I think the specification of "fractional_seconds" in the time_duration class is unclear. Using the multi-argument constructor, and passing 1000 for the 4th argument, am I specifying milliseconds? Microseconds? Does it depend on the compile-time resolution of the class? If so (which is my suspicion), then using the 4-argument form of the ctor can lead to incorrect results if the resolution is changed. That seems like a bad thing. Looking at the implementation, I can see that I can divide the result of fractional_seconds() by the return value of ticks_per_second() to get a floating point value < 1.0, but I think that this should be more clearly spelled out in the section: doc/html/date_time/posix_time.html#date_time.posix_time.time_duration I'm not sure of the wording, but something like: *Construction* Mention that fractional_seconds depends on the compile-time resolution of the time_duration. I'd personally suggest it not be used and that the count-based construction be used instead. *Accessors Section* <tr> <td>static long ticks_per_second()</td> <td>The number of ticks per second. This is the resolution of the time_duration class</td> <td>long ticks_per_sec = time_duration::ticks_per_second();</td> <tr> <td>long fractional_seconds() const</td> <td>Get the number of fractional seconds. The return value will be in the range [0, ticks_per_second())</td> <td> time_duration td(1,2,3, 1000); td.fractional_seconds() --> 1000 <br/> double subsec = td.fractional_seconds() / time_duration::ticks_per_second () --> 0.001 (?) </td> Is this reasonable? -- Caleb Epstein caleb dot epstein at gmail dot com