bkohler@ noticed that the i486 catalyst builds were failing with the same symptoms as were reported in bug 893776. sam identified that fstat() was failing, leaving `st` uninitialized. This didn't reproduce on another system because it was dependent on the underlying file system and whether the particular file fstat'd had an inode number > 2^32. It might be nice if sandbox warned loudly if stat/fstat fails with EOVERFLOW.