@atthebunga/gurmukhi-search-utils

Building blocks for searching Gurmukhi text, extracted from the search used by search.atthebunga.com. Pure TypeScript with no dependencies.

npm install @atthebunga/gurmukhi-search-utils

Spelling-variant folding

import { foldGurmukhi } from '@atthebunga/gurmukhi-search-utils';

foldGurmukhi('ਟੀਕਾ') === foldGurmukhi('ਤੀਕਾ');   // retroflex ↔ dental
foldGurmukhi('ਖ਼ਾਲਸਾ') === foldGurmukhi('ਖਾਲਸਾ'); // nukta
foldGurmukhi('ਸਭੁ') === foldGurmukhi('ਸਭ');      // final short matra

foldGurmukhi reduces a string to a lossy key that spelling variants share:

• retroflex consonants fold to dental, and ੜ to ਰ
• nukta, vowel length and nasal signs fold away
• addak and visarga are dropped
• one grammatical final short matra is dropped

Aspiration and kanna are kept, because they distinguish words. Fold both the stored text and the query, then match on prefix. Non-Gurmukhi characters are ignored.

The key is stable. It is identical to words.search_fold in gurmukhi-kosh, so changing it means rebuilding that column.

First-letter search

import { firstLetters, matchesFirstLetters, tokenize } from '@atthebunga/gurmukhi-search-utils';

firstLetters('ਸਤਿ ਨਾਮੁ ਕਰਤਾ ਪੁਰਖੁ ॥੧॥');              // 'ਸਨਕਪ'
matchesFirstLetters('ਸਤਿ ਨਾਮੁ ਕਰਤਾ ਪੁਰਖੁ ॥੧॥', 'ਸਨਕ');  // true
tokenize('ਗਉੜੀ ਮਹਲਾ ੫ ॥');                          // ['ਗਉੜੀ', 'ਮਹਲਾ']

tokenize is the word split the search index uses: it drops dandas, numerals and pure punctuation. matchesFirstLetters follows the same rule as the search_first_letters database function: word i must start with letter i, counting from the start of the line. Client-side and database results therefore agree.

Pattern language

The same syntax <gurmukhi-input pattern-mode> leaves unconverted while you type:

| Syntax | Matches | |---|---| | ਸਤਿ | the literal text | | _ | any one character | | * | any run of characters | | {velar} | any character in a category (see listCategories()) | | (ਮੁਖਿ\|ਮਤਿ) | either alternative |

A pattern that starts with a lone vowel sign (ਾ, {ooraVowels}) gets an implied base letter in front of it.

import { compilePattern, patternToRegExp } from '@atthebunga/gurmukhi-search-utils';

compilePattern('{velar}ਾ*');          // { ok: true, regex: '[ਕਖਗਘਙ]ਾ.*' }  — for Postgres ~
patternToRegExp('ਗੁਰ(ਮੁਖਿ|ਮਤਿ)').test('ਗੁਰਮਤਿ');           // true — JavaScript
patternToRegExp('ਸਤਿ', { anchored: true }).test('ਸਤਿਗੁਰ'); // false

Characters and syllables

The package also exports:

• the character sets CONSONANTS, VOWEL_CARRIERS, VOWEL_DIACRITICS and NASAL_MARKS
• the pattern categories, as CHAR_CATEGORIES
• parseSyllables(word), which splits a word into akhar syllables: a base letter plus its vowel sign and marks

Related

To match on pronunciation, such as voice search or romanized queries like "waheguru", use @atthebunga/gurmukhi-phonetics (phoneticKey and romanPhoneticKey).

License

MIT