Skip to content

Length is three different numbers

Level: 101 · start here

One line: "How long is this string?" has three correct answers in Java — UTF-16 units, code points, and grapheme clusters — and length() returns the one that is almost never the one you meant.

The three questions

They are genuinely different questions, and each has a real use:

  • length() — UTF-16 code units. How much memory, how big a char[], what index substring wants. A storage number.
  • codePointCount() — Unicode characters. How many entries from the Unicode table. The number a specification means by "character".
  • Grapheme clusters — how many things a reader would count. é written as e + a combining accent is one. A family emoji built from four people and three joiners is one.

Measured

string         length()   codePoints  graphemes
------------------------------------------------
hello                 5            5          5
zażółć                6            6          6
🎯                    2            1          1
👨‍👩‍👧              8            5          1
café                 5            5          4

length()    counts UTF-16 code units  — storage
codePoints  counts Unicode characters — the table
graphemes   counts what a reader calls a character — the screen

Read the last two rows. The family emoji is 8 by length(), 5 by code points, and 1 to a human being. café written in decomposed form is 5 and 5 and 4. No single number is wrong; they answer different questions.

Where Java leaves you

There is no String method for the third column. Grapheme segmentation lives in java.text.BreakIterator, which is a 1996-era API with a cursor and a DONE sentinel — you write a loop, as the example does. It is correct and it is in the JDK, but nothing about String's own surface suggests you need it.

That gap is why "truncate this to 20 characters for the preview" is a bug factory: substring(0, 20) is a storage operation, and it will split a surrogate pair or orphan a combining accent without a word.

Beside Python

Python 3 stores code points, not UTF-16 units, so its len() answers the middle column and the first column does not exist for a user to trip over:

string            len()   codePoints
------------------------------------
hello                 5            5
zażółć                6            6
🎯                     1            1
👨‍👩‍👧                 5            5
café                 5            5

len() IS the code point count — Python never shows you UTF-16 units.
Grapheme clusters need a third-party library (regex, grapheme).

Worth being fair about: Python removes one of the three traps, not all three. Grapheme clusters need a third-party library there too. What Python does not have is a length() that silently means something else.

See also