Why is the following date conversion in Java 8 not appropriate?

Question

I have seen a lot of debates on the following date conversion:

timeStamp.toLocalDateTime().toLocalDate();

Some people say that it is not appropriate because the timezone has to be specified for proper conversion, otherwise the result may be unexpected. My requirement is that I have an object that contains Timestamp fields and another object that contains LocalDate fields. I have to take the date difference between both so I think that the best common type to use is LocalDate. I don't see why the timezone has to be specified as either timestamp or LocalDate just represent dates. The timezone is already implied. Can someone give an example when this conversion fails?.

Answer 1

Example: the 23rd of January becomes the 24th

You asked:

Can someone give an example when this conversion fails?.

Yes, I can.

Start with the 23rd of January.

LocalDate ld = LocalDate.of( 2020 , Month.JANUARY , 23 );
LocalTime lt = LocalTime.of( 23 , 0 );
ZoneId zMontreal = ZoneId.of( "America/Montreal" );
ZonedDateTime zdt = ZonedDateTime.of( ld , lt , zMontreal );
Instant instant = zdt.toInstant();

zdt.toString() = 2020-01-23T23:00-05:00[America/Montreal]

instant.toString() = 2020-01-24T04:00:00Z

The Instant class represents a moment as seen in UTC. Let's convert to the terribly legacy class java.sql.Timestamp using the new conversion method added to that old class.

// Convert from modern class to troubled legacy class `Timestamp`. 
java.sql.Timestamp ts = Timestamp.from( instant );

ts.toString() = 2020-01-23 20:00:00.0

Unfortunately, the Timestamp::toString method dynamically applies the JVM’s current default time zone while generating text.

ZoneOffset defaultOffset = ZoneId.systemDefault().getRules().getOffset( ts.toInstant() );
System.out.println( "JVM’s current default time zone: " + ZoneId.systemDefault() + " had an offset then of: " + defaultOffset );

JVM’s current default time zone: America/Los_Angeles had an offset then of: -08:00

So Timestamp::toString misreports the object’s UTC value after adjusting back eight hours from 4 AM to 8 PM. This anti-feature is one of several severe problems with this poorly designed class. For more discussion of the screwy behavior of Timestamp, see the correct Answer by Ole V.V.

Let's run your code. Imagine at runtime the JVM’s current default time zone is Asia/Tokyo.

TimeZone.setDefault( TimeZone.getTimeZone( "Asia/Tokyo" ) );
LocalDate localDate = ts.toLocalDateTime().toLocalDate();

Test for equality. Oops! We ended up with the 24th rather than the 23rd.

boolean sameDate = ld.isEqual( localDate );
System.out.println( "sameDate = " + sameDate + " | ld: " + ld + " localDate: " + localDate );

sameDate = false | ld: 2020-01-23 localDate: 2020-01-24

See this code run live at IdeOne.com.

---

So what is wrong with your code?

• Never use java.sql.Timestamp. It is one of several terrible date-time classes shipped with the earliest versions of Java. Never use these legacy classes. They have been supplanted entirely by the modern java.time classes defined in JSR 310.
• You called toLocalDateTime which strips away vital information. Any time zone or offset-from-UTC is removed, leaving only a date and a time-of-day. So this class cannot be used to represent a moment, is not a point on the timeline. Ex: 2020-12-25 at noon — is that noon in Delhi, noon in Düsseldorf, or noon in Detroit, three different moments several hours apart? A LocalDateTime is inherently ambiguous.
• You ignored the crucial issue of time zone in determining a date. For any given moment, the date varies around the globe. At one moment it may be “tomorrow” in Australia while simultaneously “yesterday” in Mexico.

Answer 2

It’s more complicated than that. While it’s true that a Timestamp is a point in time, it also tends to have a dual nature where it sometimes pretends to be a date and time of day instead.

BTW, you probably already know, the Timestamp class is poorly designed and long outdated. Best if you can avoid it completely. If you are getting a Timestamp from a legacy API, you are doing the right thing: immediately converting it to a type from java.time, the modern Java date and time API.

Timestamp is a point in time

To convert a point in time (however represented) to a date you need to decide on a time zone. It is never the same date in all time zones. So the choice of time zone will always make a difference. So one correct conversion would be:

    ZoneId zone = ZoneId.of("Africa/Cairo");
    LocalDate date = timestamp.toInstant().atZone(zone).toLocalDate();

The Timestamp class was designed for use with your SQL database. If your datatype in SQL is timestamp with time zone, then it unambiguously denotes a point in time, and you need to see it as a point in time as just described. Even when to most database engines timestamp with time zone really just means “timestamp in UTC”, it’s still a point in time.

And then again: sometimes to be thought of as date and time of day

From the documentation of Timestamp:

A Timestamp also provides formatting and parsing operations to support the JDBC escape syntax for timestamp values.

The JDBC escape syntax is defined as

yyyy-mm-dd hh:mm:ss.fffffffff, where fffffffff indicates nanoseconds.

This doesn’t define any point in time. It’s a mere date and time of day. What the documentation doesn’t even tell you is that the date and time of day is understood in the default time zone of the JVM.

I suppose that the reason for seeing a Timestamp in this way comes from the SQL Timestamp datatype. In most database engines this is a date and time without time zone. It’s not a timestamp, despite the name! It doesn’t define a point in time, which is the purpose of and is in the definition of timestamp.

I have seen a number of cases where the Timestamp prints the same date and time as in the database, but doesn’t represent the point in time implied in the database. For example, there may be a decision that “timestamps” in the database are in UTC, while the JVM uses the time zone of the place where it’s running. It’s a bad practice, but it is not one that will go away within a few years.

This must also have been the reason why Timestamp was fitted with the toLocalDateTime method that you used in the question. It gives you that date and time that were in the database, right? So in this case your conversion in the question ought to be correct, or…?

Where this can fail miserably without us having a chance to notice is, as others have mentioned already, when the default time zone of the JVM is changed. The JVM’s default time zone can be changed at any time from any place in your program or any other program running in the same JVM. When this happens, your Timestamp objects don’t change their point in time, but they do tacitly change their time of day, sometimes also their date. I’ve read horror stories — in Stack Overflow questions and elsewhere — about the wrong results and the confusion coming out of this.

Solution: don’t use Timestamp

Since JDBC 4.2 you can retrieve java.time types out of your SQL database. If your SQL datatype is timestamp with time zone (recommended for timestamps), fetch an OffsetDateTime. Some JDBC drivers also let you fetch an Instant, that’s fine too. In both cases no time zone change will play any trick on you. If the SQL type is timestamp without time zone (discouraged and all too common), fetch a LocalDateTime. Again you can be sure that your object doesn’t change its date and time no matter if the JVM time zone setting changes. Only your LocalDateTime never defined a point in time. Conversion to LocalDate is trivial, as you have already demonstrated in the question.

Links

• java.sql.Timestamp documentation
• Wikipedia article: Timestamp
• Question: Getting the date from a ResultSet for use with java.time classes
• Question: Java - Convert java.time.Instant to java.sql.Timestamp without Zone offset

Answer 3

As you can see here(taken from https://stackoverflow.com/a/32443004/1398418):

Timestamp represents a moment in UTC and is the equivalent of the modern Instant.

When you do:

timeStamp.toLocalDateTime().toLocalDate();

the timeStamp is converted from UTC to the system timezone. It's the same as doing:

timeStamp.toInstant().atZone(ZoneId.systemDefault()).toLocalDate()

For example:

Timestamp stamp = new Timestamp(TimeUnit.HOURS.toMillis(-1)); // UTC 1969-12-31
System.setProperty("user.timezone", "EET"); // Set system time zone to Eastern European EET - UTC+2
stamp.toLocalDateTime().toLocalDate(); // represents EET 1970-01-01
stamp.toInstant().atZone(ZoneId.systemDefault()).toLocalDate(); // represents EET 1970-01-01

That result (getting the date in the system time zone) is expected and if that's what you want, doing timeStamp.toLocalDateTime().toLocalDate() is appropriate and correct.

You're saying that you have a LocalDate field in some object and you want to get a period between it and a Timestamp, well that's just not possible without aditional information. LocalDate just represents a date, it has no time zone information, you need to know how it was created and what time zone was used.

If it represent a date in the system time zone then getting the period by using timeStamp.toLocalDateTime().toLocalDate() would be correct, if it represents a date in UTC or any other time zone then you might get a wrong result.

For example if the LocalDate field represents a date in UTC you will need to use:

timeStamp.toInstant().atZone(ZoneId.of("UTC")).toLocalDate();

Answer 4

The problem lies in what is being represented by these objects. Your question forgets a crucial aspect, which is: What is the type of timeStamp?

I'm guessing it's a java.sql.Timestamp object.

Timestamp, just like java.util.Date, is old API equivalent to Instant.

It represents an instant in time, in the sense that it is milliseconds since jan 1st 1970 UTC. The system has no idea which timezone that was supposed to be in. You're supposed to know; the error, if an error is going to occur here, already occurred before you get to this code. Here's a trivial explanation of how it COULD go wrong:

1. you start off with a user entering a date in a date field on a webform; it's 2020-04-01.

2. Your server, running in Amsterdam, saves it to a DB column that is internally represented as UTC, no zone. This is a mistake (you're not saving an instant in time, you're saving a date, these two are not the same thing). What is actually stored in the DB is the exact moment in time that it is midnight, 2020-04-01 in amsterdam (in UTC, that'd be 22:00 the previous day!).

3. Later, you query this moment in time back into a java.sql.Timestamp object, and you're doing this when the server's tz is elsewhere (say, London time). You then convert this to a localdatetime, and from there to a localdate, and.... you get 2020-03-31 out.

Whoops.

Dates should remain dates. Never convert LocalX (be it Time, Date, or DateTime) to Instant (or anything that effectively is an instant, including j.s.Timestamp, or j.u.Date - yes, j.u.Date does NOT represent a date, it is very badly named), or vice versa, or pain will ensue. If you must because of backward APIs take extreme care; it's hard to test that 'moving the server's timezone around' breaks stuff!