Make lookup ignore accent marks.

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

diff --git a/pokedex/lookup.py b/pokedex/lookup.py
index 69eedc4..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.


@@ -119,11 +142,12 @@ def open_index(directory=None, session=None, recreate=False):
                            forme_name=u'XXX')
 
             def add(name, language, iso3166, score):
                            forme_name=u'XXX')
 
             def add(name, language, iso3166, score):

-                writer.add_document(name=name.lower(), display_name=name,
+                normalized_name = normalize(name)
+                writer.add_document(name=normalized_name, display_name=name,

                                     language=language,
                                     iso3166=iso3166,
                                     **row_key)
                                     language=language,
                                     iso3166=iso3166,
                                     **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):
 
             # If this is a form, mark it as such
             if getattr(row, 'forme_base_pokemon_id', None):


@@ -285,7 +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
 


@@ -332,7 +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)


@@ -377,11 +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:


@@ -414,7 +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.lower())
+    query = whoosh.query.Prefix(u'name', normalize(prefix))

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