fovea-core

The text transform behind fovea: splits prose into words and works out how many leading characters of each to emphasize.

npm install fovea-core

import { toHtml, tokenize } from "fovea-core";

toHtml("Fixation points guide the eye");
// '<b class="fovea-stem">Fixa</b>tion <b class="fovea-stem">poi</b>nts …'

Tokenizing and rendering are separate steps. tokenize returns { text, kind, stem }[], so one pass over the text can feed a terminal, a DOM, or a Markdown file:

for (const token of tokenize(source)) {
  if (token.kind === "word") emphasize(token.text.slice(0, token.stem));
}

renderAnsi, renderHtml and renderMarkdown ship with it. Anything else is a fold over the same tokens.

What it leaves alone

Emphasizing src/index.ts or --max-stem is what makes naive implementations unreadable, so code-shaped chunks are detected before word segmentation and passed through verbatim: paths, flags, URLs, camelCase, CONSTANT_CASE, and anything mixing letters with digits.

Word breaking goes through Intl.Segmenter rather than an ASCII regex, so accented and non-Latin alphabetic scripts work properly. Han, Kana, Hangul and Thai are skipped on purpose — there is no leading stem to emphasize in a logographic or abugida script.

Options

| Option | Default | Meaning | | --- | --- | --- | | ratio | 0.4 | fraction of a word to emphasize | | maxStem | 5 | cap on emphasized characters | | minWordLength | 2 | leave shorter words alone | | skipStopwords | false | skip common function words | | locale | "en" | locale for word segmentation | | skip | — | predicate for words to leave alone |

Words of three characters or fewer get one emphasized character and words of four or five get two, regardless of ratio; a proportional rule produces fractional stems at those lengths and rounding either way looks wrong beside its neighbours.

For streaming or ANSI-coloured input, use fovea-stream instead.

MIT