[Bug tree-optimization/19853] [4.0 Regression] ICE with address in struct assignment

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2005-02-12 01:36 -------
Diego, I think it's clear that adding a new global variable is much trickier
business than we've previously given it credit.

For the special case of calls, we could get away with recording which 
to-be-renamed variables are actually new and if the variable is call-clobbered,
add a vop at the call site.

But that doesn't handle any pointer operation referencing global non-conflicting
TMT's.  And I'm definitely not certain if or how NMTs might need updating to cope
with the newly exposed variable.

I'm not sure how to do anything but run a brand-new alias pass here and anywhere
else we expose new global variables (e.g DOM?).

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dnovillo at gcc dot gnu dot
                   |                            |org
   Last reconfirmed|2005-02-09 14:38:06         |2005-02-12 01:36:31
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19853