lipimala — Indic Script Converter (Dart) #

A high-fidelity Dart library that separates two concepts that cannot honestly be merged into a single Brahmic string:

1. Exact source exact round-tripness — Code-point-equivalent recovery of original Latin input, including capitalization, NFC/NFD form, alias choices, punctuation, and combining-mark order.
2. Rendered transliteration view — Visual output in Devanagari, Gujarati, Plain English, or Hunterian.

Because rendered Indic scripts cannot natively represent Latin capitalization or lossy alias selections (e.g., distinguishing Kṛṣṇa vs kṛṣṇa or x vs ḫ), this package provides two robust strategies for exact source recovery:

1. TransliterationResult (Envelope Pattern): Keeps metadata alongside the rendered view (ideal for JSON APIs and databases).
2. Unicode Tag Trailer (Embedded Metadata Pattern): Appends an invisible, checksummed Unicode tag sequence directly into the string output.

Installation #

Add the package to your pubspec.yaml or run:

dart pub add lipimala

Package link on pub.dev: pub.dev/packages/lipimala

This package uses unorm_dart for Unicode 17 NFC/NFD normalization and unicode for complete Unicode 17 Mn/Mc/Me mark classification.

Bulk String List Transliteration #

Convert a list of text values in a single call without manual iteration logic:

import 'package:lipimala/lipimala.dart';

final items = ['Kṛṣṇa', 'Rāma', 'jñāna'];

// Bulk convert to Devanagari script strings
final devaList = items.toDevanagariFromIast(); 
// -> ['कृष्ण', 'राम', 'ज्ञान']

// Bulk convert to Gujarati script strings
final gujrList = items.toGujaratiFromIast(); 
// -> ['કૃષ્ણ', 'રામ', 'જ્ઞાન']

// Bulk convert to Plain English strings
final plainList = items.toPlainEnglishFromIast(); 
// -> ['Krishna', 'Ram', 'gyan']

// Bulk convert to TransliterationResult envelopes
final envList = items.toDevanagari();

Usage & Exact Recovery Modes #

Strategy 1: Envelope Pattern (TransliterationResult) #

Recommended for database storage, network transport, and clean UI separation.

import 'dart:convert';
import 'package:lipimala/transliteration_result.dart';

void main() {
  const source = 'Kṛṣṇa ā́tman ḷa';

  // 1. Transliterate to an envelope object
  final result = source.toDevanagari();

  print(result.rendered);        // Output: कृष्ण आ॑त्मन् ऌअ
  print(result.restoreOriginal()); // Output: Kṛṣṇa ā́tman ḷa (Exact match)

  // 2. JSON Serialization
  final jsonString = jsonEncode(result.toJson());
  final restoredResult = TransliterationResult.fromJson(
    Map<String, Object?>.from(jsonDecode(jsonString) as Map),
  );

  assert(restoredResult.restoreOriginal() == source);
}

Strategy 2: Embedded Metadata Pattern (In-String Unicode Tags) #

Appends a hidden, checksummed Unicode-tag sequence containing original UTF-16 code units directly to the Brahmic string. The visible Brahmic text remains visually unchanged.

import 'package:lipimala/brahmic_to_latn_iast.dart';
import 'package:lipimala/latn_iast_to_deva.dart';
import 'package:lipimala/latn_iast_to_gujr.dart';

void main() {
  const source = 'Kṛṣṇa / Kr̥ṣṇa / ḫāna';

  // --- Devanagari ---
  final devanagari = source.toDevanagariFromIast(
    options: const IastToDevanagariOptions(
      embedExactSourceMetadata: true,
    ),
  );

  // Exact recovery using the invisible trailer
  final exactFromDevanagari = devanagari.toExactIastFromDevanagari();
  assert(exactFromDevanagari == source);

  // --- Gujarati ---
  final gujarati = source.toGujaratiFromIast(
    options: const IastToGujaratiOptions(
      embedExactSourceMetadata: true,
    ),
  );

  final exactFromGujarati = gujarati.toExactIastFromGujarati();
  assert(exactFromGujarati == source);
}

Reverse Transliteration APIs

When using embedded metadata, the reverse functions handle recovery gracefully:

⚠️ Transport Warning: Embedded Unicode Tag characters are default-ignorable. Some sanitizers, search indexes, clipboard filters, or legacy databases may strip them. For unconditional persistence, retain the TransliterationResult envelope or store original keys in dedicated columns.

Core Guarantees & Engine Features #

1. Unicode Normalization #

Parsing defaults to NFD and rendered output defaults to NFC. Both are configurable via UnicodeNormalizationForm.

2. Combining Mark Integrity #

The parser recognizes all Unicode 17 combining-mark general categories (Mn, Mc, and Me) beyond standard diacritics (U+0300–U+036F). Residual or unmapped combining marks (e.g., combining underline on a̱) are retained on rendered outputs rather than silently dropped.

3. Extended-Letter Policy (Documented Many-to-One) #

Approximated extended Latin mappings are explicitly defined in the iso15919Core and extendedIndic profiles. strictIast preserves or diagnoses non-IAST input instead of silently applying these extensions.

4. Dotted-Circle Carriers for Orphaned Nasals #

Anusvāra and candrabindu marks (ṃ, ṁ, m̐) require a base syllable. When occurring word-initially, renderers introduce a dotted-circle carrier (U+25CC):

• ṁaṅgala → ◌ंअङ्गल / ◌ંઅઙ્ગલ
• m̐tra → ◌ᳪत्र / ◌ᳪત્ર

Post-vocalic forms map normally (aham̐ → अहᳪ). Unsupported nasal extensions (e.g., ṃ̄) preserve residual marks on script nasals (◌ं̄).

5. Vedic Accent Preservation #

Encoded Vedic accent ranges are preserved and mapped appropriately:

• U+0951–U+0952
• U+A8E0–U+A8FF
• U+1CD0–U+1CFF

Established Latin accent aliases (e.g., Yajurveda fixtures) map directly to Brahmic Vedic marks. Unmapped marks remain encoded and exact round-trip.

Canonical reverse transliteration renders script-order Vedic marks back into Latin syllable order: वः॑ → váḥ, जुष्टं॑ → juṣṭáṃ, and अहᳪ → aham̐. Script dandas reverse to the ASCII corpus convention (। → |, ॥ / ।। → ||).

Transliteration Profiles #

The default profile for forward Devanagari, Gujarati, and plain transcription is extendedIndic. Use strictIast explicitly when you want classical Sanskrit IAST boundaries.

• strictIast: Strictly limits input processing to classical Sanskrit IAST inventory.
• iso15919Core: Table-driven ISO-style inventory.
• extendedIndic: Practical extensions, regional transcriptions, and common input aliases.
• hunterian: Explicitly lossy display view (Hunterian English rendering). Exact recovery remains available via envelope metadata.

Devanagari ↔ Gujarati Direct Converter #

This package includes a direct converter between Devanagari and Gujarati.

Canonical visible conversion #

import 'package:lipimala/deva_gujr_converter.dart';

void main() {
  print('कृष्ण'.toCanonicalGujaratiFromDevanagari()); // કૃષ્ણ
  print('કૃષ્ણ'.toCanonicalDevanagariFromGujarati()); // कृष्ण
}

Exact exact round-trip round trip #

The visible Gujarati and Devanagari repertoires are not one-to-one. Therefore, exact round-trip recovery uses a checksummed Unicode-tag trailer.

import 'package:lipimala/deva_gujr_converter.dart';

void main() {
  const source = 'ऄ ऎ ऍ ॲ ऒ ऑ ॵ ळ ऴ ग़ ॻ ड़ ॸ ॾ';
  const options = IndicScriptConversionOptions(
    embedExactSourceMetadata: true,
  );

  final taggedGujarati = source.toCanonicalGujaratiFromDevanagari(
    options: options,
  );

  assert(taggedGujarati.toExactDevanagariFromGujarati() == source);
}

Smart exact-or-canonical APIs #

These recover a correctly typed exact source trailer if present, and otherwise fall back to canonical visible conversion:

• toDevanagariFromGujarati()
• toGujaratiFromDevanagari()

Strict exact APIs throw when typed metadata is absent or damaged:

• toExactDevanagariFromGujarati()
• toExactGujaratiFromDevanagari()

Standards Boundary & Disclaimers #

1. ISO 15919: ISO 15919 is a normative, licensed publication. This package provides an iso15919Core profile based on common table definitions, but does not claim complete ISO conformance without independent audit against official ISO documentation.
2. Hunterian & English Displays: Renderings like Hunterian or plain English are inherently many-to-one and non-injective. Original source recovery is guaranteed only when retaining the envelope or metadata trailer.

Testing & Verification #

Run the primary test suite:

dart pub get
dart test

Direct Script Execution & Corpus Workflows #

For direct execution workflows, shared edge-case corpora are maintained under tools/latn_iast_transliteration_verification/:

# Individual script testing
dart tools/latn_iast_transliteration_verification/latn_iast_to_deva_test.dart > latn_iast_to_deva_output.txt
dart tools/latn_iast_transliteration_verification/latn_iast_to_gujr_test.dart > latn_iast_to_gujr_output.txt
dart tools/latn_iast_transliteration_verification/latn_iast_transcription_test.dart > latn_iast_transcription_output.txt
dart tools/latn_iast_transliteration_verification/deva_to_latn_iast_test.dart > deva_to_latn_iast_output.txt
dart tools/latn_iast_transliteration_verification/gujr_to_latn_iast_test.dart > gujr_to_latn_iast_output.txt
dart tools/latn_iast_transliteration_verification/deva_to_gujr_test.dart > deva_to_gujr_output.txt
dart tools/latn_iast_transliteration_verification/gujr_to_deva_test.dart > gujr_to_deva_output.txt

# Combined console suite execution
dart tools/latn_iast_transliteration_verification/latn_iast_to_deva_gujr_transcription_suite_test.dart

The five redirected output files are JSON Lines. Each physical line is a complete JSON object with at least type, source, and result, so embedded quotes and embedded newlines in samples remain machine-readable. The forward Deva/Gujr sample streams use the default extendedIndic profile so extended corpus entries render as script instead of mixed Latin passthrough; strict/ISO/extended behavior is still listed explicitly in the profile records.

The Deva/Gujr reverse runners use canonical IAST because their corpora are normal visible script strings without embedded source metadata. Exact source recovery is available for tagged outputs through toExactIastFromDevanagari() and toExactIastFromGujarati().

Before submitting changes or releases, ensure strict formatting and static code checks pass:

dart format --set-exit-if-changed .
dart analyze
dart test