Fix Spanish number normalizer dropping the mil multiplier after a hundreds-block

normalization/languages/spanish/number_normalizer.py

@@ -126,17 +126,25 @@ def _parse_number(self, words: list[str], i: int, n: int) -> tuple[int, int] | N
                     end, v2 = tail
                     return end, base + v2
                 return j, base
-            if h == 100:
-                sub = self._parse_0_99(words, j, n)
-                if sub is not None:
-                    je, v = sub
-                    return je, 100 + v
-                return j, 100
+            # Hundred block optionally followed by a 0–99 remainder
+            # (e.g. "ciento cincuenta" → 150, "doscientos uno" → 201).
+            hundreds_value = h
             sub = self._parse_0_99(words, j, n)
             if sub is not None:
                 je, v = sub
-                return je, h + v
-            return j, h
+                hundreds_value = h + v
+                j = je
+            # The whole hundred block may itself be a "mil" multiplier
+            # (e.g. "ciento cincuenta mil" → 150000).
+            if j < n and _fold(words[j]) == "mil":
+                base = hundreds_value * 1000
+                j += 1
+                tail = self._parse_number(words, j, n)
+                if tail is not None:
+                    end, v2 = tail
+                    return end, base + v2
+                return j, base
+            return j, hundreds_value
 
         # --- 0–99 or leading multiplier for "mil" ---
         sub99 = self._parse_0_99(words, i, n)

tests/unit/languages/spanish_number_normalizer_test.py

@@ -0,0 +1,30 @@
+import pytest
+
+from normalization.languages.spanish.number_normalizer import (
+    SpanishNumberNormalizer,
+)
+
+
+@pytest.fixture
+def normalizer():
+    return SpanishNumberNormalizer()
+
+
+@pytest.mark.parametrize(
+    ("text", "expected"),
+    [
+        ("veintitres", "23"),
+        ("cien", "100"),
+        ("doscientos", "200"),
+        ("doscientos cincuenta", "250"),
+        # A hundreds-block followed by a sub-part before "mil" must keep the
+        # thousands multiplier (previously dropped: "doscientos cincuenta mil"
+        # returned "250 mil" instead of "250000").
+        ("doscientos cincuenta mil", "250000"),
+        ("ciento cincuenta mil", "150000"),
+        ("novecientos noventa y nueve mil", "999000"),
+        ("doscientos cincuenta mil quinientos", "250500"),
+    ],
+)
+def test_spanish_number_normalizer(normalizer, text: str, expected: str):
+    assert normalizer(text) == expected