I got bit by a bug from a number in JSON that was really a date and the generic deserializer that provides an Object that you’ll need to cast to something to use it. Lets start off with the code and the error.

package net.shagie;

import com.fasterxml.jackson.databind.ObjectMapper;

import java.io.IOException;
import java.util.*;

public class Christmas {
    public static void main(String[] args) throws IOException {
        List<Date> cals = new ArrayList<Date>(11);
        for (int year = 1965; year <= 1975; year++) {
            Calendar cal = Calendar.getInstance();
            cal.set(year,  Calendar.DECEMBER, 25);
        ObjectMapper om = new ObjectMapper();

        String json = om.writeValueAsString(cals);


        List<Object> fromJson = om.readValue(json, ArrayList.class);
        for(Object value : fromJson) {
            Date date = new Date((Long) value);

This generates the list of Dates for Christmas for the years 1965 to 1975. It then prints out the JSON, and then converts the list back to a list of Dates and prints out that date.

Sat Dec 25 00:00:00 CST 1965
Sun Dec 25 00:00:00 CST 1966
Mon Dec 25 00:00:00 CST 1967
Wed Dec 25 00:00:00 CST 1968
Exception in thread "main" java.lang.ClassCastException: java.lang.Integer cannot be cast to java.lang.Long
	at net.shagie.Christmas.main(Christmas.java:25)

Ok, something happened here. Somehow, that list of Longs had an Integer sneak in there for December 25, 1969. That should catch some eyes because its rather close to a very special date - January 1, 1970. The time for midnight UTC January 1, 1970 is the epoch time for Unix - the 0 time.

Java uses the number of milliseconds since the epoch for its internal date representation.

The deserializer for JSON doesn’t know that this is a Date underneath and so is just pulling it into a List of number types. It checks to see if the number is possibly an Integer (between -231 and 231 - 1). If not, it tries to put it in a Long instead.

231 milliseconds is 24.85 days. So, from the 7th of December, 1969 to the 25th of January, 1970; the Java internal representation of the time will fit in an Integer. And Jackson nicely puts it in as an Integer.

There are a two ways to fix this. One is to tell Jackson to have always use a Long. DeserializationFeature.USE_LONG_FOR_INTS

And indeed, tossing the following line into the above code will let it run:


But this ignores that the data is really a Date behind that number. It may work in this instance where the data is simple, but in situations where there are actually Integers and Longs (and BigInteger too) all side by side this will cause more problems than it is worth. The thing is, these aren’t really Long numbers. They are Dates that happen to be serialized as a Long.

Thus, the proper approach to this is not to allow a sloppy Object to be tossed around (however convenient it may be to not specify the serialization contract and handle the parse exceptions), but rather to properly deserialize back to the Date from which it came.

List<Date> listFromJson = om.readValue(json, new TypeReference<List<Date>>() {});
Date[] arrayFromJson = om.readValue(json, Date[].class);

Those are the approaches for the List or array types. The type erasure from Java necessitates the TypeReference if you want a List, but an array type is its own type and doesn’t need such contortions.

This is a problem that is particularly notorious in dates and times when going through JSON, which lacks a Date typed object. Thus, different languages have taken different approaches to serializing Dates through JSON. Some Java and C# like to use milliseconds since the epoch. Others may use ISO 8601 dates. There are differing opinions as to which one is more right or least wrong.

Many (if not all) serialization libraries will properly handle the round trip serialization of a Date (or however it is named in the language). However, this depends on the hint (or outright statement) to the library that the object is a Date. Not using the library for both sides of the serialization, and you’ve good a non-zero chance that someone has a date that will break the assumptions about what a Date is.

The bugs that may occur with the improper deserialization of data are subtle and they come out at runtime with nuances of the deserializer that someone may not be aware of. The simplest and best approach to this is to maintain the type system that the language provides - give the deserializer every bit of information that it may want in order to produce the proper object. For a few more seconds of work to deserialize to the contracted object, hours of debugging and maintenance are saved.