Paul Eggert wrote: > setitimer has nanosecond resolution, which is better than the > one-second resolution that 'alarm' has. timeout should use > setitimer if available, to take advantage of this. On 64-bit > hosts, this has the additional advantage of increasing the > upper bound for timeouts from 2**31 seconds to 2**63 seconds > (about 68 years to about 292 billion years, which should be > long enough for most practical purposes :-).
I like the idea of supporting a sub-second timeout interval, but it probably deserves a warning in the documentation. Even a command like "timeout 3 sleep 1" will timeout on a system under heavy load.
