The no-break space does the quiet work

The no-break space (U+00A0) looks exactly like a space and does one thing a space does not: it refuses to end a line. Typeset text is full of pairs that must not be separated — 21 September, Prof. Oksana, 5 % — and the only way to keep them together is a character the reader never sees. In this article a no-break space is shown as ~, the same convention the rules reference uses.

Where it belongs

More than half of the engine’s rules exist to place this one character:

• A number and the short word after it. 15~people, 21~September, 5~%. A line that ends on 15 and begins on people makes the reader re-parse the number (space.number).
• An initial and the surname. K.~Słoboda — an initial stranded at a line end reads as a sentence that stopped (space.initial), and a title stays with the name: Prof.~Dr.~Oksana (space.title).
• The parts of an abbreviation. German z.~B., Ukrainian т.~д., English Fig.~2 and p.~14 (space.abbreviation).
• One-letter words. Polish and Ukrainian typography does not let i, w, z end a line; English keeps A~cat together. Which words count comes from the language pack (space.orphan).
• Percent and euro, by language. German, French, Czech and fourteen other languages set 5~% and most of them 49~€; English 5% is left alone (space.percent).
• After © (©~2026 Studio), after § and ¶ (§~5), before a sentence dash (word~— word, so a line never opens with the dash), before →, and after the dialogue dash in the languages that space it.

Why CSS cannot do this

text-wrap: pretty fixes orphan words at a paragraph’s end and white-space: nowrap can pin one phrase, but neither knows that Polish forbids a trailing i, that French spaces % and English does not, or that Fig. belongs to its number. These are language conventions, not layout preferences, and they live in the text itself — which also means they survive into RSS, email, and anywhere else your CSS does not follow.

One boundary worth knowing: French famously sets a space before : ; ! ?. The engine does not insert one where the writer typed none — the French guides disagree with each other on which space and where, and the sources registry would rather record the dispute than guess. The optional space.before-punctuation rule, which strips stray spaces before punctuation, does keep the French ones the writer typed.

The nbsp option

A raw U+00A0 looks like a space in every diff and editor. In the HTML mode the nbsp option chooses how a changed block writes it: as   (the default, visible in source) or as the bare character; see the API reference.

Try it

Paste On 21 September, Prof. K. Słoboda paid 5 % into the playground and tick Mark no-break spaces to see every one the engine placed.

Something wrong or unclear here? Report an issue.