Excel 2007 (also 2003)

For the number of days between March 30, 2009, and April 1, 2009 (for
example), the following formulas give the following results:

=B6-A6 2 days
=DATEDIF(A6,B6,"d") 2 days
=DATEDIF(A6,B6,"md") 1 day

I didn't realize this until I happened to fill in two dates that
crossed two months but were less than one month apart. My actual
formula is the following, which I thought would safely cover date
differences of any length:

DATEDIF(A2,B2,"y") & " year(s), " & DATEDIF(A2,B2,"ym") & " month(s),
" & DATEDIF(A2,B2,"md") & " day(s)")

The problem is the "md" argument in the day component. Chip Pearson
at http://www.cpearson.com/excel/datedif.aspx provides the definition
of "md" as "Complete calendar days between the dates AS IF THEY WERE
of the same month and same year" (apparently from the Excel 2000 Help
files). What does that mean --- "as if they were"?

If you substitute in my long formula above "d" in place of "md", you
get a poor result when the dates are farther apart: e.g., March 12
2008 to April 6 2009 gives the result 1 year, 1 months, 419 days.

In short, you can't use my long formula above as a foolproof general
way to calculate date differences, because if your result is less than
a month but happens to encompass two calendar months, you'll get an
incorrect figure.

I guess the safest solution is to have an embedded "IF" function: If
year difference is 0 and month difference is 0, simply subtract the
dates; otherwise, use the long formula above -- ??

Thanks for any comments.