v0.2.0

[0.2.0] – 2026-04-28

Fixed

• tra compound consonant (roman_rules.json): moved tr into the
  consonants section so the standard loop generates the correct schwa form
  tra → त्र (not the halanta form त्र्). All derived matras (tri, traa,
  trai, …) are also correct.
• lRi independent vowel (loader.py): consonant + matra combinations are
  now inserted into the trie before independent vowels, so lRi → ऌ wins
  over the l + Ri-matra combination.
• }} closing-brace escape (tokenizer.py): }} now produces a literal
  } token, symmetric with the existing {{ escape.
• Leading-dot decimal tokenization (tokenizer.py): .5 is now tokenized
  as a single NUMBER token instead of a ROMAN . followed by a NUMBER
5, so the . is no longer converted to the Devanagari danda ।.
• README output example (README.md): corrected the expected output for
  "mero naam {Sanjeev} ho" from नाम to नाम् — naam without a trailing
  vowel correctly produces the halanta form.

Changed

• Converter public API (__init__.py): Converter is now importable
  directly from the package root (from nepali_unicoder import Converter).
• Mode validation (engine.py): Engine (and therefore Converter) now
  raises a clear ValueError immediately for an unrecognised mode argument
  (e.g. Converter(mode="Preeti")), instead of silently falling back to roman
  mode.

Improved (code quality)

• import re moved to module level (tokenizer.py): the lazy import re
  inside Tokenizer.tokenize() has been moved to the top of the module,
  following PEP 8 and making the dependency visible.
• Regex patterns compiled once (tokenizer.py): _RE_BLOCK and
  _RE_NUMBER are now class-level constants on Tokenizer, compiled once at
  class definition time instead of on every tokenize() call.
• PreetiLoader reads JSON once (loader.py): the preeti_rules.json
  file is now parsed once in PreetiLoader.__init__ and cached in
  self._data. Previously load() and get_post_rules() each called
  load_json_data() independently, causing the file to be read twice per
  Converter(mode="preeti") construction.
• except Exception replaced with specific exceptions (loader.py):
  _load_custom_mappings no longer silently swallows all errors. The
  bare except Exception: print(...) has been removed; json.JSONDecodeError,
  UnicodeDecodeError, and OSError now propagate naturally so callers can
  detect corrupt or unreadable word_maps.json files.
• print replaced with logging (loader.py): the "Warning: Rule file not found" message in load_json_data now uses logging.getLogger(__name__)
  (nepali_unicoder.loader) instead of print(), giving library users full
  control over verbosity via the standard logging framework.
• PEP 8 import order (loader.py): the local import
  from nepali_unicoder.trie import Trie is now grouped with all other imports,
  above the module-level _logger assignment.
• No-op try/except removed (loader.py): after the exception-handling
  fix, the try/except block in _load_custom_mappings only re-raised every
  branch. The block has been removed; the file is opened directly and exceptions
  propagate naturally. The os.path.exists() guard above it already handles
  the missing-file case.