[Bug c++/70001] [5/6 regression] Infinity compilation time

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70001

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 37924
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37924&action=edit
gcc6-pr70001.patch

So do you mean something like this?  The *-2.C testcase is what I believe
you've been talking about and indeed Patrick's patch broke that.
I'm still wondering if there could be dependency on this even without
relocation, but e.g. for:
constexpr __UINTPTR_TYPE__ foo (void *p) { return (__UINTPTR_TYPE__) p; }
struct X
{
  __UINTPTR_TYPE__ a;
  constexpr X () : a (foo (this)) { }
};
constexpr signed char bar (void *p) { return (signed char) (__UINTPTR_TYPE__)
p; }
struct Y
{
  signed char a;
  constexpr Y () : a (bar (this)) { }
};

constexpr int N = 1 << 4;
struct A { X elems[N]; } F;
struct B { Y elems[N]; } G;

with (unpatched) g++ F contains relocations and G contains all zeros.  clang++
3.8 rejects both constexpr functions as having reinterpret cast, so maybe that
just isn't valid (shall g++ reject that too)?