Interning and compact strings

For this lesson, we'll focus on the exact semantics of intern() and what it costs, when interning saves memory and when it quietly hurts, how G1's -XX:+UseStringDeduplication differs from interning, how compact strings change the layout of every String, and how to measure all of this with JOL and a heap dump. This topic comes up in senior and performance-focused rounds, usually as "your service uses 6 GB of heap and half of it is strings, what do you do?".

01Picture it first: a library catalogue and a photocopier

Imagine a warehouse that stores a printed label on every one of 5 million parcels. Most labels say one of 40 city names. Three ways to save paper:

• Interning is a clerk who, every time a label is printed, checks a central catalogue and says "we already have a 'MUMBAI' label, staple a pointer to that one instead". It works, but every parcel waits for the clerk.
• Deduplication is a night-shift worker who walks the shelves while nobody's looking, finds parcels with identical labels and makes them all point to one sheet. Daytime work isn't slowed, but the parcels themselves (the String objects) are still there.
• Compact strings is printing labels in a smaller font whenever the text is plain English letters, and the big font only when a label contains a symbol like ₹.

Let's start with the font, because it applies to every string in every Java 9+ program. The picture below compares the same text before and after JEP 254.

Fig. 1

On Java 8 "hello" costs 56 bytes because each char takes two bytes; with compact strings the LATIN1 form costs 48 bytes; a single non-Latin-1 character like the rupee sign switches the whole string to two bytes per char

02Compact strings under the hood (JEP 254)

Here's how I measure it with JOL, the Java Object Layout tool from OpenJDK:

import org.openjdk.jol.info.ClassLayout;
import org.openjdk.jol.info.GraphLayout;

public class StringFootprint {
    public static void main(String[] args) {
        System.out.println(ClassLayout.parseClass(String.class).toPrintable());
        for (String s : new String[] {"hello", "héllo wörld 2026 ok", "₹ payment"}) {
            System.out.println(s + " -> " + GraphLayout.parseInstance(s).totalSize() + " bytes");
        }
    }
}