On Wed, Aug 6, 2014 at 1:31 AM, Jordan Rose <[email protected]> wrote:
> > On Aug 5, 2014, at 10:18 , Manuel Klimek <[email protected]> wrote: > > On Tue, Aug 5, 2014 at 7:15 PM, Jordan Rose <[email protected]> wrote: > >> I'm sorry, I don't understand what you mean by non-leaking. Do you mean >> that if we decide not to run a destructor, we shouldn't leave that trace in >> the ProgramState? I think it's okay because it will just get postponed to >> when the destructor is actually run later, right? >> > > I mean that until the lifetime extension is fixed, there are lifetime > extended temporaries for which we will mark the ProgramState, but never > clear it, as with the current modeling of lifetime extended temporaries the > destructors for them look like normal destructors, not like temporary > destructors (if we correctly run into them at all, which also has some > bugs). > > > Ah, I see what you mean. Uh...this doesn't make me happy. Data that stays > in the state like this stays in the state forever, including post-inlining. > But I guess it only happens when you turn on temporary destructors, and > we're planning to fix it before we turn that on generally, so okay. > I'll continue to beat that horse until it is thoroughly beaten (so I hope that situation will not stay that way for too long). If necessary, I'll just chain you to a chair at the next LLVM dev meeting and we'll work through it :D
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
