diff-cover 10.4.0__tar.gz → 10.4.1__tar.gz

{diff_cover-10.4.0 → diff_cover-10.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: diff_cover
-Version: 10.4.0
+Version: 10.4.1
 Summary: Run coverage and linting reports on diffs
 License: Apache-2.0
 Author: See Contributors

{diff_cover-10.4.0 → diff_cover-10.4.1}/diff_cover/snippets.py

@@ -17,6 +17,17 @@ from pygments.util import ClassNotFound
 
 from diff_cover.git_path import GitPathTool
 
+# Below this, chardet is essentially guessing. Short files with only a
+# handful of non-ascii bytes decode cleanly under every single byte
+# encoding, so which one comes back is down to the detector's internals
+# (chardet 7 answers MacRoman at 0.16 confidence for latin-1 text that
+# chardet 5 called ISO-8859-1 at 0.73).
+MIN_DETECTION_CONFIDENCE = 0.5
+
+# What to try before an unconfident guess. cp1252 is the de facto default
+# for undeclared western text and is a superset of latin-1 in practice.
+FALLBACK_ENCODING = "cp1252"
+
 
 class Snippet:
     """
@@ -284,9 +295,15 @@ class Snippet:
                 contents = src_file.read()
 
         if isinstance(contents, bytes):
-            encoding = chardet.detect(contents).get("encoding", "utf-8")
-            with contextlib.suppress(UnicodeDecodeError):
-                contents = contents.decode(encoding)
+            detected = chardet.detect(contents)
+            candidates = [detected.get("encoding") or "utf-8"]
+            if (detected.get("confidence") or 0) < MIN_DETECTION_CONFIDENCE:
+                candidates.insert(0, FALLBACK_ENCODING)
+
+            for encoding in candidates:
+                with contextlib.suppress(UnicodeDecodeError, LookupError):
+                    contents = contents.decode(encoding)
+                    break
 
         if isinstance(contents, bytes):
             # We failed to decode the file.

{diff_cover-10.4.0 → diff_cover-10.4.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "diff_cover"
-version = "10.4.0"
+version = "10.4.1"
 description = "Run coverage and linting reports on diffs"
 authors = ["See Contributors"]
 homepage = "https://github.com/Bachmann1234/diff-cover"

{diff_cover-10.4.0 → diff_cover-10.4.1}/tests/test_snippets.py

@@ -429,3 +429,42 @@ def test_latin_one_undeclared(tmp_path):
 
     contents = Snippet.load_contents(str(file))
     assert contents == "I am some latin 1 Â encoded text"
+
+
+def test_unconfident_detection_falls_back(tmp_path, mocker):
+    # Which single byte encoding chardet picks for a short file varies by
+    # version (5 said ISO-8859-1, 6 WINDOWS-1252, 7 MacRoman) so pin the
+    # low confidence branch rather than depending on the installed one.
+    mocker.patch(
+        "diff_cover.snippets.chardet.detect",
+        return_value={"encoding": "MacRoman", "confidence": 0.16},
+    )
+    file = tmp_path / "tmp"
+    file.write_bytes("I am some latin 1 Â encoded text".encode("latin1"))
+
+    contents = Snippet.load_contents(str(file))
+    assert contents == "I am some latin 1 Â encoded text"
+
+
+def test_confident_detection_is_trusted(tmp_path, mocker):
+    mocker.patch(
+        "diff_cover.snippets.chardet.detect",
+        return_value={"encoding": "MacRoman", "confidence": 0.99},
+    )
+    file = tmp_path / "tmp"
+    file.write_bytes("I am some latin 1 Â encoded text".encode("latin1"))
+
+    contents = Snippet.load_contents(str(file))
+    assert contents == "I am some latin 1 ¬ encoded text"
+
+
+def test_detection_returns_no_encoding(tmp_path, mocker):
+    mocker.patch(
+        "diff_cover.snippets.chardet.detect",
+        return_value={"encoding": None, "confidence": None},
+    )
+    file = tmp_path / "tmp"
+    file.write_bytes("I am some latin 1 Â encoded text".encode("latin1"))
+
+    contents = Snippet.load_contents(str(file))
+    assert contents == "I am some latin 1 Â encoded text"