Merge pull request #2227 from nkbeast/fix-config-backslash-continuation

fix: join backslash line continuations when reading config values

Author: Byron

git/config.py

@@ -467,6 +467,45 @@ def string_decode(v: str) -> str:
 
         # END string_decode
 
+        def is_line_continuation(value: str) -> bool:
+            quoted = escaped = False
+            for char in value:
+                if escaped:
+                    escaped = False
+                elif char == "\\":
+                    escaped = True
+                elif char == '"':
+                    quoted = not quoted
+                elif char in "#;" and not quoted:
+                    return False
+            return escaped
+
+        def parse_value(value: str) -> str:
+            parsed: List[str] = []
+            whitespace: List[str] = []
+            quoted = escaped = False
+            escapes = {"b": "\b", "n": "\n", "t": "\t", '"': '"', "\\": "\\"}
+            for char in value:
+                if escaped:
+                    parsed.append(escapes.get(char, "\\" + char))
+                    escaped = False
+                    continue
+                if char.isspace() and not quoted:
+                    if parsed:
+                        whitespace.append(char)
+                    continue
+                if char in "#;" and not quoted:
+                    break
+                parsed.extend(whitespace)
+                whitespace.clear()
+                if char == "\\":
+                    escaped = True
+                elif char == '"':
+                    quoted = not quoted
+                else:
+                    parsed.append(char)
+            return "".join(parsed)
+
         while True:
             # We assume to read binary!
             line = fp.readline().decode(defenc)
@@ -513,7 +552,29 @@ def string_decode(v: str) -> str:
 
                     if len(optval) < 2 or optval[0] != '"':
                         # Does not open quoting.
-                        pass
+                        # A value ending in an odd number of backslashes
+                        # continues on the next line, exactly as git does: the
+                        # final backslash and the newline are removed and the
+                        # next line is appended before the complete value is
+                        # parsed. An even number means the last backslash is
+                        # escaped and the value ends there.
+                        continued = False
+                        while True:
+                            if not is_line_continuation(optval):
+                                break
+                            continuation = fp.readline()
+                            if not continuation:
+                                # Backslash at end of file: git drops it.
+                                optval = optval[:-1]
+                                break
+                            lineno = lineno + 1
+                            joined = continuation.decode(defenc)
+                            while joined.endswith("\n") or joined.endswith("\r"):
+                                joined = joined[:-1]
+                            optval = optval[:-1] + joined
+                            continued = True
+                        if continued:
+                            optval = parse_value(optval)
                     elif optval[-1] != '"':
                         # Opens quoting and does not close: appears to start multi-line quoting.
                         is_multi_line = True

test/test_config.py

@@ -143,6 +143,44 @@ def test_multi_line_config(self):
             )
             self.assertEqual(len(config.sections()), 23)
 
+    def test_backslash_line_continuation(self):
+        """An unquoted value ending in a backslash continues on the next line,
+        exactly as git config parses it: the final backslash and the newline
+        are removed before the complete logical value is parsed."""
+        cases = [
+            (b"[a]\n\tk = line1\\\n line2\n", "line1 line2"),
+            (b"[a]\n\tk = one\\\n two\\\n three\n", "one two three"),
+            (b"[a]\n\tk = one\\\n two   \n", "one two"),
+            (b"[a]\n\tk = one\\\n two ; ignored\n", "one two"),
+            (b'[a]\n\tk = one\\\n "two"\n', "one two"),
+            (b"[a]\n\tk = one\\\n two\\tthree\n", "one two\tthree"),
+            (b"[a]\n\tk = val\\\\\n next\n", "val\\\\"),
+            (b"[a]\n\tk = end\\\n", "end"),
+            (b"[alias]\n\tco = checkout \\\n\t\t-v\n", "checkout \t\t-v"),
+        ]
+        for content, expected in cases:
+            config_file = io.BytesIO(content)
+            config_file.name = "backslash_continuation.config"
+            config = GitConfigParser(config_file)
+            config.read()
+            section = "alias" if b"[alias]" in content else "a"
+            key = "co" if section == "alias" else "k"
+            self.assertEqual(config.get_value(section, key), expected)
+
+    @with_rw_directory
+    def test_comment_backslash_does_not_continue_value(self, rw_dir):
+        config_path = osp.join(rw_dir, "config")
+        with open(config_path, "wb") as config_file:
+            config_file.write(b"[a]\n\tk = one\\\n two ; ignored \\\n\tx = two\n")
+
+        with GitConfigParser(config_path, read_only=False) as config:
+            self.assertEqual(config.get_value("a", "k"), "one two")
+            self.assertEqual(config.get_value("a", "x"), "two")
+            config.set_value("a", "added", "three")
+
+        with GitConfigParser(config_path) as config:
+            self.assertEqual(config.get_value("a", "x"), "two")
+
     def test_config_value_with_trailing_new_line(self):
         config_content = b'[section-header]\nkey:"value\n"'
         config_file = io.BytesIO(config_content)