bdd tests: do not query word table directly

The BDD tests cannot make assumptions about the structure of the
word table anymore because it depends on the tokenizer. Use more
abstract descriptions instead that ask for specific kinds of
tokens.

test/bdd/db/import/postcodes.feature

@@ -134,9 +134,7 @@ Feature: Import of postcodes
         Then location_postcode contains exactly
            | country | postcode | geometry |
            | de      | 01982    | country:de |
-        And word contains
-           | word  | class | type |
-           | 01982 | place | postcode |
+        And there are word tokens for postcodes 01982
 
     Scenario: Different postcodes with the same normalization can both be found
         Given the places

test/bdd/db/update/postcode.feature

@@ -18,10 +18,7 @@ Feature: Update of postcode
            | country | postcode | geometry |
            | de      | 01982    | country:de |
            | ch      | 4567     | country:ch |
-        And word contains
-           | word  | class | type |
-           | 01982 | place | postcode |
-           | 4567  | place | postcode |
+        And there are word tokens for postcodes 01982,4567
 
      Scenario: When the last postcode is deleted, it is deleted from postcode and word
         Given the places
@@ -34,12 +31,8 @@ Feature: Update of postcode
         Then location_postcode contains exactly
            | country | postcode | geometry |
            | ch      | 4567     | country:ch |
-        And word contains not
-           | word  | class | type |
-           | 01982 | place | postcode |
-        And word contains
-           | word  | class | type |
-           | 4567  | place | postcode |
+        And there are word tokens for postcodes 4567
+        And there are no word tokens for postcodes 01982
 
      Scenario: A postcode is not deleted from postcode and word when it exist in another country
         Given the places
@@ -52,9 +45,7 @@ Feature: Update of postcode
         Then location_postcode contains exactly
            | country | postcode | geometry |
            | ch      | 01982    | country:ch |
-        And word contains
-           | word  | class | type |
-           | 01982 | place | postcode |
+        And there are word tokens for postcodes 01982
 
      Scenario: Updating a postcode is reflected in postcode table
         Given the places
@@ -68,9 +59,7 @@ Feature: Update of postcode
         Then location_postcode contains exactly
            | country | postcode | geometry |
            | de      | 20453    | country:de |
-        And word contains
-           | word  | class | type |
-           | 20453 | place | postcode |
+        And there are word tokens for postcodes 20453
 
      Scenario: When changing from a postcode type, the entry appears in placex
         When importing
@@ -91,9 +80,7 @@ Feature: Update of postcode
         Then location_postcode contains exactly
            | country | postcode | geometry |
            | de      | 20453    | country:de |
-        And word contains
-           | word  | class | type |
-           | 20453 | place | postcode |
+        And there are word tokens for postcodes 20453
 
      Scenario: When changing to a postcode type, the entry disappears from placex
         When importing
@@ -114,6 +101,4 @@ Feature: Update of postcode
         Then location_postcode contains exactly
            | country | postcode | geometry |
            | de      | 01982    | country:de |
-        And word contains
-           | word  | class | type |
-           | 01982 | place | postcode |
+        And there are word tokens for postcodes 01982

test/bdd/steps/steps_db_ops.py

@@ -281,6 +281,39 @@ def check_word_table(context, exclude):
             else:
                 assert cur.rowcount > 0, "Row not in word table: %s" % '/'.join(values)
 
+
+@then("there are(?P<exclude> no)? word tokens for postcodes (?P<postcodes>.*)")
+def check_word_table_for_postcodes(context, exclude, postcodes):
+    """ Check that the tokenizer produces postcode tokens for the given
+        postcodes. The postcodes are a comma-separated list of postcodes.
+        Whitespace matters.
+    """
+    nctx = context.nominatim
+    tokenizer = tokenizer_factory.get_tokenizer_for_db(nctx.get_test_config())
+    with tokenizer.name_analyzer() as ana:
+        plist = [ana.normalize_postcode(p) for p in postcodes.split(',')]
+
+    plist.sort()
+
+    with context.db.cursor(cursor_factory=psycopg2.extras.DictCursor) as cur:
+        if nctx.tokenizer == 'legacy_icu':
+            cur.execute("""SELECT info->>'postcode' FROM word
+                           WHERE type = 'P' and info->>'postcode' = any(%s)""",
+                        (plist,))
+        else:
+            cur.execute("""SELECT word FROM word WHERE word = any(%s)
+                             and class = 'place' and type = 'postcode'""",
+                        (plist,))
+
+        found = [row[0] for row in cur]
+        assert len(found) == len(set(found)), f"Duplicate rows for postcodes: {found}"
+
+    if exclude:
+        assert len(found) == 0, f"Unexpected postcodes: {found}"
+    else:
+        assert set(found) == set(plist), \
+        f"Missing postcodes {set(plist) - set(found)}. Found: {found}"
+
 @then("place_addressline contains")
 def check_place_addressline(context):
     """ Check the contents of the place_addressline table. Each row represents