that the number of seconds in an isc_time_t does not
exceed the range of a time_t, or return ISC_R_RANGE.
Similarly, isc_time_now(), isc_time_nowplusinterval(),
isc_time_add() and isc_time_subtract() now check the
range for overflow/underflow. In the case of
isc_time_subtract, this changed a calling requirement
(ie, something that could generate an assertion)
into merely a condition that returns an error result.
isc_time_add() and isc_time_subtract() were void-
valued before but now return isc_result_t.
The seconds member isc_time_t on Unix platforms was changed from time_t
to unsigned int.
unix/time.c now uses macros for nanoseconds per second, nanoseconds per
microsecond and microseconds per second to make sure that the right
number of zeros appears each place the constant is used.
unix/time.c functions which take initialized isc_(interval|time)_t arguments
INSIST() that the nanoseconds value is less than one full second.
unix/time.c's isc_time_microdiff was broken because it did multiplication and
addition with unsigned integers and attempted to set them a 64 bit int to
avoid overflow, but C's ints don't promote to 64 bits on machines that only
have 32 bit longs. Fixed.
Added all the pertinent documentation to time.h.
Though source modules things compile, no programs have yet been
built and tested that actually use networking (as opposed to support
functions, like inet_aton).
should be decided on; it is kind of a toss-up to me personally.
One method is slightly more complicated but the other uses a spin lock.
See once.c for details.
Wrote isc_time_nowplusinterval, a new function added to unix/time.c.
Made some LARGE_INTEGERS to be ULARGE_INTEGERS to ameliorate any
possible hint of a signed vs unsigned problem, even though it would
not happen for 28,847 more years from now.
Assert t1 and t2 not NULL in isc_time_millidiff.
