Source code for indic_transliteration.detect

# -*- coding: utf-8 -*-
"""
    detect
    ~~~~~~

    Code for automatically detecting a transliteration scheme.

    :license: MIT and BSD

Example usage:

::

    from indic_transliteration import detect
    detect.detect('pitRRIn') == Scheme.ITRANS
    detect.detect('pitRRn') == Scheme.HK

When handling a Sanskrit string, it's almost always best to explicitly
state its transliteration scheme. This avoids embarrassing errors with
words like ``pitRRIn``. But most of the time, it's possible to infer the
encoding from the text itself.

``detect.py`` automatically detects a string's transliteration scheme:

::

    detect('pitRRIn') == Scheme.ITRANS
    detect('pitRRn') == Scheme.HK
    detect('pitFn') == Scheme.SLP1
    detect('पितॄन्') == Scheme.Devanagari
    detect('পিতৄন্') == Scheme.Bengali

Supported schemes
-----------------

All schemes are attributes on the ``Scheme`` class. You can also just
use the scheme name:

::

    Scheme.IAST == 'IAST'
    Scheme.Devanagari == 'Devanagari'

Scripts:

-  Bengali (``'Bengali'``)
-  Devanagari (``'Devanagari'``)
-  Gujarati (``'Gujarati'``)
-  Gurmukhi (``'Gurmukhi'``)
-  Kannada (``'Kannada'``)
-  Malayalam (``'Malayalam'``)
-  Oriya (``'Oriya'``)
-  Tamil (``'Tamil'``)
-  Telugu (``'Telugu'``)

Romanizations:

-  Harvard-Kyoto (``'HK'``)
-  IAST (``'IAST'``)
-  ITRANS (``'ITRANS'``)
-  Kolkata (``'Kolkata'``)
-  SLP1 (``'SLP1'``)
-  Velthuis (``'Velthuis'``)

"""

import re

#: Scheme data. This is split into separate classes, but here it's DRY.
import sys

SCHEMES = [
  ('Bengali', 0x0980),
  ('Devanagari', 0x0900),
  ('Gujarati', 0x0a80),
  ('Gurmukhi', 0x0a00),
  ('Kannada', 0x0c80),
  ('Malayalam', 0x0d00),
  ('Oriya', 0x0b00),
  ('Tamil', 0x0b80),
  ('Telugu', 0x0c00),
  ('HK', None),
  ('IAST', None),
  ('ITRANS', None),
  ('Kolkata', None),
  ('SLP1', None),
  ('Velthuis', None),
]

#: Start of the Devanagari block.
BRAHMIC_FIRST_CODE_POINT = 0x0900

#: End of the Malayalam block.
BRAHMIC_LAST_CODE_POINT = 0x0d7f

#: Schemes sorted by Unicode code point. Ignore schemes with none defined.
BLOCKS = sorted([x for x in SCHEMES if x[-1]], key=lambda x: -x[1])

#: Enum for Sanskrit schemes.
Scheme = type('Enum', (), {name: name for name, code in SCHEMES})


class Regex:
  #: Match on special Roman characters
  IAST_OR_KOLKATA_ONLY = re.compile(u'[āīūṛṝḷḹēōṃḥṅñṭḍṇśṣḻ]')

  #: Match on chars shared by ITRANS and Velthuis
  ITRANS_OR_VELTHUIS_ONLY = re.compile(u'aa|ii|uu|~n')

  #: Match on ITRANS-only
  ITRANS_ONLY = re.compile(u'ee|oo|\^[iI]|RR[iI]|L[iI]|'
                           u'~N|N\^|Ch|chh|JN|sh|Sh|\\.a')

  #: Match on Kolkata-specific Roman characters
  KOLKATA_ONLY = re.compile(u'[ēō]')

  #: Match on SLP1-only characters and bigrams
  SLP1_ONLY = re.compile(u'[fFxXEOCYwWqQPB]|kz|Nk|Ng|tT|dD|Sc|Sn|'
                         u'[aAiIuUfFxXeEoO]R|'
                         u'G[yr]|(\\W|^)G')

  #: Match on Velthuis-only characters
  VELTHUIS_ONLY = re.compile(u'\\.[mhnrltds]|"n|~s')


# noinspection PyUnresolvedReferences
def detect(text):
  """Detect the input's transliteration scheme.

    :param text: some text data, either a `unicode` or a `str` encoded
                 in UTF-8.
    """
  if sys.version_info < (3, 0):
    # Verify encoding
    try:
      text = text.decode('utf-8')
    except UnicodeError:
      pass

  # Brahmic schemes are all within a specific range of code points.
  for L in text:
    code = ord(L)
    if code >= BRAHMIC_FIRST_CODE_POINT:
      for name, start_code in BLOCKS:
        if start_code <= code <= BRAHMIC_LAST_CODE_POINT:
          return name

  # Romanizations
  if Regex.IAST_OR_KOLKATA_ONLY.search(text):
    if Regex.KOLKATA_ONLY.search(text):
      return Scheme.Kolkata
    else:
      return Scheme.IAST

  if Regex.ITRANS_ONLY.search(text):
    return Scheme.ITRANS

  if Regex.SLP1_ONLY.search(text):
    return Scheme.SLP1

  if Regex.VELTHUIS_ONLY.search(text):
    return Scheme.Velthuis

  if Regex.ITRANS_OR_VELTHUIS_ONLY.search(text):
    return Scheme.ITRANS

  return Scheme.HK