When you're trying to do calendar arithmetic using time_t, you naturally have to worry about the type and representation of time_t, which is of course implementation-defined. It's almost always a signed, integral type. On Unix and modern MacOS systems it's seconds since 1970, and for compatibility I think it might be used that way on Windows, too. It tends to be 32 bits. Putting that all together, it can typically represent dates between December 13, 1901 and January 18, 2038.
And indeed when I changed the tm_year line in your code to
referenceDateComponent.tm_year = 60;
the code worked and printed 1721200226, which is about 19921 days or 54.5 years, which is exactly the difference between December 31, 1960 and today.
But if you set tm_year to be negative, you'd be asking for a date before 1900, and that's not going to work using the typical definition of time_t we've been discussing.
(It's true there are other possibilities for time_t. It could be floating point. It could be unsigned instead of signed. It could be a 64-bit type, meaning it'd have a range of almost 600,000,000,000 years, which is incidentally more than a 32-bit tm_year can hold.)
So although there are several naysayers here telling you not to, and although there are certainly plenty of obscure difficulties having to do with time zones and leap seconds and calendars other than Gregorian, you can usually get away with using time_t to do basic calendar math for dates in the 20th century and in this century up until 2038, when the infamous "Y2.038K problem" is going to hit. (It will, I fear, be somewhat worse than the not-so-infamous Y2K problem, but that's another story.)
As I said, your code worked for me for dates before 1970. Here's what I'd recommend using for simple time_t-based date calculations (and with caveats as already mentioned):
time_t makedate(int year, int month, int day)
{
struct tm tm = {0};
tm.tm_hour = 12;
tm.tm_min = tm.tm_sec = 0;
tm.tm_year = year - 1900;
tm.tm_mon = month - 1;
tm.tm_mday = day;
tm.tm_isdst = -1;
return mktime(&tm);
}
int main()
{
long int d = difftime(makedate(2015, 7, 17), makedate(2015, 6, 1));
printf("%ld sec = %ld days = %.2f years\n", d, d/86400, d/86400./365.25);
d = difftime(makedate(2015, 7, 17), makedate(2014, 7, 17));
printf("%ld sec = %ld days = %.2f years\n", d, d/86400, d/86400./365.25);
d = difftime(makedate(1975, 7, 17), makedate(1965, 7, 17));
printf("%ld sec = %ld days = %.2f years\n", d, d/86400, d/86400./365.25);
d = difftime(makedate(1950, 1, 11), makedate(1925, 1, 1));
printf("%ld sec = %ld days = %.2f years\n", d, d/86400, d/86400./365.25);
d = difftime(makedate(2030, 12, 31), makedate(2025, 12, 31));
printf("%ld sec = %ld days = %.2f years\n", d, d/86400, d/86400./365.25);
}
Just like your code, this leverages the surprisingly powerful mktime function, and can do everything it can do. It handles leap years, no problem. It does not handle leap seconds or calendar changes.
And if, as you say, you're interested in dates before 1900, I'm afraid you're out of luck. time_t simply cannot represent those dates on most systems, so you're going to have to pursue some other solution.
