disallow address searches that start with a postcode

[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 e9c03d3f96a76a490f7849be3bd97173fa162487..edddd1004534a13dc88d7c925ed498d7bb3668a5 100644 (file)
--- a/nominatim/api/search/token_assignment.py
+++ b/nominatim/api/search/token_assignment.py
@@ -257,6 +257,26 @@ class _TokenSequence:
         return True
 
 
+    def _get_assignments_postcode(self, base: TokenAssignment,
+                                  query_len: int)  -> Iterator[TokenAssignment]:
+        """ Yield possible assignments of Postcode searches with an
+            address component.
+        """
+        assert base.postcode is not None
+
+        if (base.postcode.start == 0 and self.direction != -1)\
+           or (base.postcode.end == query_len and self.direction != 1):
+            log().comment('postcode search')
+            # <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
+            else:
+                penalty = self.penalty + 0.1
+                self.direction = 1 # name searches are only possbile forewards
+            yield dataclasses.replace(base, penalty=penalty)
+
+
     def get_assignments(self, query: qmod.QueryStruct) -> Iterator[TokenAssignment]:
         """ Yield possible assignments for the current sequence.
 
@@ -265,12 +285,13 @@ class _TokenSequence:
         """
         base = TokenAssignment.from_ranges(self.seq)
 
+        num_addr_tokens = sum(t.end - t.start for t in base.address)
+        if num_addr_tokens > 50:
+            return
+
         # Postcode search (postcode-only search is covered in next case)
         if base.postcode is not None and base.address:
-            if (base.postcode.start == 0 and self.direction != -1)\
-               or (base.postcode.end == query.num_token_slots() and self.direction != 1):
-                log().comment('postcode search')
-                yield dataclasses.replace(base, penalty=self.penalty)
+            yield from self._get_assignments_postcode(base, query.num_token_slots())
 
         # Postcode or country-only search
         if not base.address:
@@ -278,6 +299,9 @@ class _TokenSequence:
                 log().comment('postcode/country search')
                 yield dataclasses.replace(base, penalty=self.penalty)
         else:
+            # <postcode>,<address> should give preference to postcode search
+            if base.postcode and base.postcode.start == 0:
+                self.penalty += 0.1
             # Use entire first word as name
             if self.direction != -1:
                 log().comment('first word = name')
@@ -292,14 +316,22 @@ class _TokenSequence:
                                           penalty=self.penalty,
                                           address=base.address[:-1])
 
+            # variant for special housenumber searches
+            if base.housenumber:
+                yield dataclasses.replace(base, penalty=self.penalty)
+
             # Use beginning of first word as name
             if self.direction != -1:
                 first = base.address[0]
                 if (not base.housenumber or first.end >= base.housenumber.start)\
                    and (not base.qualifier or first.start >= base.qualifier.end):
+                    base_penalty = self.penalty
+                    if (base.housenumber and base.housenumber.start > first.start) \
+                       or len(query.source) > 1:
+                        base_penalty += 0.25
                     for i in range(first.start + 1, first.end):
                         name, addr = first.split(i)
-                        penalty = self.penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype]
+                        penalty = base_penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype]
                         log().comment(f'split first word = name ({i - first.start})')
                         yield dataclasses.replace(base, name=name, penalty=penalty,
                                                   address=[addr] + base.address[1:])
@@ -309,9 +341,14 @@ class _TokenSequence:
                 last = base.address[-1]
                 if (not base.housenumber or last.start <= base.housenumber.end)\
                    and (not base.qualifier or last.end <= base.qualifier.start):
+                    base_penalty = self.penalty
+                    if base.housenumber and base.housenumber.start < last.start:
+                        base_penalty += 0.4
+                    if len(query.source) > 1:
+                        base_penalty += 0.25
                     for i in range(last.start + 1, last.end):
                         addr, name = last.split(i)
-                        penalty = self.penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype]
+                        penalty = base_penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype]
                         log().comment(f'split last word = name ({i - last.start})')
                         yield dataclasses.replace(base, name=name, penalty=penalty,
                                                   address=base.address[:-1] + [addr])