Fixed bug #74851: Improve uniqid() performance

uniqid() relies on microsecond-precise system clock to produce an
unique identifier. In order to avoid  using the same value, the
original implementation calls usleep(1) to wait for the next microsecond.

Unfortunately, usleep() specification says "The suspension time may be
longer than requested due to the scheduling of other activity by the
system." Indeed, the pause may as as long as an entire execution slice,
causing a uniqid() call to last more than 10 ms.

This is fixed by replacing the usleep() call by time polling using
gettimeofday() until the microsecond changes. Since the getttimeoday()
system call lasts around a microsecond, only a small time is wasted
calling multiple gettimeofday. On the benefit side, uniqid() performance
in increased 10000 fold without changing its behavior.

0 files changed, 0 insertions, 0 deletions