postcodes: exclude seen places later

[nominatim.git] / nominatim / api / search / token_assignment.py

diff --git a/nominatim/api/search/token_assignment.py b/nominatim/api/search/token_assignment.py
index 0ae2cd4306c9e4ca1c78cbd9acd9c0363aa7b9c8..ca907b797dfe53b24e48d1602cc19ca97ff0fc2c 100644 (file)
--- a/nominatim/api/search/token_assignment.py
+++ b/nominatim/api/search/token_assignment.py
@@ -46,7 +46,7 @@ class TokenAssignment: # pylint: disable=too-many-instance-attributes
     housenumber: Optional[qmod.TokenRange] = None
     postcode: Optional[qmod.TokenRange] = None
     country: Optional[qmod.TokenRange] = None
-    category: Optional[qmod.TokenRange] = None
+    near_item: Optional[qmod.TokenRange] = None
     qualifier: Optional[qmod.TokenRange] = None
 
 
@@ -64,15 +64,15 @@ class TokenAssignment: # pylint: disable=too-many-instance-attributes
                 out.postcode = token.trange
             elif token.ttype == qmod.TokenType.COUNTRY:
                 out.country = token.trange
-            elif token.ttype == qmod.TokenType.CATEGORY:
-                out.category = token.trange
+            elif token.ttype == qmod.TokenType.NEAR_ITEM:
+                out.near_item = token.trange
             elif token.ttype == qmod.TokenType.QUALIFIER:
                 out.qualifier = token.trange
         return out
 
 
 class _TokenSequence:
-    """ Working state used to put together the token assignements.
+    """ Working state used to put together the token assignments.
 
         Represents an intermediate state while traversing the tokenized
         query.
@@ -109,7 +109,7 @@ class _TokenSequence:
         """
         # Country and category must be the final term for left-to-right
         return len(self.seq) > 1 and \
-               self.seq[-1].ttype in (qmod.TokenType.COUNTRY, qmod.TokenType.CATEGORY)
+               self.seq[-1].ttype in (qmod.TokenType.COUNTRY, qmod.TokenType.NEAR_ITEM)
 
 
     def appendable(self, ttype: qmod.TokenType) -> Optional[int]:
@@ -132,6 +132,11 @@ class _TokenSequence:
 
         # Name tokens are always acceptable and don't change direction
         if ttype == qmod.TokenType.PARTIAL:
+            # qualifiers cannot appear in the middle of the query. They need
+            # to be near the next phrase.
+            if self.direction == -1 \
+               and any(t.ttype == qmod.TokenType.QUALIFIER for t in self.seq[:-1]):
+                return None
             return self.direction
 
         # Other tokens may only appear once
@@ -165,22 +170,22 @@ class _TokenSequence:
         if ttype == qmod.TokenType.COUNTRY:
             return None if self.direction == -1 else 1
 
-        if ttype == qmod.TokenType.CATEGORY:
+        if ttype == qmod.TokenType.NEAR_ITEM:
             return self.direction
 
         if ttype == qmod.TokenType.QUALIFIER:
             if self.direction == 1:
                 if (len(self.seq) == 1
-                    and self.seq[0].ttype in (qmod.TokenType.PARTIAL, qmod.TokenType.CATEGORY)) \
+                    and self.seq[0].ttype in (qmod.TokenType.PARTIAL, qmod.TokenType.NEAR_ITEM)) \
                    or (len(self.seq) == 2
-                       and self.seq[0].ttype == qmod.TokenType.CATEGORY
+                       and self.seq[0].ttype == qmod.TokenType.NEAR_ITEM
                        and self.seq[1].ttype == qmod.TokenType.PARTIAL):
                     return 1
                 return None
             if self.direction == -1:
                 return -1
 
-            tempseq = self.seq[1:] if self.seq[0].ttype == qmod.TokenType.CATEGORY else self.seq
+            tempseq = self.seq[1:] if self.seq[0].ttype == qmod.TokenType.NEAR_ITEM else self.seq
             if len(tempseq) == 0:
                 return 1
             if len(tempseq) == 1 and self.seq[0].ttype == qmod.TokenType.HOUSENUMBER:
@@ -233,10 +238,10 @@ class _TokenSequence:
 
     def recheck_sequence(self) -> bool:
         """ Check that the sequence is a fully valid token assignment
-            and addapt direction and penalties further if necessary.
+            and adapt direction and penalties further if necessary.
 
             This function catches some impossible assignments that need
-            forward context and can therefore not be exluded when building
+            forward context and can therefore not be excluded when building
             the assignment.
         """
         # housenumbers may not be further than 2 words from the beginning.
@@ -253,6 +258,8 @@ class _TokenSequence:
                 priors = sum(1 for t in self.seq[hnrpos+1:] if t.ttype == qmod.TokenType.PARTIAL)
                 if not self._adapt_penalty_from_priors(priors, 1):
                     return False
+            if any(t.ttype == qmod.TokenType.NEAR_ITEM for t in self.seq):
+                self.penalty += 1.0
 
         return True
 
@@ -270,10 +277,10 @@ class _TokenSequence:
             # <address>,<postcode> should give preference to address search
             if base.postcode.start == 0:
                 penalty = self.penalty
-                self.direction = -1 # name searches are only possbile backwards
+                self.direction = -1 # name searches are only possible backwards
             else:
                 penalty = self.penalty + 0.1
-                self.direction = 1 # name searches are only possbile forwards
+                self.direction = 1 # name searches are only possible forwards
             yield dataclasses.replace(base, penalty=penalty)
 
 
@@ -366,7 +373,7 @@ class _TokenSequence:
 
         # Postcode or country-only search
         if not base.address:
-            if not base.housenumber and (base.postcode or base.country or base.category):
+            if not base.housenumber and (base.postcode or base.country or base.near_item):
                 log().comment('postcode/country search')
                 yield dataclasses.replace(base, penalty=self.penalty)
         else:
@@ -383,7 +390,7 @@ class _TokenSequence:
                 yield from self._get_assignments_address_backward(base, query)
 
             # variant for special housenumber searches
-            if base.housenumber:
+            if base.housenumber and not base.qualifier:
                 yield dataclasses.replace(base, penalty=self.penalty)