Fix buffering unit test failing on some platforms (#896)

tests/test_add_option_kwargs.py

@@ -1,5 +1,3 @@
-import io
-
 import pytest
 
 from loguru import logger
@@ -21,13 +19,33 @@ def test_file_mode_w(tmp_path):
     assert file.read_text() == "msg\n"
 
 
-def test_file_buffering(tmp_path):
-    file = tmp_path / "test.log"
-    logger.add(file, format="{message}", buffering=-1)
-    logger.debug("x" * (io.DEFAULT_BUFFER_SIZE // 2))
-    assert file.read_text() == ""
-    logger.debug("x" * (io.DEFAULT_BUFFER_SIZE * 2))
-    assert file.read_text() != ""
+def test_file_auto_buffering(tmp_path):
+    # There doesn't seem to be a reliable way to known buffer size for text files.
+    # We perform a preliminary test to ensure empirically that 128 <= buffer size <= 65536.
+    dummy_filepath = tmp_path / "dummy.txt"
+    with open(str(dummy_filepath), buffering=-1, mode="w") as dummy_file:
+        dummy_file.write("." * 127)
+        if dummy_filepath.read_text() != "":
+            pytest.skip("Size buffer for text files is too small.")
+        dummy_file.write("." * (65536 - 127))
+        if dummy_filepath.read_text() == "":
+            pytest.skip("Size buffer for text files is too big.")
+
+    filepath = tmp_path / "test.log"
+    logger.add(filepath, format="{message}", buffering=-1)
+    logger.debug("A short message.")
+    assert filepath.read_text() == ""
+    logger.debug("A long message" + "." * 65536)
+    assert filepath.read_text() != ""
+
+
+def test_file_line_buffering(tmp_path):
+    filepath = tmp_path / "test.log"
+    logger.add(filepath, format=lambda _: "{message}", buffering=1)
+    logger.debug("Without newline")
+    assert filepath.read_text() == ""
+    logger.debug("With newline\n")
+    assert filepath.read_text() != ""
 
 
 def test_invalid_function_kwargs():