Question

I have encountered a very weird behavior while using SimpleDateFormat for parsing a string to a date. Consider the following unit test:

@Test
public void testParse() throws ParseException
{
    DateFormat dateFormat = new SimpleDateFormat("yyyyMMdd");

    String dateStr = "2012-12-21";
    Date parsedDate = dateFormat.parse(dateStr);
    Calendar date = Calendar.getInstance();
    date.setTime(parsedDate);

    Assert.assertEquals(2012, date.get(Calendar.YEAR));
    Assert.assertEquals(11, date.get(Calendar.MONTH)); // yeah, Calendar sucks
    Assert.assertEquals(21, date.get(Calendar.DAY_OF_MONTH));
}

As it can be seen there is an intentional mistake in the above code: the SimpleDateFormat is initialized with "yyyyMMdd" but the string to be parsed is in the format "yyyy-MM-dd". I would expect that such thing results in a ParseException, or at least be parsed on a best-effort basis correctly. Instead, for some weird reason the date is parsed as 2011-11-02. Eh?!

This is unacceptable as one single mistake while handling the inputs would result in something totally unexpected / devastating. Switched to JodaTime in the meantime, but it would be nice to understand what went wrong there.

Answer 1

Well, the input would be split into 3 components: year, month, day and you’d get month = -12 and day = -21 (for correction see below). Try to parse 2012/12/21 and you’ll get the exception 🙂

Edit: excerpt from the JavaDoc:

Month: If the number of pattern letters is 3 or more, the month is interpreted as text; otherwise, it is interpreted as a number.

Edit2: Correction

Looking at the source of SimpleDateFormat it seems that 2012-12-21 is actually split into this:

year = "2012"
month = "-1"
day = "2-" 

The source comments state that a - following a number might either denote a negative number (depending on the locale) or be a delimiter. In your case it seems to be taken as a delimiter, thus day = "2-" results in day = 2, hence the second of November.