Reference

How software reads text

The surprising thing about strings is that “length” is not one number. The moment a word leaves the keyboard, the software measures it by a UNIT you chose — and the unit changes the answer. Here’s the model. Open the strings lab →

• A string is measured by a UNIT

  There is no single “length” of a word. A person counts LETTERS; a program can count grapheme clusters, Unicode code points, or UTF-8 bytes — and get three different numbers for the same word. Decide which unit you mean before you count, compare, or truncate.

• Grapheme cluster = what a person calls a letter

  A grapheme cluster is a base letter plus any marks that ride on it. Vietnamese “ệ” is one cluster even though it carries a circumflex and a dot. Use grapheme clusters (Intl.Segmenter) whenever you mean “letters” — for counting, cursor movement, or a character limit.

• Normalization: one letter, two encodings

  The same letter can be stored “composed” (NFC, one code point) or “decomposed” (NFD, a base letter plus combining marks). They look identical but have different lengths and won’t match with ==. Normalize (usually to NFC) before you compare or store.

• Plural is a CATEGORY, not “add an s”

  Software picks a word-form by CLDR category (one · few · many · other · …). English has two categories, Vietnamese has one, Polish has four, Arabic six. Never hard-code an English pluralize() — ask Intl.PluralRules for the user’s language.

• Casing and sorting are locale-specific

  Turkish “i” upper-cases to a dotted “İ”; Swedish sorts Ö after Z; German sorts it as O. A naïve A–Z uppercase or a raw string sort is wrong for real users — pass the locale to toLocaleUpperCase and Intl.Collator.

• Layout must survive translation

  Translations are often 30–50% longer (German especially) — and some scripts are taller. Never size a control to the English word; let text grow and wrap, and test with a long language before you ship.

The units & the tools

• grapheme clusterWhat a person calls a letter (base + its marks). Count with Intl.Segmenter. Use for character limits and “length in letters.”
• Unicode scalarA single code point. [...string] iterates these. A decomposed accented letter is several scalars — more than the eye sees.
• UTF-8 byteWhat’s stored/sent. ASCII = 1 byte; accented and CJK letters = 2–4. A byte limit is not a letter limit.
• NFC / NFDComposed vs decomposed encoding of the same text. Normalize before comparing so “é” always equals “é”.
• plural categoryone · few · many · other — Intl.PluralRules picks the right one for the language and number.
• collationLocale sort order. Intl.Collator, not a raw string sort — the alphabet is not universal.

And one the lab leaves for later: right-to-left

Arabic, Hebrew, Persian and Urdu read right-to-left, and their whole layout mirrors: a “next” arrow points left, the back button moves to the right, padding flips from the leading edge to the trailing edge. Numbers and embedded Latin words stay left-to-right inside the mirrored page. Write your layout in leading/trailing terms, never fixed left/right, and it flips for free. Read a string →