From: Julian S. <js...@ac...> - 2008-12-16 09:36:17
|
On Sunday 14 December 2008, Bart Van Assche wrote: > Both algorithms are valuable. Unfortunately there is not yet any known > way to combine those two algorithms (Eraser and happens-before) > without also combining the disadvantages of both algorithms. It could be argued that, because of this problem, race detection is really not something very suitable for a dynamic tool. For Miguel's example it would be better in fact to construct a formal proof (somehow!) which shows it is race-free regardless of the scheduling. In other words, maybe static analysis of the program is better. But obviously that is far beyond the scope of Valgrind, and I would guess far beyond the scope of the art for anything but the smallest programs. J |