Make lookup ignore accent marks.

[zzz-pokedex.git] / pokedex / lookup.py

diff --git a/pokedex/lookup.py b/pokedex/lookup.py
index c5a22ab..b31dce8 100644 (file)
--- a/pokedex/lookup.py
+++ b/pokedex/lookup.py
@@ -5,6 +5,7 @@ import pkg_resources
 import random
 import re
 import shutil
 import random
 import re
 import shutil

+import unicodedata

 
 from sqlalchemy.sql import func
 import whoosh
 
 from sqlalchemy.sql import func
 import whoosh


@@ -37,6 +38,28 @@ for cls in [
     ]:
     indexed_tables[cls.__tablename__] = cls
 
     ]:
     indexed_tables[cls.__tablename__] = cls
 

+def normalize(name):
+    """Strips irrelevant formatting junk from name input.
+
+    Specifically: everything is lowercased, and accents are removed.
+    """
+    # http://stackoverflow.com/questions/517923/what-is-the-best-way-to-remove-accents-in-a-python-unicode-string
+    # Makes sense to me.  Decompose by Unicode rules, then remove combining
+    # characters, then recombine.  I'm explicitly doing it this way instead of
+    # testing combining() because Korean characters apparently decompose!  But
+    # the results are considered letters, not combining characters, so testing
+    # for Mn works well, and combining them again makes them look right.
+    nkfd_form = unicodedata.normalize('NFKD', unicode(name))
+    name = u"".join(c for c in nkfd_form
+                    if unicodedata.category(c) != 'Mn')
+    name = unicodedata.normalize('NFC', name)
+
+    name = name.strip()
+    name = name.lower()
+
+    return name
+
+

 def open_index(directory=None, session=None, recreate=False):
     """Opens the whoosh index stored in the named directory and returns (index,
     speller).  If the index doesn't already exist, it will be created.
 def open_index(directory=None, session=None, recreate=False):
     """Opens the whoosh index stored in the named directory and returns (index,
     speller).  If the index doesn't already exist, it will be created.


@@ -94,6 +117,7 @@ def open_index(directory=None, session=None, recreate=False):
         table=whoosh.fields.ID(stored=True),
         row_id=whoosh.fields.ID(stored=True),
         language=whoosh.fields.STORED,
         table=whoosh.fields.ID(stored=True),
         row_id=whoosh.fields.ID(stored=True),
         language=whoosh.fields.STORED,

+        iso3166=whoosh.fields.STORED,

         display_name=whoosh.fields.STORED,  # non-lowercased name
         forme_name=whoosh.fields.ID,
     )
         display_name=whoosh.fields.STORED,  # non-lowercased name
         forme_name=whoosh.fields.ID,
     )


@@ -117,18 +141,20 @@ def open_index(directory=None, session=None, recreate=False):
                            row_id=unicode(row.id),
                            forme_name=u'XXX')
 
                            row_id=unicode(row.id),
                            forme_name=u'XXX')
 

-            def add(name, language, score):
-                writer.add_document(name=name.lower(), display_name=name,
+            def add(name, language, iso3166, score):
+                normalized_name = normalize(name)
+                writer.add_document(name=normalized_name, display_name=name,

                                     language=language,
                                     language=language,

+                                    iso3166=iso3166,

                                     **row_key)
                                     **row_key)

-                speller_entries.append((name.lower(), score))
+                speller_entries.append((normalized_name, score))

 
             # If this is a form, mark it as such
             if getattr(row, 'forme_base_pokemon_id', None):
                 row_key['forme_name'] = row.forme_name
 
             name = row.name
 
             # If this is a form, mark it as such
             if getattr(row, 'forme_base_pokemon_id', None):
                 row_key['forme_name'] = row.forme_name
 
             name = row.name

-            add(name, None, 1)
+            add(name, None, u'us', 1)

 
             # Pokemon also get other languages
             for foreign_name in getattr(row, 'foreign_names', []):
 
             # Pokemon also get other languages
             for foreign_name in getattr(row, 'foreign_names', []):


@@ -138,12 +164,14 @@ def open_index(directory=None, session=None, recreate=False):
                     # no point and it makes spell results confusing
                     continue
 
                     # no point and it makes spell results confusing
                     continue
 

-                add(moonspeak, foreign_name.language.name, 3)
+                add(moonspeak, foreign_name.language.name,
+                               foreign_name.language.iso3166,
+                               3)

 
                 # Add Roomaji too
                 if foreign_name.language.name == 'Japanese':
                     roomaji = romanize(foreign_name.name)
 
                 # Add Roomaji too
                 if foreign_name.language.name == 'Japanese':
                     roomaji = romanize(foreign_name.name)

-                    add(roomaji, u'Roomaji', 8)
+                    add(roomaji, u'Roomaji', u'jp', 8)

 
     writer.commit()
 
 
     writer.commit()
 


@@ -176,7 +204,7 @@ class LanguageWeighting(whoosh.scoring.Weighting):
 rx_is_number = re.compile('^\d+$')
 
 LookupResult = namedtuple('LookupResult',
 rx_is_number = re.compile('^\d+$')
 
 LookupResult = namedtuple('LookupResult',

-                          ['object', 'name', 'language', 'exact'])
+                          ['object', 'name', 'language', 'iso3166', 'exact'])

 
 def _parse_table_name(name):
     """Takes a singular table name, table name, or table object and returns the
 
 def _parse_table_name(name):
     """Takes a singular table name, table name, or table object and returns the


@@ -215,6 +243,7 @@ def _whoosh_records_to_results(records, session, exact=True):
         results.append(LookupResult(object=obj,
                                     name=record['display_name'],
                                     language=record['language'],
         results.append(LookupResult(object=obj,
                                     name=record['display_name'],
                                     language=record['language'],

+                                    iso3166=record['iso3166'],

                                     exact=exact))
 
     return results
                                     exact=exact))
 
     return results


@@ -223,10 +252,11 @@ def _whoosh_records_to_results(records, session, exact=True):
 def lookup(input, valid_types=[], session=None, indices=None, exact_only=False):
     """Attempts to find some sort of object, given a database session and name.
 
 def lookup(input, valid_types=[], session=None, indices=None, exact_only=False):
     """Attempts to find some sort of object, given a database session and name.
 

-    Returns a list of named (object, name, language, exact) tuples.  `object`
-    is a database object, `name` is the name under which the object was found,
-    `language` is the name of the language in which the name was found, and
-    `exact` is True iff this was an exact match.
+    Returns a list of named (object, name, language, iso3166, exact) tuples.
+    `object` is a database object, `name` is the name under which the object
+    was found, `language` and `iso3166` are the name and country code of the
+    language in which the name was found, and `exact` is True iff this was an
+    exact match.

 
     This function currently ONLY does fuzzy matching if there are no exact
     matches.
 
     This function currently ONLY does fuzzy matching if there are no exact
     matches.


@@ -279,7 +309,7 @@ def lookup(input, valid_types=[], session=None, indices=None, exact_only=False):
     else:
         index, speller = open_index()
 
     else:
         index, speller = open_index()
 

-    name = unicode(input).strip().lower()
+    name = normalize(input)

     exact = True
     form = None
 
     exact = True
     form = None
 


@@ -326,7 +356,9 @@ def lookup(input, valid_types=[], session=None, indices=None, exact_only=False):
     type_terms = []
     for valid_type in valid_types:
         table_name = _parse_table_name(valid_type)
     type_terms = []
     for valid_type in valid_types:
         table_name = _parse_table_name(valid_type)

-        type_terms.append(whoosh.query.Term(u'table', table_name))
+        if table_name:
+            # Quietly ignore bogus valid_types; more likely to DTRT
+            type_terms.append(whoosh.query.Term(u'table', table_name))

 
     if type_terms:
         query = query & whoosh.query.Or(type_terms)
 
     if type_terms:
         query = query & whoosh.query.Or(type_terms)


@@ -371,11 +403,17 @@ def random_lookup(valid_types=[], session=None, indices=None):
             tables.append(indexed_tables[table_name])
 
     if not tables:
             tables.append(indexed_tables[table_name])
 
     if not tables:

+        # n.b.: It's possible we got a list of valid_types and none of them
+        # were valid, but this function is guaranteed to return *something*, so
+        # it politely selects from the entire index isntead

         tables = indexed_tables.values()
 
     # Rather than create an array of many hundred items and pick randomly from
     # it, just pick a number up to the total number of potential items, then
         tables = indexed_tables.values()
 
     # Rather than create an array of many hundred items and pick randomly from
     # it, just pick a number up to the total number of potential items, then

-    # pick randomly from that, and partition the whole range into chunks
+    # pick randomly from that, and partition the whole range into chunks.
+    # This also avoids the slight problem that the index contains more rows
+    # (for languages) for some items than others.
+    # XXX ought to cache this (in the index?) if possible

     total = 0
     partitions = []
     for table in tables:
     total = 0
     partitions = []
     for table in tables:


@@ -408,7 +446,7 @@ def prefix_lookup(prefix, session=None, indices=None):
     else:
         index, speller = open_index()
 
     else:
         index, speller = open_index()
 

-    query = whoosh.query.Prefix(u'name', prefix)
+    query = whoosh.query.Prefix(u'name', normalize(prefix))

 
     searcher = index.searcher()
     searcher.weighting = LanguageWeighting()
 
     searcher = index.searcher()
     searcher.weighting = LanguageWeighting()