Fixers

Each fixer implements Fixer — fix(Run $run): Run — and can be used alone, in a Typographer of your own or through a preset. They are listed here in the order the presets run them; the order matters: quotes are made before French spaces are placed around them, and hyphenation comes last.

Ellipsis

Three full stops close together or spaced by single spaces. Four stops are an ellipsis and a full stop.

Dimension

The letter x or X, with or without a space on either side, between a number and a number. The first number may carry a unit — mm, cm, dm, m, km, in, ft, px, pt, em, rem — or a foot or inch mark. A chain such as 10x20x30 is fixed throughout. A hexadecimal number (0x1F), a number after a letter (A4x5) or after a decimal mark is left alone.

Dash

A phone number written in two groups, 555-1234, looks like a range and gets an en dash; write it with a no-break hyphen (U+2011) or leave the fixer out where such numbers are common.

Smart quotes

A double quote opens after the start of the block, a space or an opening mark — ( [ { « “ ‘ „ ‹ - – — / > or another quote — when a character follows it; otherwise it closes. > counts as an opening mark because a tag escaped as text, <b>"word", ends with it, as does a marker of quoted text. A single quote is an inner quotation mark only when it opens in that position and a matching single quote closes before a space, a punctuation mark or <; every other single quote is an apostrophe, left to CurlyQuote. The marks come from a QuoteStyle, English by default:

new SmartQuotes(new QuoteStyle('«', '»', '‹', '›'));

Known limits: rock 'n' roll is read as an inner quotation of n; a double quote inside a double quotation is a quotation of its own, never an inner one.

Curly quote

A plain replacement of ' by ’: placed after SmartQuotes, it turns the apostrophes that fixer left. Used alone, it turns every straight single quote.

No space before a comma

Spaces before a comma are removed; when a letter follows the comma directly, a space is put after it. 1,5 and a ,5 keep their digits together.

French spacing

SpaceBeforePunctuation writes the space before high punctuation that ends a word — whatever the author typed: none, a space, a no-break space, a thin space — and inside guillemets. See Locales for the rules of France and of Canada. The colon of 12:30 or of https:// is not followed by a space and is left alone; ?! takes a single space, before the first mark.

Trademark

(tm) and (r), in either case, are attached to the word before them. (c) and (p) become © and ℗ only when a year follows, held to it by a no-break space, so the item (c) of a list stays as written; © 2026 gets its no-break space too.

Unit

Only an existing space is changed — the fixer never adds one. The units are listed in Unit::UNITS: SI units and their common multiples, data sizes, typographic units, speeds, temperatures, percentages and currency signs. A word that merely begins like a unit (5 mice) is left alone.

Hyphen

Hyphen takes a Hyphenator and a minimum length, five characters by default. A word is a run of letters and combining marks; digits, emoji and symbols end it. A word that already holds a soft hyphen is left as its author wrote it, so fixing twice gives the same text. See Hyphenation.

Word break

WordBreak puts a zero-width space (U+200B) between the words of each run of Thai characters, at the line-break opportunities ICU finds with its Thai dictionary, never before the repetition mark ๆ. Other scripts are left alone, and a zero-width space already written ends a run, so fixing twice gives the same text. See Hyphenation.

Writing a fixer

final readonly class NoBreakAfterShortWords implements Fixer
{
    public function fix(Run $run): Run
    {
        return $run->replace(static fn (string $text): string => preg_replace('/\b(\p{L}{1,2}) /u', "\$1\u{A0}", $text) ?? $text);
    }
}

Run::replace() gives the fix the part between its neighbouring characters and keeps the result only if they are unchanged; Run::map() gives the part and its neighbours separately; Run::characters() gives every character of the run at once. A fixer must only rewrite text: what it returns is written into text nodes, so < stays a character.