Merge latest main into gatekeeper/fix-4392-1

Integrate the upstream tokenizer default fix after validating the combined remote error behavior and Python binding.

python/python/lancedb/__init__.py

@@ -321,8 +321,8 @@ def tokenize(
     language: str = "English",
     max_token_length: Optional[int] = 40,
     lower_case: bool = True,
-    stem: bool = True,
-    remove_stop_words: bool = True,
+    stem: Optional[bool] = None,
+    remove_stop_words: Optional[bool] = None,
     custom_stop_words: Optional[List[str]] = None,
     ascii_folding: bool = True,
     ngram_min_length: int = 3,
@@ -333,6 +333,8 @@ def tokenize(
 
     This does not require an FTS index. The tokenizer options match
     :class:`lancedb.index.FTS`. ``custom_stop_words`` accepts a list of strings.
+    When ``stem`` or ``remove_stop_words`` is None, the base tokenizer's default
+    applies: False for ``code`` and ``ngram``, True otherwise.
     """
 
     return _tokenize(

python/python/lancedb/_lancedb.pyi

@@ -58,8 +58,8 @@ def tokenize(
     language: str = "English",
     max_token_length: Optional[int] = 40,
     lower_case: bool = True,
-    stem: bool = True,
-    remove_stop_words: bool = True,
+    stem: Optional[bool] = None,
+    remove_stop_words: Optional[bool] = None,
     custom_stop_words: Optional[List[str]] = None,
     ascii_folding: bool = True,
     ngram_min_length: int = 3,

python/python/lancedb/index.py

@@ -138,6 +138,7 @@ class FTS:
         - "simple": Splits text by whitespace and punctuation.
         - "whitespace": Split text by whitespace, but not punctuation.
         - "raw": No tokenization. The entire text is treated as a single token.
+        - "code": Tokenizes source code and identifiers.
         - "ngram": N-gram tokenizer for substring-style matching.
         - "icu": ICU dictionary-based word segmentation.
         - "icu/split": ICU segmentation with simple-style delimiter splitting.
@@ -151,12 +152,16 @@ class FTS:
         ignored.
     lower_case : bool, default True
         Whether to convert the token to lower case. This makes queries case-insensitive.
-    stem : bool, default True
+    stem : bool, optional
         Whether to stem the token. Stemming reduces words to their root form.
         For example, in English "running" and "runs" would both be reduced to "run".
-    remove_stop_words : bool, default True
+        ``None`` uses the base tokenizer's default: False for ``code`` and
+        ``ngram``, True otherwise.
+    remove_stop_words : bool, optional
         Whether to remove stop words. Stop words are common words that are often
         removed from text before indexing. For example, in English "the" and "and".
+        ``None`` uses the base tokenizer's default: False for ``code`` and
+        ``ngram``, True otherwise.
     custom_stop_words : list of str, optional
         Custom words replace the built-in language stop words
         and only take effect when ``remove_stop_words`` is True. ``None`` uses
@@ -197,8 +202,8 @@ class FTS:
     language: str = "English"
     max_token_length: Optional[int] = 40
     lower_case: bool = True
-    stem: bool = True
-    remove_stop_words: bool = True
+    stem: Optional[bool] = None
+    remove_stop_words: Optional[bool] = None
     ascii_folding: bool = True
     ngram_min_length: int = 3
     ngram_max_length: int = 3

python/python/lancedb/remote/table.py

@@ -380,8 +380,8 @@ class RemoteTable(Table):
         language: str = "English",
         max_token_length: Optional[int] = 40,
         lower_case: bool = True,
-        stem: bool = True,
-        remove_stop_words: bool = True,
+        stem: Optional[bool] = None,
+        remove_stop_words: Optional[bool] = None,
         custom_stop_words: Optional[List[str]] = None,
         ascii_folding: bool = True,
         ngram_min_length: int = 3,

python/python/lancedb/table.py

@@ -1612,8 +1612,8 @@ class Table(ABC):
         language: str = "English",
         max_token_length: Optional[int] = 40,
         lower_case: bool = True,
-        stem: bool = True,
-        remove_stop_words: bool = True,
+        stem: Optional[bool] = None,
+        remove_stop_words: Optional[bool] = None,
         custom_stop_words: Optional[List[str]] = None,
         ascii_folding: bool = True,
         ngram_min_length: int = 3,
@@ -1663,6 +1663,7 @@ class Table(ABC):
             - "simple": Splits text by whitespace and punctuation.
             - "whitespace": Split text by whitespace, but not punctuation.
             - "raw": No tokenization. The entire text is treated as a single token.
+            - "code": Tokenizes source code and identifiers.
             - "ngram": N-Gram tokenizer.
             - "icu": ICU dictionary-based word segmentation.
             - "icu/split": ICU segmentation with simple-style delimiter splitting.
@@ -1677,12 +1678,16 @@ class Table(ABC):
         lower_case : bool, default True
             Whether to convert the token to lower case. This makes queries
             case-insensitive.
-        stem : bool, default True
+        stem : bool, optional
             Whether to stem the token. Stemming reduces words to their root form.
             For example, in English "running" and "runs" would both be reduced to "run".
-        remove_stop_words : bool, default True
+            ``None`` uses the base tokenizer's default: False for ``code`` and
+            ``ngram``, True otherwise.
+        remove_stop_words : bool, optional
             Whether to remove stop words. Stop words are common words that are often
             removed from text before indexing. For example, in English "the" and "and".
+            ``None`` uses the base tokenizer's default: False for ``code`` and
+            ``ngram``, True otherwise.
         custom_stop_words : list of str, optional
             Custom words that replace the built-in language stop words. ``None``
             uses the built-in list; an empty list explicitly uses no stop words.
@@ -3820,8 +3825,8 @@ class LanceTable(Table):
         language: str = "English",
         max_token_length: Optional[int] = 40,
         lower_case: bool = True,
-        stem: bool = True,
-        remove_stop_words: bool = True,
+        stem: Optional[bool] = None,
+        remove_stop_words: Optional[bool] = None,
         custom_stop_words: Optional[List[str]] = None,
         ascii_folding: bool = True,
         ngram_min_length: int = 3,

python/python/tests/test_fts.py

@@ -895,6 +895,78 @@ def test_tokenize_uses_simple_index_tokenizer(mem_db: DBConnection):
     ]
 
 
+@pytest.mark.parametrize("legacy", [False, True])
+@pytest.mark.parametrize(
+    ("base_tokenizer", "options", "expected_filters", "expected_tokens"),
+    [
+        (
+            "code",
+            {},
+            (False, False),
+            [("running", 0), ("in", 1), ("cafe", 2)],
+        ),
+        (
+            "code",
+            {"stem": None, "remove_stop_words": None},
+            (False, False),
+            [("running", 0), ("in", 1), ("cafe", 2)],
+        ),
+        (
+            "code",
+            {"stem": True},
+            (True, False),
+            [("run", 0), ("in", 1), ("cafe", 2)],
+        ),
+        (
+            "code",
+            {"remove_stop_words": True},
+            (False, True),
+            [("running", 0), ("cafe", 2)],
+        ),
+        (
+            "code",
+            {"stem": True, "remove_stop_words": True},
+            (True, True),
+            [("run", 0), ("cafe", 2)],
+        ),
+        (
+            "simple",
+            {},
+            (True, True),
+            [("run", 0), ("cafe", 2)],
+        ),
+        (
+            "simple",
+            {"stem": False, "remove_stop_words": False},
+            (False, False),
+            [("running", 0), ("in", 1), ("cafe", 2)],
+        ),
+    ],
+)
+def test_fts_tokenizer_filter_defaults(
+    mem_db, legacy, base_tokenizer, options, expected_filters, expected_tokens
+):
+    text = "Running in café"
+    table = mem_db.create_table("test_filter_defaults", data=[{"text": text}])
+    if legacy:
+        with pytest.warns(DeprecationWarning, match="create_fts_index"):
+            table.create_fts_index("text", base_tokenizer=base_tokenizer, **options)
+    else:
+        table.create_index("text", config=FTS(base_tokenizer=base_tokenizer, **options))
+
+    details = table.list_indices()[0].index_details
+    assert (details["stem"], details["remove_stop_words"]) == expected_filters
+    assert [
+        (token.text, token.position) for token in table.tokenize(text, column="text")
+    ] == expected_tokens
+    assert [
+        (token.text, token.position)
+        for token in ldb.tokenize(text, base_tokenizer=base_tokenizer, **options)
+    ] == expected_tokens
+    if base_tokenizer == "code" and not expected_filters[1]:
+        assert table.search("in", query_type="fts").to_list()[0]["text"] == text
+
+
 def test_tokenize_uses_icu_index_tokenizer_by_name(mem_db: DBConnection):
     data = pa.table({"text": ["Hello, こんにちは世界!"]})
     table = mem_db.create_table("test_tokenize_icu", data=data)

python/python/tests/test_remote_db.py

@@ -1128,6 +1128,77 @@ def test_table_create_indices():
         table.drop_index("custom_fts_idx")
 
 
+@pytest.mark.parametrize("legacy", [False, True])
+@pytest.mark.parametrize(
+    ("base_tokenizer", "options", "expected_filters"),
+    [
+        ("code", {}, (False, False)),
+        ("code", {"stem": None, "remove_stop_words": None}, (False, False)),
+        ("code", {"stem": True}, (True, False)),
+        ("code", {"remove_stop_words": True}, (False, True)),
+        ("code", {"stem": True, "remove_stop_words": True}, (True, True)),
+        ("simple", {}, (True, True)),
+        ("simple", {"stem": False, "remove_stop_words": False}, (False, False)),
+    ],
+)
+def test_remote_fts_tokenizer_filter_defaults(
+    legacy, base_tokenizer, options, expected_filters
+):
+    from lancedb.index import FTS
+
+    received_requests = []
+
+    def handler(request):
+        if request.path == "/v1/table/test/create_index/":
+            content_len = int(request.headers["Content-Length"])
+            received_requests.append(json.loads(request.rfile.read(content_len)))
+            request.send_response(200)
+            request.end_headers()
+        elif request.path == "/v1/table/test/create/?mode=create":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"{}")
+        elif request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(
+                json.dumps(
+                    {
+                        "version": 1,
+                        "schema": {
+                            "fields": [
+                                {
+                                    "name": "text",
+                                    "type": {"type": "string"},
+                                    "nullable": False,
+                                }
+                            ]
+                        },
+                    }
+                ).encode()
+            )
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    with mock_lancedb_connection(handler) as db:
+        table = db.create_table("test", [{"text": "Running in cafes"}])
+        if legacy:
+            with pytest.warns(DeprecationWarning, match="create_fts_index"):
+                table.create_fts_index("text", base_tokenizer=base_tokenizer, **options)
+        else:
+            table.create_index(
+                "text", config=FTS(base_tokenizer=base_tokenizer, **options)
+            )
+
+    assert len(received_requests) == 1
+    params = received_requests[0]
+    assert params["base_tokenizer"] == base_tokenizer
+    assert (params["stem"], params["remove_stop_words"]) == expected_filters
+
+
 def test_remote_create_index_async_returns_job():
     from lancedb.index import BTree
 

python/src/index.rs

@@ -54,8 +54,6 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                     .with_position(params.with_position)
                     .lower_case(params.lower_case)
                     .max_token_length(params.max_token_length)
-                    .remove_stop_words(params.remove_stop_words)
-                    .stem(params.stem)
                     .ascii_folding(params.ascii_folding)
                     .ngram_min_length(params.ngram_min_length)
                     .ngram_max_length(params.ngram_max_length)
@@ -65,6 +63,12 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                         DocumentGranularity::try_from(params.document_granularity.as_str())
                             .map_err(|err| PyValueError::new_err(err.to_string()))?,
                     );
+                if let Some(stem) = params.stem {
+                    inner_opts = inner_opts.stem(stem);
+                }
+                if let Some(remove_stop_words) = params.remove_stop_words {
+                    inner_opts = inner_opts.remove_stop_words(remove_stop_words);
+                }
                 if let Some(memory_limit) = params.memory_limit {
                     inner_opts = inner_opts.memory_limit_mb(memory_limit);
                 }
@@ -215,8 +219,8 @@ struct FtsParams {
     language: String,
     max_token_length: Option<usize>,
     lower_case: bool,
-    stem: bool,
-    remove_stop_words: bool,
+    stem: Option<bool>,
+    remove_stop_words: Option<bool>,
     custom_stop_words: Option<Vec<String>>,
     ascii_folding: bool,
     ngram_min_length: u32,

python/src/table.rs

@@ -823,8 +823,8 @@ impl From<LanceDbFtsToken> for FtsToken {
     language = "English".to_string(),
     max_token_length = Some(40),
     lower_case = true,
-    stem = true,
-    remove_stop_words = true,
+    stem = None,
+    remove_stop_words = None,
     custom_stop_words = None,
     ascii_folding = true,
     ngram_min_length = 3,
@@ -838,15 +838,15 @@ pub fn tokenize(
     language: String,
     max_token_length: Option<u32>,
     lower_case: bool,
-    stem: bool,
-    remove_stop_words: bool,
+    stem: Option<bool>,
+    remove_stop_words: Option<bool>,
     custom_stop_words: Option<Vec<String>>,
     ascii_folding: bool,
     ngram_min_length: u32,
     ngram_max_length: u32,
     prefix_only: bool,
 ) -> PyResult<Vec<FtsToken>> {
-    let params = FtsIndexBuilder::default()
+    let mut params = FtsIndexBuilder::default()
         .base_tokenizer(base_tokenizer)
         .language(&language)
         .map_err(|_| {
@@ -857,13 +857,17 @@ pub fn tokenize(
         })?
         .max_token_length(max_token_length.map(|value| value as usize))
         .lower_case(lower_case)
-        .stem(stem)
-        .remove_stop_words(remove_stop_words)
         .ascii_folding(ascii_folding)
         .ngram_min_length(ngram_min_length)
         .ngram_max_length(ngram_max_length)
         .ngram_prefix_only(prefix_only)
         .custom_stop_words(custom_stop_words);
+    if let Some(stem) = stem {
+        params = params.stem(stem);
+    }
+    if let Some(remove_stop_words) = remove_stop_words {
+        params = params.remove_stop_words(remove_stop_words);
+    }
     let tokens = lancedb_tokenize(&query, &params).infer_error()?;
     Ok(tokens.into_iter().map(FtsToken::from).collect())
 }