Our design [for kill-safe abstractions] builds on the observation that … to manage a particular synchronization abstract, and by … is suspended when it might otherwise be killed. … out to be "only mostly dead," and a … [p. 49] When the [surviving] task later accesses the … [instance] … and safely continues.  If both  … tasks are terminated, the [instance] becomes both suspended … 
     Google-1
     Google-2
   
   Copyright clearance needed for quotation.