字符串长度歧义

What Is String Length and Why Is It Complicated?

Asking "how long is this string?" seems simple but has multiple valid answers depending on what you mean by "length":

1. Bytes: How many bytes does the encoded string occupy?
2. Code units: How many code units in the string's internal encoding (UTF-16, UTF-8)?
3. Code points: How many Unicode code points (scalar values) does the string contain?
4. Grapheme clusters: How many user-perceived characters (glyphs the user would call "one letter") does the string contain?

These four counts can all be different for the same string.

A Concrete Example

Consider the string "café" followed by a family emoji "👨‍👩‍👧":

import unicodedata

s = "café"
# c-a-f-e+combining-accent OR c-a-f-é (precomposed)
# Decomposed NFC vs NFD matters here

nfc = unicodedata.normalize("NFC", "café")   # 4 code points
nfd = unicodedata.normalize("NFD", "café")   # 5 code points (e + combining acute)

len(nfc)  # 4
len(nfd)  # 5

# Family emoji: man + ZWJ + woman + ZWJ + girl = 8 code points, 1 grapheme
family = "👨‍👩‍👧"
len(family)   # 8 (code points: 3 emoji + 2 ZWJ)

const family = "👨‍👩‍👧";
family.length;          // 8 (UTF-16 code units: surrogates + ZWJ)
[...family].length;     // 8 (code points — spread operator)

// True grapheme count requires Intl.Segmenter (ES2022)
const segmenter = new Intl.Segmenter();
[...segmenter.segment(family)].length;  // 1 (one grapheme cluster!)

Byte Length

Byte length depends on the encoding:

s = "Hello, 世界 🌍"

len(s.encode("utf-8"))    # 19 bytes
len(s.encode("utf-16"))   # 24 bytes (with BOM: 26)
len(s.encode("utf-32"))   # 44 bytes (with BOM: 48)

For database column sizing (e.g., VARCHAR(100) in PostgreSQL with UTF-8), the limit is in bytes. A 100-character CJK string needs up to 300 bytes in UTF-8.

Grapheme Cluster Length

A grapheme cluster is what a user perceives as a single character. It may span multiple code points:

• Precomposed vs. decomposed: é can be one code point (U+00E9) or two (e + U+0301).
• Emoji sequences: 👨‍👩‍👧 = 3 base emoji + 2 ZWJ = 8 code points → 1 grapheme.
• Flag emoji: 🇺🇸 = 2 regional indicator letters → 1 grapheme.
• Keycap sequences: 1️⃣ = digit + VS16 + combining enclosing keycap → 1 grapheme.

# Python: grapheme cluster length
# Standard library has no built-in grapheme segmenter
# Use the 'grapheme' package
import grapheme

grapheme.length("👨‍👩‍👧")    # 1
grapheme.length("café")      # 4 (NFC) or 4 (NFD renders as 4 visible chars)
grapheme.length("🇺🇸")       # 1
grapheme.length("e\u0301")   # 1 (e + combining accent = 1 grapheme)

Database Implications

PostgreSQL char_length() counts code points; octet_length() counts bytes:

SELECT
  char_length('café'),           -- 4 (code points)
  octet_length('café'),          -- 5 (UTF-8 bytes: é = 2 bytes)
  length('café'),                 -- alias for char_length in text context
  char_length('👨‍👩‍👧'),     -- 8 (code points)
  octet_length('👨‍👩‍👧');    -- 25 (UTF-8 bytes)

Quick Facts