enclosing variable template...