From 83ad96b1b3fa3519bccf2fe82c9c683a8f16ce37 Mon Sep 17 00:00:00 2001
From: shamoon <4887959+shamoon@users.noreply.github.com>
Date: Sun, 2 Aug 2026 17:50:45 -0700
Subject: [PATCH] make allow_infix order-stable

---
 src/documents/search/_query.py             | 16 ++++++++-----
 src/documents/tests/search/test_backend.py | 26 ++++++++++++----------
 2 files changed, 24 insertions(+), 18 deletions(-)

diff --git a/src/documents/search/_query.py b/src/documents/search/_query.py
index 15b672e1f..3584767bc 100644
--- a/src/documents/search/_query.py
+++ b/src/documents/search/_query.py
@@ -148,10 +148,10 @@ def _build_simple_token_query(
 ) -> tantivy.Query:
     escaped = regex.escape(token)
     # The simple analyzer keeps punctuation inside whitespace-delimited terms.
-    # Later query tokens may therefore begin either at the indexed term boundary
-    # or after punctuation within a term (for example, ``medical-history``).
-    # Do not allow an arbitrary infix for later tokens: a query ending in ``6``
-    # must not match the middle of ``16``.
+    # Boundary-constrained query tokens may therefore begin either at the indexed
+    # term boundary or after punctuation within a term (for example,
+    # ``medical-history``). This avoids matching a numeric token such as ``6``
+    # in the middle of ``16``.
     pattern = (
         f".*{escaped}.*"
         if allow_infix
@@ -282,10 +282,14 @@ def parse_simple_query(
                     index,
                     fields,
                     token,
-                    allow_infix=idx == 0,
+                    # Preserve historical infix matching for single-token
+                    # searches. In multi-token searches, constrain numeric
+                    # tokens to boundaries to avoid partial-number overlap.
+                    # This depends on token content, not query order.
+                    allow_infix=len(tokens) == 1 or not token.isdecimal(),
                 ),
             )
-            for idx, token in enumerate(tokens)
+            for token in tokens
         ]
         simple_query = (
             token_queries[0][1]
diff --git a/src/documents/tests/search/test_backend.py b/src/documents/tests/search/test_backend.py
index 3f8989180..8615dc0ee 100644
--- a/src/documents/tests/search/test_backend.py
+++ b/src/documents/tests/search/test_backend.py
@@ -163,10 +163,11 @@ class TestSearch:
         assert (
             len(backend.search_ids("sswo", user=None, search_mode=SearchMode.TEXT)) == 1
         )
-        assert (
-            len(backend.search_ids("sswo re", user=None, search_mode=SearchMode.TEXT))
-            == 1
-        )
+        for query in ["sswo re", "re sswo"]:
+            assert (
+                len(backend.search_ids(query, user=None, search_mode=SearchMode.TEXT))
+                == 1
+            ), query
 
     def test_text_mode_matches_all_terms_without_requiring_adjacency(
         self,
@@ -230,11 +231,11 @@ class TestSearch:
             == 0
         )
 
-    def test_text_mode_anchors_later_query_tokens_to_token_starts(
+    def test_text_mode_anchors_numeric_tokens_regardless_of_query_order(
         self,
         backend: TantivyBackend,
     ) -> None:
-        """Multi-token simple search should not match later tokens in the middle of a word."""
+        """Numeric tokens must not match in the middle of a larger number."""
         exact_doc = Document.objects.create(
             title="Z-Berichte 6",
             content="monthly report",
@@ -257,13 +258,14 @@ class TestSearch:
         backend.add_or_update(prefix_doc)
         backend.add_or_update(false_positive)
 
-        result_ids = set(
-            backend.search_ids("Z-Berichte 6", user=None, search_mode=SearchMode.TEXT),
-        )
+        for query in ["Z-Berichte 6", "6 Z-Berichte"]:
+            result_ids = set(
+                backend.search_ids(query, user=None, search_mode=SearchMode.TEXT),
+            )
 
-        assert exact_doc.id in result_ids
-        assert prefix_doc.id in result_ids
-        assert false_positive.id not in result_ids
+            assert exact_doc.id in result_ids, query
+            assert prefix_doc.id in result_ids, query
+            assert false_positive.id not in result_ids, query
 
     def test_text_mode_ignores_queries_without_searchable_tokens(
         self,