[Bug middle-end/65074] [5 Regression] r220674 broke C++ PIEs

[rth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65074

Richard Henderson <rth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #34770|0                           |1
        is obsolete|                            |
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rth at gcc dot gnu.org

--- Comment #4 from Richard Henderson <rth at gcc dot gnu.org> ---
Created attachment 34785
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34785&action=edit
second patch

I prefer to remove the test against node->definition entirely.
It appears to mean only that we have a definition that we could
emit, not that we're going to actually emit the symbol.

In particular, for functions, we'd have to also check cgraph_node::process.
At which point I suspect that DECL_EXTERNAL is sufficient.  Which means
that we can at least set defined_locally based on DECL_EXTERNAL even if
there is no symtab_node.  I don't know that we'd ever not have one, but
that's certainly not obvious from here.

Starting regression testing on several platforms...