From 6bc1d33d583c54bd69fbdd2071117e2d38c354a1 Mon Sep 17 00:00:00 2001
From: Simon Willison <swillison@gmail.com>
Date: Tue, 22 Sep 2026 11:34:27 -0700
Subject: [PATCH] Fix resource leaks caused by buffered readers

Spotted running 'just test -Wall' on the LLM project.
---
 sqlite_utils/utils.py        |  6 +++-
 tests/test_rows_from_file.py | 70 ++++++++++++++++++++++++++++++++++++
 2 files changed, 75 insertions(+), 1 deletion(-)

diff --git a/sqlite_utils/utils.py b/sqlite_utils/utils.py
index ee6695b..91fd24d 100644
--- a/sqlite_utils/utils.py
+++ b/sqlite_utils/utils.py
@@ -380,10 +380,14 @@ def rows_from_file(
                 "rows_from_file() requires a file-like object that supports peek(), such as io.BytesIO"
             )
         if not first_bytes:
+            buffered.close()
             return (), Format.CSV
         if first_bytes.startswith((b"[", b"{")):
+            # JSON is read eagerly, so the detection wrapper can close now,
+            # including when parsing raises an error.
             # TODO: Detect newline-JSON
-            return rows_from_file(buffered, format=Format.JSON)
+            with buffered:
+                return rows_from_file(buffered, format=Format.JSON)
         else:
             dialect = csv.Sniffer().sniff(
                 first_bytes.decode(encoding or "utf-8-sig", "ignore")
diff --git a/tests/test_rows_from_file.py b/tests/test_rows_from_file.py
index 3de3582..a8e7f9d 100644
--- a/tests/test_rows_from_file.py
+++ b/tests/test_rows_from_file.py
@@ -1,3 +1,5 @@
+import io
+import json
 from io import BytesIO, StringIO
 
 import pytest
@@ -61,3 +63,71 @@ def test_rows_from_file_error_on_string_io():
     assert ex.value.args == (
         "rows_from_file() requires a file-like object that supports peek(), such as io.BytesIO",
     )
+
+
+@pytest.fixture
+def buffered_readers(monkeypatch):
+    # Keep wrappers alive so these checks cannot pass due to garbage collection.
+    readers = []
+    original = io.BufferedReader
+
+    def buffered_reader(*args, **kwargs):
+        reader = original(*args, **kwargs)
+        readers.append(reader)
+        return reader
+
+    monkeypatch.setattr(io, "BufferedReader", buffered_reader)
+    yield readers
+    for reader in readers:
+        reader.close()
+
+
+@pytest.mark.parametrize(
+    "content, expected_format, expected_rows",
+    [
+        (b'[{"id": 1}]', Format.JSON, [{"id": 1}]),
+        (b'{"id": 1}', Format.JSON, [{"id": 1}]),
+        (b"[]", Format.JSON, []),
+        (b"", Format.CSV, []),
+        (b" \n\t", Format.CSV, []),
+    ],
+)
+def test_detect_format_closes_eager_reader(
+    tmp_path, buffered_readers, content, expected_format, expected_rows
+):
+    path = tmp_path / "input"
+    path.write_bytes(content)
+    with path.open("rb") as fp:
+        rows, detected = rows_from_file(fp)
+        assert detected == expected_format
+        assert list(rows) == expected_rows
+        assert len(buffered_readers) == 1
+        assert buffered_readers[0].closed
+
+
+@pytest.mark.parametrize("content", [b"[", b'{"id":'])
+def test_detect_format_closes_reader_on_invalid_json(
+    tmp_path, buffered_readers, content
+):
+    path = tmp_path / "input.json"
+    path.write_bytes(content)
+    with path.open("rb") as fp:
+        with pytest.raises(json.JSONDecodeError):
+            rows_from_file(fp)
+        assert len(buffered_readers) == 1
+        assert buffered_readers[0].closed
+
+
+@pytest.mark.parametrize(
+    "delimiter, expected_format", [(b",", Format.CSV), (b"\t", Format.TSV)]
+)
+def test_detect_format_keeps_streaming_reader_open(
+    buffered_readers, delimiter, expected_format
+):
+    content = b"id" + delimiter + b"name\n" + (b"1" + delimiter + b"Cleo\n") * 2000
+    rows, detected = rows_from_file(BytesIO(content))
+    assert detected == expected_format
+    assert len(buffered_readers) == 1
+    assert not buffered_readers[0].closed
+    assert len(list(rows)) == 2000
+    assert buffered_readers[0].closed