Granularity of difftime

Question

In section 7.23.2.2 paragraph 2 and 3 of the C99 standard:

2 The difftime function computes the difference between two calendar times: time1 - time0.

3 The difftime function returns the difference expressed in seconds as a double.

The C99 standard does appear to define a minimum granularity for the difftime function only a resolution (seconds). Are there any implicit guarantees about the granularity of the difftime function?

I can reason that the minimum possible granularity on any given implementation depends on the granularity of the underlying representation of that implementation's time_t data type (since that is what is passed to difftime) -- but from my understanding of the text, difftime isn't required to return the difference at that minimum granularity, and could in fact return at a much larger one.

Answer 1

There are no such guarantees made by the C standard. time_t has an arithmetic type, and is able to represent the implementation's best approximation of a calendar time (with implementation defined range and precision). difftime calculates the difference between two such time_t values, and returns the result in seconds. That's about all that the standard guarantees. Specifically, it does not say how granular/precise the value returned by difftime is (but it can be reasonably assumed to have the same granularity as the time_t type on the same implementation).

The POSIX standard gives one further guarantee, by specifying time_t to hold the seconds since the epoch (either as an integer or floating point type) - ie. it fixes the unit. Again, it does not say anything about the granularity though.

The main reason I can think of that neither of these standards specify the granularity, is that that is hardware dependent, and both standards specify the behavior of software (potentially running on a wide variety of hardware). They don't control the system clock, so can't make any assumptions about its granularity.