On Wed, Jun 11, 2014 at 11:08 AM, Eric Botcazou <ebotca...@adacore.com> wrote: >> Note that I'm happy to revert the change. > > Thanks. > >> I am hesitant to any approach that overloads TREE_ADDRESSABLE even more. >> It already is used for two (slightly) different things - first the >> "old" meaning that the address of the symbol is needed, second, that >> the symbol is aliased by pointers. Those are of course related, but >> as you see they are not 100% equivalent. >> >> As I already added DECL_NONALIASED (for VAR_DECLs) to "fix" that >> coverage counter issue (those are TREE_STATIC but they have their >> address taken - still we know that no pointers alias the accesses), >> we can as well rely on that flag - but then we should set it whenever >> a TU-local decl does not have its address taken (!TREE_ADDRESSABLE). > > Why not just make the change to may_be_aliased in LTO mode, with a comment > saying that TREE_PUBLIC and DECL_EXTERNAL aren't fully correct any longer?
Because that's not the point and because it feels like a hack ;) I could do || in_lto_p and make LTO "properly" set TREE_ADDRESSABLE. But then we can make the IPA machinery "properly" set TREE_ADDRESSABLE as well, unconditionally, just like with my proposed patch to fix the fallout. Richard. > -- > Eric Botcazou
