Hey hackers, I was doing some work in backend/utils/adt/formatting.c, and found the following:
case DCH_D: INVALID_FOR_INTERVAL; if (is_to_char) { sprintf(inout, "%d", tm->tm_wday + 1); if (S_THth(suf)) str_numth(p_inout, inout, S_TH_TYPE(suf)); return strlen(p_inout); } else { sscanf(inout, "%1d", &tmfc->d); return strspace_len(inout) + 1 + SKIP_THth(suf); } The tm_wday field is internally stored as an integer 0 - 6, with 0 being Sunday. The 'D' formatting field, as per the documentation, gives 1 - 7 with 1 being Sunday. So to convert tm_wday to 'D' in to_char(), you add one. This works as expected. However, in from_char(), the reverse is not true. Looking at the code snippet above, the digit is scanned straight into tmfc->d unaltered (this value is later copied directly to tm->tm_wday circa line 3394). Unless I'm missing something, when converting to text, 'D' yields 1-7, but when converting back from text, 'D' expects 0-6. It's not a major problem, since there's not really a use-case for specifying dates for conversion with the 'D' field, but this behaviour appears to be incorrect, or at the very least, incorrectly documented. The fix should be trivial; subtract one from tmfc->d after the call to sscanf() Regards, BJ ---------------------------(end of broadcast)--------------------------- TIP 3: Have you checked our extensive FAQ? http://www.postgresql.org/docs/faq