]> git.madduck.net Git - etc/vim.git/blobdiff - src/black/__init__.py

madduck's git repository

Every one of the projects in this repository is available at the canonical URL git://git.madduck.net/madduck/pub/<projectpath> — see each project's metadata for the exact URL.

All patches and comments are welcome. Please squash your changes to logical commits before using git-format-patch and git-send-email to patches@git.madduck.net. If you'd read over the Git project's submission guidelines and adhered to them, I'd be especially grateful.

SSH access, as well as push access can be individually arranged.

If you use my repositories frequently, consider adding the following snippet to ~/.gitconfig and using the third clone URL listed for each project:

[url "git://git.madduck.net/madduck/"]
  insteadOf = madduck:

Bump pathspec to >= 0.8.1 to solve invalid .gitignore exclusion handling (#2084)
[etc/vim.git] / src / black / __init__.py
index 3ab4bc76bb23836a1f12bbe4953a5fb6bcbd7b55..051de5d13df58bfe7c0b6b300cfe629da27018d9 100644 (file)
@@ -42,14 +42,26 @@ from typing import (
     cast,
     TYPE_CHECKING,
 )
-from typing_extensions import Final
 from mypy_extensions import mypyc_attr
 
 from appdirs import user_cache_dir
 from dataclasses import dataclass, field, replace
 import click
 import toml
-from typed_ast import ast3, ast27
+
+try:
+    from typed_ast import ast3, ast27
+except ImportError:
+    if sys.version_info < (3, 8):
+        print(
+            "The typed_ast package is not installed.\n"
+            "You can install it with `python3 -m pip install typed-ast`.",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+    else:
+        ast3 = ast27 = ast
+
 from pathspec import PathSpec
 
 # lib2to3 fork
@@ -61,13 +73,19 @@ from blib2to3.pgen2.parse import ParseError
 
 from _black_version import version as __version__
 
+if sys.version_info < (3, 8):
+    from typing_extensions import Final
+else:
+    from typing import Final
+
 if TYPE_CHECKING:
     import colorama  # noqa: F401
 
 DEFAULT_LINE_LENGTH = 88
-DEFAULT_EXCLUDES = r"/(\.eggs|\.git|\.hg|\.mypy_cache|\.nox|\.tox|\.venv|\.svn|_build|buck-out|build|dist)/"  # noqa: B950
+DEFAULT_EXCLUDES = r"/(\.direnv|\.eggs|\.git|\.hg|\.mypy_cache|\.nox|\.tox|\.venv|venv|\.svn|_build|buck-out|build|dist)/"  # noqa: B950
 DEFAULT_INCLUDES = r"\.pyi?$"
 CACHE_DIR = Path(user_cache_dir("black", version=__version__))
+STDIN_PLACEHOLDER = "__BLACK_STDIN_FILENAME__"
 
 STRING_PREFIX_CHARS: Final = "furbFURB"  # All possible string prefix characters.
 
@@ -88,7 +106,7 @@ Transformer = Callable[["Line", Collection["Feature"]], Iterator["Line"]]
 Timestamp = float
 FileSize = int
 CacheInfo = Tuple[Timestamp, FileSize]
-Cache = Dict[Path, CacheInfo]
+Cache = Dict[str, CacheInfo]
 out = partial(click.secho, bold=True, err=True)
 err = partial(click.secho, fg="red", err=True)
 
@@ -112,6 +130,10 @@ class InvalidInput(ValueError):
     """Raised when input source code fails all parse attempts."""
 
 
+class BracketMatchError(KeyError):
+    """Raised when an opening bracket is unable to be matched to a closing bracket."""
+
+
 T = TypeVar("T")
 E = TypeVar("E", bound=Exception)
 
@@ -174,14 +196,12 @@ class TargetVersion(Enum):
     PY36 = 6
     PY37 = 7
     PY38 = 8
+    PY39 = 9
 
     def is_python2(self) -> bool:
         return self is TargetVersion.PY27
 
 
-PY36_VERSIONS = {TargetVersion.PY36, TargetVersion.PY37, TargetVersion.PY38}
-
-
 class Feature(Enum):
     # All string literals are unicode
     UNICODE_LITERALS = 1
@@ -195,6 +215,8 @@ class Feature(Enum):
     ASYNC_KEYWORDS = 7
     ASSIGNMENT_EXPRESSIONS = 8
     POS_ONLY_ARGUMENTS = 9
+    RELAXED_DECORATORS = 10
+    FORCE_OPTIONAL_PARENTHESES = 50
 
 
 VERSION_TO_FEATURES: Dict[TargetVersion, Set[Feature]] = {
@@ -232,6 +254,17 @@ VERSION_TO_FEATURES: Dict[TargetVersion, Set[Feature]] = {
         Feature.ASSIGNMENT_EXPRESSIONS,
         Feature.POS_ONLY_ARGUMENTS,
     },
+    TargetVersion.PY39: {
+        Feature.UNICODE_LITERALS,
+        Feature.F_STRINGS,
+        Feature.NUMERIC_UNDERSCORES,
+        Feature.TRAILING_COMMA_IN_CALL,
+        Feature.TRAILING_COMMA_IN_DEF,
+        Feature.ASYNC_KEYWORDS,
+        Feature.ASSIGNMENT_EXPRESSIONS,
+        Feature.RELAXED_DECORATORS,
+        Feature.POS_ONLY_ARGUMENTS,
+    },
 }
 
 
@@ -241,6 +274,8 @@ class Mode:
     line_length: int = DEFAULT_LINE_LENGTH
     string_normalization: bool = True
     is_pyi: bool = False
+    magic_trailing_comma: bool = True
+    experimental_string_processing: bool = False
 
     def get_cache_key(self) -> str:
         if self.target_versions:
@@ -255,6 +290,8 @@ class Mode:
             str(self.line_length),
             str(int(self.string_normalization)),
             str(int(self.is_pyi)),
+            str(int(self.magic_trailing_comma)),
+            str(int(self.experimental_string_processing)),
         ]
         return ".".join(parts)
 
@@ -267,11 +304,15 @@ def supports_feature(target_versions: Set[TargetVersion], feature: Feature) -> b
     return all(feature in VERSION_TO_FEATURES[version] for version in target_versions)
 
 
-def find_pyproject_toml(path_search_start: str) -> Optional[str]:
+def find_pyproject_toml(path_search_start: Tuple[str, ...]) -> Optional[str]:
     """Find the absolute filepath to a pyproject.toml if it exists"""
     path_project_root = find_project_root(path_search_start)
     path_pyproject_toml = path_project_root / "pyproject.toml"
-    return str(path_pyproject_toml) if path_pyproject_toml.is_file() else None
+    if path_pyproject_toml.is_file():
+        return str(path_pyproject_toml)
+
+    path_user_pyproject_toml = find_user_pyproject_toml()
+    return str(path_user_pyproject_toml) if path_user_pyproject_toml.is_file() else None
 
 
 def parse_pyproject_toml(path_config: str) -> Dict[str, Any]:
@@ -306,11 +347,19 @@ def read_pyproject_toml(
 
     if not config:
         return None
+    else:
+        # Sanitize the values to be Click friendly. For more information please see:
+        # https://github.com/psf/black/issues/1458
+        # https://github.com/pallets/click/issues/1567
+        config = {
+            k: str(v) if not isinstance(v, (list, dict)) else v
+            for k, v in config.items()
+        }
 
     target_version = config.get("target_version")
     if target_version is not None and not isinstance(target_version, list):
         raise click.BadOptionUsage(
-            "target-version", f"Config key target-version must be a list"
+            "target-version", "Config key target-version must be a list"
         )
 
     default_map: Dict[str, Any] = {}
@@ -333,6 +382,17 @@ def target_version_option_callback(
     return [TargetVersion[val.upper()] for val in v]
 
 
+def validate_regex(
+    ctx: click.Context,
+    param: click.Parameter,
+    value: Optional[str],
+) -> Optional[Pattern]:
+    try:
+        return re_compile_maybe_verbose(value) if value is not None else None
+    except re.error:
+        raise click.BadParameter("Not a valid regular expression")
+
+
 @click.command(context_settings=dict(help_option_names=["-h", "--help"]))
 @click.option("-c", "--code", type=str, help="Format the code passed in as a string.")
 @click.option(
@@ -368,12 +428,27 @@ def target_version_option_callback(
     is_flag=True,
     help="Don't normalize string quotes or prefixes.",
 )
+@click.option(
+    "-C",
+    "--skip-magic-trailing-comma",
+    is_flag=True,
+    help="Don't use trailing commas as a reason to split lines.",
+)
+@click.option(
+    "--experimental-string-processing",
+    is_flag=True,
+    hidden=True,
+    help=(
+        "Experimental option that performs more normalization on string literals."
+        " Currently disabled because it leads to some crashes."
+    ),
+)
 @click.option(
     "--check",
     is_flag=True,
     help=(
-        "Don't write the files back, just return the status.  Return code 0 means"
-        " nothing would change.  Return code 1 means some files would be reformatted."
+        "Don't write the files back, just return the status. Return code 0 means"
+        " nothing would change. Return code 1 means some files would be reformatted."
         " Return code 123 means there was an internal error."
     ),
 )
@@ -396,11 +471,12 @@ def target_version_option_callback(
     "--include",
     type=str,
     default=DEFAULT_INCLUDES,
+    callback=validate_regex,
     help=(
         "A regular expression that matches files and directories that should be"
-        " included on recursive searches.  An empty value means all files are included"
-        " regardless of the name.  Use forward slashes for directories on all platforms"
-        " (Windows, too).  Exclusions are calculated first, inclusions later."
+        " included on recursive searches. An empty value means all files are included"
+        " regardless of the name. Use forward slashes for directories on all platforms"
+        " (Windows, too). Exclusions are calculated first, inclusions later."
     ),
     show_default=True,
 )
@@ -408,20 +484,40 @@ def target_version_option_callback(
     "--exclude",
     type=str,
     default=DEFAULT_EXCLUDES,
+    callback=validate_regex,
     help=(
         "A regular expression that matches files and directories that should be"
-        " excluded on recursive searches.  An empty value means no paths are excluded."
-        " Use forward slashes for directories on all platforms (Windows, too). "
+        " excluded on recursive searches. An empty value means no paths are excluded."
+        " Use forward slashes for directories on all platforms (Windows, too)."
         " Exclusions are calculated first, inclusions later."
     ),
     show_default=True,
 )
+@click.option(
+    "--extend-exclude",
+    type=str,
+    callback=validate_regex,
+    help=(
+        "Like --exclude, but adds additional files and directories on top of the"
+        " excluded ones. (Useful if you simply want to add to the default)"
+    ),
+)
 @click.option(
     "--force-exclude",
     type=str,
+    callback=validate_regex,
     help=(
         "Like --exclude, but files and directories matching this regex will be "
-        "excluded even when they are passed explicitly as arguments"
+        "excluded even when they are passed explicitly as arguments."
+    ),
+)
+@click.option(
+    "--stdin-filename",
+    type=str,
+    help=(
+        "The name of the file when passing it through stdin. Useful to make "
+        "sure Black will respect --force-exclude option on some "
+        "editors that rely on using stdin."
     ),
 )
 @click.option(
@@ -439,7 +535,7 @@ def target_version_option_callback(
     is_flag=True,
     help=(
         "Also emit messages to stderr about files that were not changed or were ignored"
-        " due to --exclude=."
+        " due to exclusion patterns."
     ),
 )
 @click.version_option(version=__version__)
@@ -463,7 +559,7 @@ def target_version_option_callback(
     ),
     is_eager=True,
     callback=read_pyproject_toml,
-    help="Read configuration from PATH.",
+    help="Read configuration from FILE path.",
 )
 @click.pass_context
 def main(
@@ -477,11 +573,15 @@ def main(
     fast: bool,
     pyi: bool,
     skip_string_normalization: bool,
+    skip_magic_trailing_comma: bool,
+    experimental_string_processing: bool,
     quiet: bool,
     verbose: bool,
-    include: str,
-    exclude: str,
-    force_exclude: Optional[str],
+    include: Pattern,
+    exclude: Pattern,
+    extend_exclude: Optional[Pattern],
+    force_exclude: Optional[Pattern],
+    stdin_filename: Optional[str],
     src: Tuple[str, ...],
     config: Optional[str],
 ) -> None:
@@ -497,6 +597,8 @@ def main(
         line_length=line_length,
         is_pyi=pyi,
         string_normalization=not skip_string_normalization,
+        magic_trailing_comma=not skip_magic_trailing_comma,
+        experimental_string_processing=experimental_string_processing,
     )
     if config and verbose:
         out(f"Using configuration from {config}.", bold=False, fg="blue")
@@ -511,8 +613,10 @@ def main(
         verbose=verbose,
         include=include,
         exclude=exclude,
+        extend_exclude=extend_exclude,
         force_exclude=force_exclude,
         report=report,
+        stdin_filename=stdin_filename,
     )
 
     path_empty(
@@ -548,58 +652,62 @@ def get_sources(
     src: Tuple[str, ...],
     quiet: bool,
     verbose: bool,
-    include: str,
-    exclude: str,
-    force_exclude: Optional[str],
+    include: Pattern[str],
+    exclude: Pattern[str],
+    extend_exclude: Optional[Pattern[str]],
+    force_exclude: Optional[Pattern[str]],
     report: "Report",
+    stdin_filename: Optional[str],
 ) -> Set[Path]:
     """Compute the set of files to be formatted."""
-    try:
-        include_regex = re_compile_maybe_verbose(include)
-    except re.error:
-        err(f"Invalid regular expression for include given: {include!r}")
-        ctx.exit(2)
-    try:
-        exclude_regex = re_compile_maybe_verbose(exclude)
-    except re.error:
-        err(f"Invalid regular expression for exclude given: {exclude!r}")
-        ctx.exit(2)
-    try:
-        force_exclude_regex = (
-            re_compile_maybe_verbose(force_exclude) if force_exclude else None
-        )
-    except re.error:
-        err(f"Invalid regular expression for force_exclude given: {force_exclude!r}")
-        ctx.exit(2)
 
     root = find_project_root(src)
     sources: Set[Path] = set()
     path_empty(src, "No Path provided. Nothing to do 😴", quiet, verbose, ctx)
-    exclude_regexes = [exclude_regex]
-    if force_exclude_regex is not None:
-        exclude_regexes.append(force_exclude_regex)
+    gitignore = get_gitignore(root)
 
     for s in src:
-        p = Path(s)
-        if p.is_dir():
+        if s == "-" and stdin_filename:
+            p = Path(stdin_filename)
+            is_stdin = True
+        else:
+            p = Path(s)
+            is_stdin = False
+
+        if is_stdin or p.is_file():
+            normalized_path = normalize_path_maybe_ignore(p, root, report)
+            if normalized_path is None:
+                continue
+
+            normalized_path = "/" + normalized_path
+            # Hard-exclude any files that matches the `--force-exclude` regex.
+            if force_exclude:
+                force_exclude_match = force_exclude.search(normalized_path)
+            else:
+                force_exclude_match = None
+            if force_exclude_match and force_exclude_match.group(0):
+                report.path_ignored(p, "matches the --force-exclude regular expression")
+                continue
+
+            if is_stdin:
+                p = Path(f"{STDIN_PLACEHOLDER}{str(p)}")
+
+            sources.add(p)
+        elif p.is_dir():
             sources.update(
                 gen_python_files(
                     p.iterdir(),
                     root,
-                    include_regex,
-                    exclude_regexes,
+                    include,
+                    exclude,
+                    extend_exclude,
+                    force_exclude,
                     report,
-                    get_gitignore(root),
+                    gitignore,
                 )
             )
         elif s == "-":
             sources.add(p)
-        elif p.is_file():
-            sources.update(
-                gen_python_files(
-                    [p], root, None, exclude_regexes, report, get_gitignore(root)
-                )
-            )
         else:
             err(f"invalid path: {s}")
     return sources
@@ -611,10 +719,9 @@ def path_empty(
     """
     Exit if there is no `src` provided for formatting
     """
-    if len(src) == 0:
-        if verbose or not quiet:
-            out(msg)
-            ctx.exit(0)
+    if not src and (verbose or not quiet):
+        out(msg)
+        ctx.exit(0)
 
 
 def reformat_one(
@@ -627,15 +734,27 @@ def reformat_one(
     """
     try:
         changed = Changed.NO
-        if not src.is_file() and str(src) == "-":
+
+        if str(src) == "-":
+            is_stdin = True
+        elif str(src).startswith(STDIN_PLACEHOLDER):
+            is_stdin = True
+            # Use the original name again in case we want to print something
+            # to the user
+            src = Path(str(src)[len(STDIN_PLACEHOLDER) :])
+        else:
+            is_stdin = False
+
+        if is_stdin:
             if format_stdin_to_stdout(fast=fast, write_back=write_back, mode=mode):
                 changed = Changed.YES
         else:
             cache: Cache = {}
-            if write_back != WriteBack.DIFF:
+            if write_back not in (WriteBack.DIFF, WriteBack.COLOR_DIFF):
                 cache = read_cache(mode)
                 res_src = src.resolve()
-                if res_src in cache and cache[res_src] == get_cache_info(res_src):
+                res_src_s = str(res_src)
+                if res_src_s in cache and cache[res_src_s] == get_cache_info(res_src):
                     changed = Changed.CACHED
             if changed is not Changed.CACHED and format_file_in_place(
                 src, fast=fast, write_back=write_back, mode=mode
@@ -647,6 +766,8 @@ def reformat_one(
                 write_cache(cache, [src], mode)
         report.done(src, changed)
     except Exception as exc:
+        if report.verbose:
+            traceback.print_exc()
         report.failed(src, str(exc))
 
 
@@ -659,13 +780,13 @@ def reformat_many(
     worker_count = os.cpu_count()
     if sys.platform == "win32":
         # Work around https://bugs.python.org/issue26903
-        worker_count = min(worker_count, 61)
+        worker_count = min(worker_count, 60)
     try:
         executor = ProcessPoolExecutor(max_workers=worker_count)
-    except OSError:
+    except (ImportError, OSError):
         # we arrive here if the underlying system does not support multi-processing
-        # like in AWS Lambda, in which case we gracefully fallback to
-        # a ThreadPollExecutor with just a single worker (more workers would not do us
+        # like in AWS Lambda or Termux, in which case we gracefully fallback to
+        # a ThreadPoolExecutor with just a single worker (more workers would not do us
         # any good due to the Global Interpreter Lock)
         executor = ThreadPoolExecutor(max_workers=1)
 
@@ -704,7 +825,7 @@ async def schedule_formatting(
     :func:`format_file_in_place`.
     """
     cache: Cache = {}
-    if write_back != WriteBack.DIFF:
+    if write_back not in (WriteBack.DIFF, WriteBack.COLOR_DIFF):
         cache = read_cache(mode)
         sources, cached = filter_cached(cache, sources)
         for src in sorted(cached):
@@ -715,7 +836,7 @@ async def schedule_formatting(
     cancelled = []
     sources_to_cache = []
     lock = None
-    if write_back == WriteBack.DIFF:
+    if write_back in (WriteBack.DIFF, WriteBack.COLOR_DIFF):
         # For diff output, we need locks to ensure we don't interleave output
         # from different processes.
         manager = Manager()
@@ -728,7 +849,7 @@ async def schedule_formatting(
         ): src
         for src in sorted(sources)
     }
-    pending: Iterable["asyncio.Future[bool]"] = tasks.keys()
+    pending = tasks.keys()
     try:
         loop.add_signal_handler(signal.SIGINT, cancel, pending)
         loop.add_signal_handler(signal.SIGTERM, cancel, pending)
@@ -791,7 +912,7 @@ def format_file_in_place(
         dst_name = f"{src}\t{now} +0000"
         diff_contents = diff(src_contents, dst_contents, src_name, dst_name)
 
-        if write_back == write_back.COLOR_DIFF:
+        if write_back == WriteBack.COLOR_DIFF:
             diff_contents = color_diff(diff_contents)
 
         with lock or nullcontext():
@@ -814,9 +935,9 @@ def color_diff(contents: str) -> str:
     for i, line in enumerate(lines):
         if line.startswith("+++") or line.startswith("---"):
             line = "\033[1;37m" + line + "\033[0m"  # bold white, reset
-        if line.startswith("@@"):
+        elif line.startswith("@@"):
             line = "\033[36m" + line + "\033[0m"  # cyan, reset
-        if line.startswith("+"):
+        elif line.startswith("+"):
             line = "\033[32m" + line + "\033[0m"  # green, reset
         elif line.startswith("-"):
             line = "\033[31m" + line + "\033[0m"  # red, reset
@@ -826,30 +947,22 @@ def color_diff(contents: str) -> str:
 
 def wrap_stream_for_windows(
     f: io.TextIOWrapper,
-) -> Union[io.TextIOWrapper, "colorama.AnsiToWin32.AnsiToWin32"]:
+) -> Union[io.TextIOWrapper, "colorama.AnsiToWin32"]:
     """
-    Wrap the stream in colorama's wrap_stream so colors are shown on Windows.
+    Wrap stream with colorama's wrap_stream so colors are shown on Windows.
 
-    If `colorama` is not found, then no change is made. If `colorama` does
-    exist, then it handles the logic to determine whether or not to change
-    things.
+    If `colorama` is unavailable, the original stream is returned unmodified.
+    Otherwise, the `wrap_stream()` function determines whether the stream needs
+    to be wrapped for a Windows environment and will accordingly either return
+    an `AnsiToWin32` wrapper or the original stream.
     """
     try:
-        from colorama import initialise
-
-        # We set `strip=False` so that we can don't have to modify
-        # test_express_diff_with_color.
-        f = initialise.wrap_stream(
-            f, convert=None, strip=False, autoreset=False, wrap=True
-        )
-
-        # wrap_stream returns a `colorama.AnsiToWin32.AnsiToWin32` object
-        # which does not have a `detach()` method. So we fake one.
-        f.detach = lambda *args, **kwargs: None  # type: ignore
+        from colorama.initialise import wrap_stream
     except ImportError:
-        pass
-
-    return f
+        return f
+    else:
+        # Set `strip=False` to avoid needing to modify test_express_diff_with_color.
+        return wrap_stream(f, convert=None, strip=False, autoreset=False, wrap=True)
 
 
 def format_stdin_to_stdout(
@@ -890,13 +1003,13 @@ def format_stdin_to_stdout(
 
 
 def format_file_contents(src_contents: str, *, fast: bool, mode: Mode) -> FileContent:
-    """Reformat contents a file and return new contents.
+    """Reformat contents of a file and return new contents.
 
     If `fast` is False, additionally confirm that the reformatted code is
     valid by calling :func:`assert_equivalent` and :func:`assert_stable` on it.
     `mode` is passed to :func:`format_str`.
     """
-    if src_contents.strip() == "":
+    if not src_contents.strip():
         raise NothingChanged
 
     dst_contents = format_str(src_contents, mode=mode)
@@ -905,7 +1018,17 @@ def format_file_contents(src_contents: str, *, fast: bool, mode: Mode) -> FileCo
 
     if not fast:
         assert_equivalent(src_contents, dst_contents)
-        assert_stable(src_contents, dst_contents, mode=mode)
+
+        # Forced second pass to work around optional trailing commas (becoming
+        # forced trailing commas on pass 2) interacting differently with optional
+        # parentheses.  Admittedly ugly.
+        dst_contents_pass2 = format_str(dst_contents, mode=mode)
+        if dst_contents != dst_contents_pass2:
+            dst_contents = dst_contents_pass2
+            assert_equivalent(src_contents, dst_contents, pass_num=2)
+            assert_stable(src_contents, dst_contents, mode=mode)
+        # Note: no need to explicitly call `assert_stable` if `dst_contents` was
+        # the same as `dst_contents_pass2`.
     return dst_contents
 
 
@@ -916,11 +1039,12 @@ def format_str(src_contents: str, *, mode: Mode) -> FileContent:
     allowed.  Example:
 
     >>> import black
-    >>> print(black.format_str("def f(arg:str='')->None:...", mode=Mode()))
+    >>> print(black.format_str("def f(arg:str='')->None:...", mode=black.Mode()))
     def f(arg: str = "") -> None:
         ...
 
     A more complex example:
+
     >>> print(
     ...   black.format_str(
     ...     "def f(arg:str='')->None: hey",
@@ -947,13 +1071,12 @@ def format_str(src_contents: str, *, mode: Mode) -> FileContent:
         versions = detect_target_versions(src_node)
     normalize_fmt_off(src_node)
     lines = LineGenerator(
+        mode=mode,
         remove_u_prefix="unicode_literals" in future_imports
         or supports_feature(versions, Feature.UNICODE_LITERALS),
-        is_pyi=mode.is_pyi,
-        normalize_strings=mode.string_normalization,
     )
     elt = EmptyLineTracker(is_pyi=mode.is_pyi)
-    empty_line = Line()
+    empty_line = Line(mode=mode)
     after = 0
     split_line_features = {
         feature
@@ -965,10 +1088,7 @@ def format_str(src_contents: str, *, mode: Mode) -> FileContent:
         before, after = elt.maybe_empty_lines(current_line)
         dst_contents.append(str(empty_line) * before)
         for line in transform_line(
-            current_line,
-            line_length=mode.line_length,
-            normalize_strings=mode.string_normalization,
-            features=split_line_features,
+            current_line, mode=mode, features=split_line_features
         ):
             dst_contents.append(str(line))
     return "".join(dst_contents)
@@ -1032,7 +1152,7 @@ def get_grammars(target_versions: Set[TargetVersion]) -> List[Grammar]:
 
 def lib2to3_parse(src_txt: str, target_versions: Iterable[TargetVersion] = ()) -> Node:
     """Given a string with source, return the lib2to3 Node."""
-    if src_txt[-1:] != "\n":
+    if not src_txt.endswith("\n"):
         src_txt += "\n"
 
     for grammar in get_grammars(set(target_versions)):
@@ -1255,6 +1375,7 @@ class BracketTracker:
     previous: Optional[Leaf] = None
     _for_loop_depths: List[int] = field(default_factory=list)
     _lambda_argument_depths: List[int] = field(default_factory=list)
+    invisible: List[Leaf] = field(default_factory=list)
 
     def mark(self, leaf: Leaf) -> None:
         """Mark `leaf` with bracket-related metadata. Keep track of delimiters.
@@ -1278,8 +1399,16 @@ class BracketTracker:
         self.maybe_decrement_after_lambda_arguments(leaf)
         if leaf.type in CLOSING_BRACKETS:
             self.depth -= 1
-            opening_bracket = self.bracket_match.pop((self.depth, leaf.type))
+            try:
+                opening_bracket = self.bracket_match.pop((self.depth, leaf.type))
+            except KeyError as e:
+                raise BracketMatchError(
+                    "Unable to match a closing bracket to the following opening"
+                    f" bracket: {leaf}"
+                ) from e
             leaf.opening_bracket = opening_bracket
+            if not leaf.value:
+                self.invisible.append(leaf)
         leaf.bracket_depth = self.depth
         if self.depth == 0:
             delim = is_split_before_delimiter(leaf, self.previous)
@@ -1292,6 +1421,8 @@ class BracketTracker:
         if leaf.type in OPENING_BRACKETS:
             self.bracket_match[self.depth, BRACKET[leaf.type]] = leaf
             self.depth += 1
+            if not leaf.value:
+                self.invisible.append(leaf)
         self.previous = leaf
         self.maybe_increment_lambda_arguments(leaf)
         self.maybe_increment_for_loop_variable(leaf)
@@ -1381,13 +1512,15 @@ class BracketTracker:
 class Line:
     """Holds leaves and comments. Can be printed with `str(line)`."""
 
+    mode: Mode
     depth: int = 0
     leaves: List[Leaf] = field(default_factory=list)
     # keys ordered like `leaves`
     comments: Dict[LeafID, List[Leaf]] = field(default_factory=dict)
     bracket_tracker: BracketTracker = field(default_factory=BracketTracker)
     inside_brackets: bool = False
-    should_explode: bool = False
+    should_split_rhs: bool = False
+    magic_trailing_comma: Optional[Leaf] = None
 
     def append(self, leaf: Leaf, preformatted: bool = False) -> None:
         """Add a new `leaf` to the end of the line.
@@ -1413,7 +1546,11 @@ class Line:
             )
         if self.inside_brackets or not preformatted:
             self.bracket_tracker.mark(leaf)
-            self.maybe_remove_trailing_comma(leaf)
+            if self.mode.magic_trailing_comma:
+                if self.has_magic_trailing_comma(leaf):
+                    self.magic_trailing_comma = leaf
+            elif self.has_magic_trailing_comma(leaf, ensure_removable=True):
+                self.remove_trailing_comma()
         if not self.append_comment(leaf):
             self.leaves.append(leaf)
 
@@ -1465,69 +1602,6 @@ class Line:
             Leaf(token.DOT, ".") for _ in range(3)
         ]
 
-    @property
-    def is_collection_with_optional_trailing_comma(self) -> bool:
-        """Is this line a collection literal with a trailing comma that's optional?
-
-        Note that the trailing comma in a 1-tuple is not optional.
-        """
-        if not self.leaves or len(self.leaves) < 4:
-            return False
-
-        # Look for and address a trailing colon.
-        if self.leaves[-1].type == token.COLON:
-            closer = self.leaves[-2]
-            close_index = -2
-        else:
-            closer = self.leaves[-1]
-            close_index = -1
-        if closer.type not in CLOSING_BRACKETS or self.inside_brackets:
-            return False
-
-        if closer.type == token.RPAR:
-            # Tuples require an extra check, because if there's only
-            # one element in the tuple removing the comma unmakes the
-            # tuple.
-            #
-            # We also check for parens before looking for the trailing
-            # comma because in some cases (eg assigning a dict
-            # literal) the literal gets wrapped in temporary parens
-            # during parsing. This case is covered by the
-            # collections.py test data.
-            opener = closer.opening_bracket
-            for _open_index, leaf in enumerate(self.leaves):
-                if leaf is opener:
-                    break
-
-            else:
-                # Couldn't find the matching opening paren, play it safe.
-                return False
-
-            commas = 0
-            comma_depth = self.leaves[close_index - 1].bracket_depth
-            for leaf in self.leaves[_open_index + 1 : close_index]:
-                if leaf.bracket_depth == comma_depth and leaf.type == token.COMMA:
-                    commas += 1
-            if commas > 1:
-                # We haven't looked yet for the trailing comma because
-                # we might also have caught noop parens.
-                return self.leaves[close_index - 1].type == token.COMMA
-
-            elif commas == 1:
-                return False  # it's either a one-tuple or didn't have a trailing comma
-
-            if self.leaves[close_index - 1].type in CLOSING_BRACKETS:
-                close_index -= 1
-                closer = self.leaves[close_index]
-                if closer.type == token.RPAR:
-                    # TODO: this is a gut feeling. Will we ever see this?
-                    return False
-
-        if self.leaves[close_index - 1].type != token.COMMA:
-            return False
-
-        return True
-
     @property
     def is_def(self) -> bool:
         """Is this a function definition? (Also returns True for async defs.)"""
@@ -1633,8 +1707,10 @@ class Line:
         # one line in the original code.
 
         # Grab the first and last line numbers, skipping generated leaves
-        first_line = next((l.lineno for l in self.leaves if l.lineno != 0), 0)
-        last_line = next((l.lineno for l in reversed(self.leaves) if l.lineno != 0), 0)
+        first_line = next((leaf.lineno for leaf in self.leaves if leaf.lineno != 0), 0)
+        last_line = next(
+            (leaf.lineno for leaf in reversed(self.leaves) if leaf.lineno != 0), 0
+        )
 
         if first_line == last_line:
             # We look at the last two leaves since a comma or an
@@ -1650,42 +1726,38 @@ class Line:
     def contains_multiline_strings(self) -> bool:
         return any(is_multiline_string(leaf) for leaf in self.leaves)
 
-    def maybe_remove_trailing_comma(self, closing: Leaf) -> bool:
-        """Remove trailing comma if there is one and it's safe."""
-        if not (self.leaves and self.leaves[-1].type == token.COMMA):
-            return False
-
-        # We remove trailing commas only in the case of importing a
-        # single name from a module.
+    def has_magic_trailing_comma(
+        self, closing: Leaf, ensure_removable: bool = False
+    ) -> bool:
+        """Return True if we have a magic trailing comma, that is when:
+        - there's a trailing comma here
+        - it's not a one-tuple
+        Additionally, if ensure_removable:
+        - it's not from square bracket indexing
+        """
         if not (
-            self.leaves
-            and self.is_import
-            and len(self.leaves) > 4
+            closing.type in CLOSING_BRACKETS
+            and self.leaves
             and self.leaves[-1].type == token.COMMA
-            and closing.type in CLOSING_BRACKETS
-            and self.leaves[-4].type == token.NAME
-            and (
-                # regular `from foo import bar,`
-                self.leaves[-4].value == "import"
-                # `from foo import (bar as baz,)
-                or (
-                    len(self.leaves) > 6
-                    and self.leaves[-6].value == "import"
-                    and self.leaves[-3].value == "as"
-                )
-                # `from foo import bar as baz,`
-                or (
-                    len(self.leaves) > 5
-                    and self.leaves[-5].value == "import"
-                    and self.leaves[-3].value == "as"
-                )
-            )
-            and closing.type == token.RPAR
         ):
             return False
 
-        self.remove_trailing_comma()
-        return True
+        if closing.type == token.RBRACE:
+            return True
+
+        if closing.type == token.RSQB:
+            if not ensure_removable:
+                return True
+            comma = self.leaves[-1]
+            return bool(comma.parent and comma.parent.type == syms.listmaker)
+
+        if self.is_import:
+            return True
+
+        if not is_one_tuple_between(closing.opening_bracket, closing, self.leaves):
+            return True
+
+        return False
 
     def append_comment(self, comment: Leaf) -> bool:
         """Add an inline or standalone comment to the line."""
@@ -1756,9 +1828,11 @@ class Line:
 
     def clone(self) -> "Line":
         return Line(
+            mode=self.mode,
             depth=self.depth,
             inside_brackets=self.inside_brackets,
-            should_explode=self.should_explode,
+            should_split_rhs=self.should_split_rhs,
+            magic_trailing_comma=self.magic_trailing_comma,
         )
 
     def __str__(self) -> str:
@@ -1864,6 +1938,10 @@ class EmptyLineTracker:
             return 0, 0
 
         if self.previous_line.is_decorator:
+            if self.is_pyi and current_line.is_stub_class:
+                # Insert an empty line after a decorated stub class
+                return 0, 1
+
             return 0, 0
 
         if self.previous_line.depth < current_line.depth and (
@@ -1887,8 +1965,11 @@ class EmptyLineTracker:
                     newlines = 0
                 else:
                     newlines = 1
-            elif current_line.is_def and not self.previous_line.is_def:
-                # Blank line between a block of functions and a block of non-functions
+            elif (
+                current_line.is_def or current_line.is_decorator
+            ) and not self.previous_line.is_def:
+                # Blank line between a block of functions (maybe with preceding
+                # decorators) and a block of non-functions
                 newlines = 1
             else:
                 newlines = 0
@@ -1907,10 +1988,9 @@ class LineGenerator(Visitor[Line]):
     in ways that will no longer stringify to valid Python code on the tree.
     """
 
-    is_pyi: bool = False
-    normalize_strings: bool = True
-    current_line: Line = field(default_factory=Line)
+    mode: Mode
     remove_u_prefix: bool = False
+    current_line: Line = field(init=False)
 
     def line(self, indent: int = 0) -> Iterator[Line]:
         """Generate a line.
@@ -1925,7 +2005,7 @@ class LineGenerator(Visitor[Line]):
             return  # Line is empty, don't emit. Creating a new one unnecessary.
 
         complete_line = self.current_line
-        self.current_line = Line(depth=complete_line.depth + indent)
+        self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
         yield complete_line
 
     def visit_default(self, node: LN) -> Iterator[Line]:
@@ -1949,7 +2029,7 @@ class LineGenerator(Visitor[Line]):
                     yield from self.line()
 
             normalize_prefix(node, inside_brackets=any_open_brackets)
-            if self.normalize_strings and node.type == token.STRING:
+            if self.mode.string_normalization and node.type == token.STRING:
                 normalize_string_prefix(node, remove_u_prefix=self.remove_u_prefix)
                 normalize_string_quotes(node)
             if node.type == token.NUMBER:
@@ -2001,16 +2081,18 @@ class LineGenerator(Visitor[Line]):
 
     def visit_suite(self, node: Node) -> Iterator[Line]:
         """Visit a suite."""
-        if self.is_pyi and is_stub_suite(node):
+        if self.mode.is_pyi and is_stub_suite(node):
             yield from self.visit(node.children[2])
         else:
             yield from self.visit_default(node)
 
     def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
         """Visit a statement without nested statements."""
+        if first_child_is_arith(node):
+            wrap_in_parentheses(node, node.children[0], visible=False)
         is_suite_like = node.parent and node.parent.type in STATEMENT
         if is_suite_like:
-            if self.is_pyi and is_stub_body(node):
+            if self.mode.is_pyi and is_stub_body(node):
                 yield from self.visit_default(node)
             else:
                 yield from self.line(+1)
@@ -2018,7 +2100,11 @@ class LineGenerator(Visitor[Line]):
                 yield from self.line(-1)
 
         else:
-            if not self.is_pyi or not node.parent or not is_stub_suite(node.parent):
+            if (
+                not self.mode.is_pyi
+                or not node.parent
+                or not is_stub_suite(node.parent)
+            ):
                 yield from self.line()
             yield from self.visit_default(node)
 
@@ -2075,19 +2161,52 @@ class LineGenerator(Visitor[Line]):
         yield from self.visit_default(node)
 
     def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
-        # Check if it's a docstring
-        if prev_siblings_are(
-            leaf.parent, [None, token.NEWLINE, token.INDENT, syms.simple_stmt]
-        ) and is_multiline_string(leaf):
-            prefix = "    " * self.current_line.depth
-            docstring = fix_docstring(leaf.value[3:-3], prefix)
-            leaf.value = leaf.value[0:3] + docstring + leaf.value[-3:]
-            normalize_string_quotes(leaf)
+        if is_docstring(leaf) and "\\\n" not in leaf.value:
+            # We're ignoring docstrings with backslash newline escapes because changing
+            # indentation of those changes the AST representation of the code.
+            prefix = get_string_prefix(leaf.value)
+            docstring = leaf.value[len(prefix) :]  # Remove the prefix
+            quote_char = docstring[0]
+            # A natural way to remove the outer quotes is to do:
+            #   docstring = docstring.strip(quote_char)
+            # but that breaks on """""x""" (which is '""x').
+            # So we actually need to remove the first character and the next two
+            # characters but only if they are the same as the first.
+            quote_len = 1 if docstring[1] != quote_char else 3
+            docstring = docstring[quote_len:-quote_len]
+
+            if is_multiline_string(leaf):
+                indent = " " * 4 * self.current_line.depth
+                docstring = fix_docstring(docstring, indent)
+            else:
+                docstring = docstring.strip()
+
+            if docstring:
+                # Add some padding if the docstring starts / ends with a quote mark.
+                if docstring[0] == quote_char:
+                    docstring = " " + docstring
+                if docstring[-1] == quote_char:
+                    docstring += " "
+                if docstring[-1] == "\\":
+                    backslash_count = len(docstring) - len(docstring.rstrip("\\"))
+                    if backslash_count % 2:
+                        # Odd number of tailing backslashes, add some padding to
+                        # avoid escaping the closing string quote.
+                        docstring += " "
+            else:
+                # Add some padding if the docstring is empty.
+                docstring = " "
+
+            # We could enforce triple quotes at this point.
+            quote = quote_char * quote_len
+            leaf.value = prefix + quote + docstring + quote
 
         yield from self.visit_default(leaf)
 
     def __post_init__(self) -> None:
         """You are in a twisty little maze of passages."""
+        self.current_line = Line(mode=self.mode)
+
         v = self.visit_stmt
         Ø: Set[str] = set()
         self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
@@ -2201,6 +2320,9 @@ def whitespace(leaf: Leaf, *, complex_subscript: bool) -> str:  # noqa: C901
         ):
             # Python 2 print chevron
             return NO
+        elif prevp.type == token.AT and p.parent and p.parent.type == syms.decorator:
+            # no space in decorators
+            return NO
 
     elif prev.type in OPENING_BRACKETS:
         return NO
@@ -2527,6 +2649,8 @@ def is_split_before_delimiter(leaf: Leaf, previous: Optional[Leaf] = None) -> Pr
 
 
 FMT_OFF = {"# fmt: off", "# fmt:off", "# yapf: disable"}
+FMT_SKIP = {"# fmt: skip", "# fmt:skip"}
+FMT_PASS = {*FMT_OFF, *FMT_SKIP}
 FMT_ON = {"# fmt: on", "# fmt:on", "# yapf: enable"}
 
 
@@ -2581,7 +2705,7 @@ def list_comments(prefix: str, *, is_endmarker: bool) -> List[ProtoComment]:
     consumed = 0
     nlines = 0
     ignored_lines = 0
-    for index, line in enumerate(prefix.split("\n")):
+    for index, line in enumerate(re.split("\r?\n", prefix)):
         consumed += len(line) + 1  # adding the length of the split '\n'
         line = line.lstrip()
         if not line:
@@ -2622,16 +2746,20 @@ def make_comment(content: str) -> str:
 
     if content[0] == "#":
         content = content[1:]
+    NON_BREAKING_SPACE = " "
+    if (
+        content
+        and content[0] == NON_BREAKING_SPACE
+        and not content.lstrip().startswith("type:")
+    ):
+        content = " " + content[1:]  # Replace NBSP by a simple space
     if content and content[0] not in " !:#'%":
         content = " " + content
     return "#" + content
 
 
 def transform_line(
-    line: Line,
-    line_length: int,
-    normalize_strings: bool,
-    features: Collection[Feature] = (),
+    line: Line, mode: Mode, features: Collection[Feature] = ()
 ) -> Iterator[Line]:
     """Transform a `line`, potentially splitting it into many lines.
 
@@ -2647,7 +2775,7 @@ def transform_line(
 
     def init_st(ST: Type[StringTransformer]) -> StringTransformer:
         """Initialize StringTransformer"""
-        return ST(line_length, normalize_strings)
+        return ST(mode.line_length, mode.string_normalization)
 
     string_merge = init_st(StringMerger)
     string_paren_strip = init_st(StringParenStripper)
@@ -2657,73 +2785,81 @@ def transform_line(
     transformers: List[Transformer]
     if (
         not line.contains_uncollapsable_type_comments()
-        and not line.should_explode
-        and not line.is_collection_with_optional_trailing_comma
+        and not line.should_split_rhs
+        and not line.magic_trailing_comma
         and (
-            is_line_short_enough(line, line_length=line_length, line_str=line_str)
+            is_line_short_enough(line, line_length=mode.line_length, line_str=line_str)
             or line.contains_unsplittable_type_ignore()
         )
-        and not (line.contains_standalone_comments() and line.inside_brackets)
+        and not (line.inside_brackets and line.contains_standalone_comments())
     ):
         # Only apply basic string preprocessing, since lines shouldn't be split here.
-        transformers = [string_merge, string_paren_strip]
+        if mode.experimental_string_processing:
+            transformers = [string_merge, string_paren_strip]
+        else:
+            transformers = []
     elif line.is_def:
         transformers = [left_hand_split]
     else:
 
         def rhs(line: Line, features: Collection[Feature]) -> Iterator[Line]:
-            for omit in generate_trailers_to_omit(line, line_length):
-                lines = list(right_hand_split(line, line_length, features, omit=omit))
-                if is_line_short_enough(lines[0], line_length=line_length):
+            """Wraps calls to `right_hand_split`.
+
+            The calls increasingly `omit` right-hand trailers (bracket pairs with
+            content), meaning the trailers get glued together to split on another
+            bracket pair instead.
+            """
+            for omit in generate_trailers_to_omit(line, mode.line_length):
+                lines = list(
+                    right_hand_split(line, mode.line_length, features, omit=omit)
+                )
+                # Note: this check is only able to figure out if the first line of the
+                # *current* transformation fits in the line length.  This is true only
+                # for simple cases.  All others require running more transforms via
+                # `transform_line()`.  This check doesn't know if those would succeed.
+                if is_line_short_enough(lines[0], line_length=mode.line_length):
                     yield from lines
                     return
 
             # All splits failed, best effort split with no omits.
             # This mostly happens to multiline strings that are by definition
-            # reported as not fitting a single line.
-            # line_length=1 here was historically a bug that somehow became a feature.
-            # See #762 and #781 for the full story.
-            yield from right_hand_split(line, line_length=1, features=features)
-
-        if line.inside_brackets:
-            transformers = [
-                string_merge,
-                string_paren_strip,
-                delimiter_split,
-                standalone_comment_split,
-                string_split,
-                string_paren_wrap,
-                rhs,
-            ]
+            # reported as not fitting a single line, as well as lines that contain
+            # trailing commas (those have to be exploded).
+            yield from right_hand_split(
+                line, line_length=mode.line_length, features=features
+            )
+
+        if mode.experimental_string_processing:
+            if line.inside_brackets:
+                transformers = [
+                    string_merge,
+                    string_paren_strip,
+                    string_split,
+                    delimiter_split,
+                    standalone_comment_split,
+                    string_paren_wrap,
+                    rhs,
+                ]
+            else:
+                transformers = [
+                    string_merge,
+                    string_paren_strip,
+                    string_split,
+                    string_paren_wrap,
+                    rhs,
+                ]
         else:
-            transformers = [
-                string_merge,
-                string_paren_strip,
-                string_split,
-                string_paren_wrap,
-                rhs,
-            ]
+            if line.inside_brackets:
+                transformers = [delimiter_split, standalone_comment_split, rhs]
+            else:
+                transformers = [rhs]
 
     for transform in transformers:
         # We are accumulating lines in `result` because we might want to abort
         # mission and return the original line in the end, or attempt a different
         # split altogether.
-        result: List[Line] = []
         try:
-            for l in transform(line, features):
-                if str(l).strip("\n") == line_str:
-                    raise CannotTransform(
-                        "Line transformer returned an unchanged result"
-                    )
-
-                result.extend(
-                    transform_line(
-                        l,
-                        line_length=line_length,
-                        normalize_strings=normalize_strings,
-                        features=features,
-                    )
-                )
+            result = run_transformer(line, transform, mode, features, line_str=line_str)
         except CannotTransform:
             continue
         else:
@@ -2752,7 +2888,7 @@ class StringTransformer(ABC):
         StringTransformer to be applied?
 
     Transformations:
-        If the given Line meets all of the above requirments, which string
+        If the given Line meets all of the above requirements, which string
         transformations can you expect to be applied to it by this
         StringTransformer?
 
@@ -2764,6 +2900,7 @@ class StringTransformer(ABC):
 
     line_length: int
     normalize_strings: bool
+    __name__ = "StringTransformer"
 
     @abstractmethod
     def do_match(self, line: Line) -> TMatchResult:
@@ -2918,11 +3055,8 @@ class StringMerger(CustomSplitMapMixin, StringTransformer):
     """StringTransformer that merges strings together.
 
     Requirements:
-        (A) The line contains adjacent strings such that at most one substring
-        has inline comments AND none of those inline comments are pragmas AND
-        the set of all substring prefixes is either of length 1 or equal to
-        {"", "f"} AND none of the substrings are raw strings (i.e. are prefixed
-        with 'r').
+        (A) The line contains adjacent strings such that ALL of the validation checks
+        listed in StringMerger.__validate_msg(...)'s docstring pass.
             OR
         (B) The line contains a string which uses line continuation backslashes.
 
@@ -3010,7 +3144,7 @@ class StringMerger(CustomSplitMapMixin, StringTransformer):
             )
 
         new_line = line.clone()
-        new_line.comments = line.comments
+        new_line.comments = line.comments.copy()
         append_leaves(new_line, line, LL)
 
         new_string_leaf = new_line.leaves[string_idx]
@@ -3171,6 +3305,7 @@ class StringMerger(CustomSplitMapMixin, StringTransformer):
             * Ok(None), if ALL validation checks (listed below) pass.
                 OR
             * Err(CannotTransform), if any of the following are true:
+                - The target string group does not contain ANY stand-alone comments.
                 - The target string is not in a string group (i.e. it has no
                   adjacent strings).
                 - The string group has more than one inline comment.
@@ -3179,6 +3314,26 @@ class StringMerger(CustomSplitMapMixin, StringTransformer):
                   length greater than one and is not equal to {"", "f"}.
                 - The string group consists of raw strings.
         """
+        # We first check for "inner" stand-alone comments (i.e. stand-alone
+        # comments that have a string leaf before them AND after them).
+        for inc in [1, -1]:
+            i = string_idx
+            found_sa_comment = False
+            is_valid_index = is_valid_index_factory(line.leaves)
+            while is_valid_index(i) and line.leaves[i].type in [
+                token.STRING,
+                STANDALONE_COMMENT,
+            ]:
+                if line.leaves[i].type == STANDALONE_COMMENT:
+                    found_sa_comment = True
+                elif found_sa_comment:
+                    return TErr(
+                        "StringMerger does NOT merge string groups which contain "
+                        "stand-alone comments."
+                    )
+
+                i += inc
+
         num_of_inline_string_comments = 0
         set_of_prefixes = set()
         num_of_strings = 0
@@ -3227,8 +3382,11 @@ class StringParenStripper(StringTransformer):
 
     Requirements:
         The line contains a string which is surrounded by parentheses and:
-            - The target string is NOT the only argument to a function call).
-            - The RPAR is NOT followed by an attribute access (i.e. a dot).
+            - The target string is NOT the only argument to a function call.
+            - The target string is NOT a "pointless" string.
+            - If the target string contains a PERCENT, the brackets are not
+              preceeded or followed by an operator with higher precedence than
+              PERCENT.
 
     Transformations:
         The parentheses mentioned in the 'Requirements' section are stripped.
@@ -3249,6 +3407,14 @@ class StringParenStripper(StringTransformer):
             if leaf.type != token.STRING:
                 continue
 
+            # If this is a "pointless" string...
+            if (
+                leaf.parent
+                and leaf.parent.parent
+                and leaf.parent.parent.type == syms.simple_stmt
+            ):
+                continue
+
             # Should be preceded by a non-empty LPAR...
             if (
                 not is_valid_index(idx - 1)
@@ -3271,14 +3437,51 @@ class StringParenStripper(StringTransformer):
             string_parser = StringParser()
             next_idx = string_parser.parse(LL, string_idx)
 
+            # if the leaves in the parsed string include a PERCENT, we need to
+            # make sure the initial LPAR is NOT preceded by an operator with
+            # higher or equal precedence to PERCENT
+            if is_valid_index(idx - 2):
+                # mypy can't quite follow unless we name this
+                before_lpar = LL[idx - 2]
+                if token.PERCENT in {leaf.type for leaf in LL[idx - 1 : next_idx]} and (
+                    (
+                        before_lpar.type
+                        in {
+                            token.STAR,
+                            token.AT,
+                            token.SLASH,
+                            token.DOUBLESLASH,
+                            token.PERCENT,
+                            token.TILDE,
+                            token.DOUBLESTAR,
+                            token.AWAIT,
+                            token.LSQB,
+                            token.LPAR,
+                        }
+                    )
+                    or (
+                        # only unary PLUS/MINUS
+                        before_lpar.parent
+                        and before_lpar.parent.type == syms.factor
+                        and (before_lpar.type in {token.PLUS, token.MINUS})
+                    )
+                ):
+                    continue
+
             # Should be followed by a non-empty RPAR...
             if (
                 is_valid_index(next_idx)
                 and LL[next_idx].type == token.RPAR
                 and not is_empty_rpar(LL[next_idx])
             ):
-                # That RPAR should NOT be followed by a '.' symbol.
-                if is_valid_index(next_idx + 1) and LL[next_idx + 1].type == token.DOT:
+                # That RPAR should NOT be followed by anything with higher
+                # precedence than PERCENT
+                if is_valid_index(next_idx + 1) and LL[next_idx + 1].type in {
+                    token.DOUBLESTAR,
+                    token.LSQB,
+                    token.LPAR,
+                    token.DOT,
+                }:
                     continue
 
                 return Ok(string_idx)
@@ -3296,11 +3499,17 @@ class StringParenStripper(StringTransformer):
                 yield TErr(
                     "Will not strip parentheses which have comments attached to them."
                 )
+                return
 
         new_line = line.clone()
         new_line.comments = line.comments.copy()
-
-        append_leaves(new_line, line, LL[: string_idx - 1])
+        try:
+            append_leaves(new_line, line, LL[: string_idx - 1])
+        except BracketMatchError:
+            # HACK: I believe there is currently a bug somewhere in
+            # right_hand_split() that is causing brackets to not be tracked
+            # properly by a shared BracketTracker.
+            append_leaves(new_line, line, LL[: string_idx - 1], preformatted=True)
 
         string_leaf = Leaf(token.STRING, LL[string_idx].value)
         LL[string_idx - 1].remove()
@@ -3308,7 +3517,7 @@ class StringParenStripper(StringTransformer):
         new_line.append(string_leaf)
 
         append_leaves(
-            new_line, line, LL[string_idx + 1 : rpar_idx] + LL[rpar_idx + 1 :],
+            new_line, line, LL[string_idx + 1 : rpar_idx] + LL[rpar_idx + 1 :]
         )
 
         LL[rpar_idx].remove()
@@ -3463,9 +3672,12 @@ class BaseStringSplitter(StringTransformer):
                 # WMA4 a single space.
                 offset += 1
 
-                # WMA4 the lengths of any leaves that came before that space.
-                for leaf in LL[: p_idx + 1]:
+                # WMA4 the lengths of any leaves that came before that space,
+                # but after any closing bracket before that space.
+                for leaf in reversed(LL[: p_idx + 1]):
                     offset += len(str(leaf))
+                    if leaf.type in CLOSING_BRACKETS:
+                        break
 
         if is_valid_index(string_idx + 1):
             N = LL[string_idx + 1]
@@ -3531,7 +3743,8 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
         MIN_SUBSTR_SIZE characters.
 
         The string will ONLY be split on spaces (i.e. each new substring should
-        start with a space).
+        start with a space). Note that the string will NOT be split on a space
+        which is escaped with a backslash.
 
         If the string is an f-string, it will NOT be split in the middle of an
         f-expression (e.g. in f"FooBar: {foo() if x else bar()}", {foo() if x
@@ -3551,13 +3764,14 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
     MIN_SUBSTR_SIZE = 6
     # Matches an "f-expression" (e.g. {var}) that might be found in an f-string.
     RE_FEXPR = r"""
-    (?<!\{)\{
+    (?<!\{) (?:\{\{)* \{ (?!\{)
         (?:
             [^\{\}]
             | \{\{
             | \}\}
+            | (?R)
         )+?
-    (?<!\})(?:\}\})*\}(?!\})
+    (?<!\}) \} (?:\}\})* (?!\})
     """
 
     def do_splitter_match(self, line: Line) -> TMatchResult:
@@ -3611,7 +3825,7 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
 
         # We MAY choose to drop the 'f' prefix from substrings that don't
         # contain any f-expressions, but ONLY if the original f-string
-        # containes at least one f-expression. Otherwise, we will alter the AST
+        # contains at least one f-expression. Otherwise, we will alter the AST
         # of the program.
         drop_pointless_f_prefix = ("f" in prefix) and re.search(
             self.RE_FEXPR, LL[string_idx].value, re.VERBOSE
@@ -3701,7 +3915,7 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
                 max_bidx = max_break_idx - 2 if line_needs_plus() else max_break_idx
                 maybe_break_idx = self.__get_break_idx(rest_value, max_bidx)
                 if maybe_break_idx is None:
-                    # If we are unable to algorthmically determine a good split
+                    # If we are unable to algorithmically determine a good split
                     # and this string has custom splits registered to it, we
                     # fall back to using them--which means we have to start
                     # over from the beginning.
@@ -3871,11 +4085,23 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
                 section of this classes' docstring would be be met by returning @i.
             """
             is_space = string[i] == " "
+
+            is_not_escaped = True
+            j = i - 1
+            while is_valid_index(j) and string[j] == "\\":
+                is_not_escaped = not is_not_escaped
+                j -= 1
+
             is_big_enough = (
                 len(string[i:]) >= self.MIN_SUBSTR_SIZE
                 and len(string[:i]) >= self.MIN_SUBSTR_SIZE
             )
-            return is_space and is_big_enough and not breaks_fstring_expression(i)
+            return (
+                is_space
+                and is_not_escaped
+                and is_big_enough
+                and not breaks_fstring_expression(i)
+            )
 
         # First, we check all indices BELOW @max_break_idx.
         break_idx = max_break_idx
@@ -3981,12 +4207,13 @@ class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
     def do_splitter_match(self, line: Line) -> TMatchResult:
         LL = line.leaves
 
-        string_idx = None
-        string_idx = string_idx or self._return_match(LL)
-        string_idx = string_idx or self._else_match(LL)
-        string_idx = string_idx or self._assert_match(LL)
-        string_idx = string_idx or self._assign_match(LL)
-        string_idx = string_idx or self._dict_match(LL)
+        string_idx = (
+            self._return_match(LL)
+            or self._else_match(LL)
+            or self._assert_match(LL)
+            or self._assign_match(LL)
+            or self._dict_match(LL)
+        )
 
         if string_idx is not None:
             string_value = line.leaves[string_idx].value
@@ -4185,7 +4412,7 @@ class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
         is_valid_index = is_valid_index_factory(LL)
         insert_str_child = insert_str_child_factory(LL[string_idx])
 
-        comma_idx = len(LL) - 1
+        comma_idx = -1
         ends_with_comma = False
         if LL[comma_idx].type == token.COMMA:
             ends_with_comma = True
@@ -4230,9 +4457,11 @@ class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
         # `StringSplitter` will break it down further if necessary.
         string_value = LL[string_idx].value
         string_line = Line(
+            mode=line.mode,
             depth=line.depth + 1,
             inside_brackets=True,
-            should_explode=line.should_explode,
+            should_split_rhs=line.should_split_rhs,
+            magic_trailing_comma=line.magic_trailing_comma,
         )
         string_leaf = Leaf(token.STRING, string_value)
         insert_str_child(string_leaf)
@@ -4278,7 +4507,7 @@ class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
 class StringParser:
     """
     A state machine that aids in parsing a string's "trailer", which can be
-    either non-existant, an old-style formatting sequence (e.g. `% varX` or `%
+    either non-existent, an old-style formatting sequence (e.g. `% varX` or `%
     (varX, varY)`), or a method-call / attribute access (e.g. `.format(varX,
     varY)`).
 
@@ -4372,7 +4601,7 @@ class StringParser:
             was directly after the string leaf in question (e.g. if our target
             string is `line.leaves[i]` then the first call to this method must
             be `line.leaves[i + 1]`).
-            * On the next call to this function, the leaf paramater passed in
+            * On the next call to this function, the leaf parameter passed in
             MUST be the leaf directly following @leaf.
 
         Returns:
@@ -4570,25 +4799,25 @@ def line_to_string(line: Line) -> str:
     return str(line).strip("\n")
 
 
-def append_leaves(new_line: Line, old_line: Line, leaves: List[Leaf]) -> None:
+def append_leaves(
+    new_line: Line, old_line: Line, leaves: List[Leaf], preformatted: bool = False
+) -> None:
     """
     Append leaves (taken from @old_line) to @new_line, making sure to fix the
     underlying Node structure where appropriate.
 
     All of the leaves in @leaves are duplicated. The duplicates are then
     appended to @new_line and used to replace their originals in the underlying
-    Node structure. Any comments attatched to the old leaves are reattached to
+    Node structure. Any comments attached to the old leaves are reattached to
     the new leaves.
 
     Pre-conditions:
         set(@leaves) is a subset of set(@old_line.leaves).
     """
     for old_leaf in leaves:
-        assert old_leaf in old_line.leaves
-
         new_leaf = Leaf(old_leaf.type, old_leaf.value)
         replace_child(old_leaf, new_leaf)
-        new_line.append(new_leaf)
+        new_line.append(new_leaf, preformatted=preformatted)
 
         for comment_leaf in old_line.comments_after(old_leaf):
             new_line.append(comment_leaf, preformatted=True)
@@ -4745,8 +4974,7 @@ def right_hand_split(
     tail = bracket_split_build_line(tail_leaves, line, opening_bracket)
     bracket_split_succeeded_or_raise(head, body, tail)
     if (
-        # the body shouldn't be exploded
-        not body.should_explode
+        Feature.FORCE_OPTIONAL_PARENTHESES not in features
         # the opening bracket is an optional paren
         and opening_bracket.type == token.LPAR
         and not opening_bracket.value
@@ -4759,7 +4987,7 @@ def right_hand_split(
         # there are no standalone comments in the body
         and not body.contains_standalone_comments(0)
         # and we can actually remove the parens
-        and can_omit_invisible_parens(body, line_length)
+        and can_omit_invisible_parens(body, line_length, omit_on_explode=omit)
     ):
         omit = {id(closing_bracket), *omit}
         try:
@@ -4824,7 +5052,7 @@ def bracket_split_build_line(
     If `is_body` is True, the result line is one-indented inside brackets and as such
     has its first leaf's prefix normalized and a trailing comma added when expected.
     """
-    result = Line(depth=original.depth)
+    result = Line(mode=original.mode, depth=original.depth)
     if is_body:
         result.inside_brackets = True
         result.depth += 1
@@ -4836,7 +5064,7 @@ def bracket_split_build_line(
             no_commas = (
                 original.is_def
                 and opening_bracket.value == "("
-                and not any(l.type == token.COMMA for l in leaves)
+                and not any(leaf.type == token.COMMA for leaf in leaves)
             )
 
             if original.is_import or no_commas:
@@ -4845,7 +5073,8 @@ def bracket_split_build_line(
                         continue
 
                     if leaves[i].type != token.COMMA:
-                        leaves.insert(i + 1, Leaf(token.COMMA, ","))
+                        new_comma = Leaf(token.COMMA, ",")
+                        leaves.insert(i + 1, new_comma)
                     break
 
     # Populate the line
@@ -4853,8 +5082,8 @@ def bracket_split_build_line(
         result.append(leaf, preformatted=True)
         for comment_after in original.comments_after(leaf):
             result.append(comment_after, preformatted=True)
-    if is_body:
-        result.should_explode = should_explode(result, opening_bracket)
+    if is_body and should_split_line(result, opening_bracket):
+        result.should_split_rhs = True
     return result
 
 
@@ -4866,9 +5095,9 @@ def dont_increase_indentation(split_func: Transformer) -> Transformer:
 
     @wraps(split_func)
     def split_wrapper(line: Line, features: Collection[Feature] = ()) -> Iterator[Line]:
-        for l in split_func(line, features):
-            normalize_prefix(l.leaves[0], inside_brackets=True)
-            yield l
+        for line in split_func(line, features):
+            normalize_prefix(line.leaves[0], inside_brackets=True)
+            yield line
 
     return split_wrapper
 
@@ -4895,7 +5124,9 @@ def delimiter_split(line: Line, features: Collection[Feature] = ()) -> Iterator[
         if bt.delimiter_count_with_priority(delimiter_priority) == 1:
             raise CannotSplit("Splitting a single attribute from its owner looks wrong")
 
-    current_line = Line(depth=line.depth, inside_brackets=line.inside_brackets)
+    current_line = Line(
+        mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
+    )
     lowest_depth = sys.maxsize
     trailing_comma_safe = True
 
@@ -4907,7 +5138,9 @@ def delimiter_split(line: Line, features: Collection[Feature] = ()) -> Iterator[
         except ValueError:
             yield current_line
 
-            current_line = Line(depth=line.depth, inside_brackets=line.inside_brackets)
+            current_line = Line(
+                mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
+            )
             current_line.append(leaf)
 
     for leaf in line.leaves:
@@ -4931,7 +5164,9 @@ def delimiter_split(line: Line, features: Collection[Feature] = ()) -> Iterator[
         if leaf_priority == delimiter_priority:
             yield current_line
 
-            current_line = Line(depth=line.depth, inside_brackets=line.inside_brackets)
+            current_line = Line(
+                mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
+            )
     if current_line:
         if (
             trailing_comma_safe
@@ -4939,7 +5174,8 @@ def delimiter_split(line: Line, features: Collection[Feature] = ()) -> Iterator[
             and current_line.leaves[-1].type != token.COMMA
             and current_line.leaves[-1].type != STANDALONE_COMMENT
         ):
-            current_line.append(Leaf(token.COMMA, ","))
+            new_comma = Leaf(token.COMMA, ",")
+            current_line.append(new_comma)
         yield current_line
 
 
@@ -4951,7 +5187,9 @@ def standalone_comment_split(
     if not line.contains_standalone_comments(0):
         raise CannotSplit("Line does not have any standalone comments")
 
-    current_line = Line(depth=line.depth, inside_brackets=line.inside_brackets)
+    current_line = Line(
+        mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
+    )
 
     def append_to_line(leaf: Leaf) -> Iterator[Line]:
         """Append `leaf` to current line or to new line if appending impossible."""
@@ -4961,7 +5199,9 @@ def standalone_comment_split(
         except ValueError:
             yield current_line
 
-            current_line = Line(depth=line.depth, inside_brackets=line.inside_brackets)
+            current_line = Line(
+                line.mode, depth=line.depth, inside_brackets=line.inside_brackets
+            )
             current_line.append(leaf)
 
     for leaf in line.leaves:
@@ -5117,31 +5357,47 @@ def normalize_numeric_literal(leaf: Leaf) -> None:
         # Leave octal and binary literals alone.
         pass
     elif text.startswith("0x"):
-        # Change hex literals to upper case.
-        before, after = text[:2], text[2:]
-        text = f"{before}{after.upper()}"
+        text = format_hex(text)
     elif "e" in text:
-        before, after = text.split("e")
-        sign = ""
-        if after.startswith("-"):
-            after = after[1:]
-            sign = "-"
-        elif after.startswith("+"):
-            after = after[1:]
-        before = format_float_or_int_string(before)
-        text = f"{before}e{sign}{after}"
+        text = format_scientific_notation(text)
     elif text.endswith(("j", "l")):
-        number = text[:-1]
-        suffix = text[-1]
-        # Capitalize in "2L" because "l" looks too similar to "1".
-        if suffix == "l":
-            suffix = "L"
-        text = f"{format_float_or_int_string(number)}{suffix}"
+        text = format_long_or_complex_number(text)
     else:
         text = format_float_or_int_string(text)
     leaf.value = text
 
 
+def format_hex(text: str) -> str:
+    """
+    Formats a hexadecimal string like "0x12B3"
+    """
+    before, after = text[:2], text[2:]
+    return f"{before}{after.upper()}"
+
+
+def format_scientific_notation(text: str) -> str:
+    """Formats a numeric string utilizing scentific notation"""
+    before, after = text.split("e")
+    sign = ""
+    if after.startswith("-"):
+        after = after[1:]
+        sign = "-"
+    elif after.startswith("+"):
+        after = after[1:]
+    before = format_float_or_int_string(before)
+    return f"{before}e{sign}{after}"
+
+
+def format_long_or_complex_number(text: str) -> str:
+    """Formats a long or complex string like `10L` or `10j`"""
+    number = text[:-1]
+    suffix = text[-1]
+    # Capitalize in "2L" because "l" looks too similar to "1".
+    if suffix == "l":
+        suffix = "L"
+    return f"{format_float_or_int_string(number)}{suffix}"
+
+
 def format_float_or_int_string(text: str) -> str:
     """Formats a float string like "1.0"."""
     if "." not in text:
@@ -5180,9 +5436,6 @@ def normalize_invisible_parens(node: Node, parens_after: Set[str]) -> None:
             check_lpar = True
 
         if check_lpar:
-            if is_walrus_assignment(child):
-                continue
-
             if child.type == syms.atom:
                 if maybe_make_parens_invisible_in_atom(child, parent=node):
                     wrap_in_parentheses(node, child, visible=False)
@@ -5222,58 +5475,80 @@ def convert_one_fmt_off_pair(node: Node) -> bool:
     for leaf in node.leaves():
         previous_consumed = 0
         for comment in list_comments(leaf.prefix, is_endmarker=False):
-            if comment.value in FMT_OFF:
-                # We only want standalone comments. If there's no previous leaf or
-                # the previous leaf is indentation, it's a standalone comment in
-                # disguise.
-                if comment.type != STANDALONE_COMMENT:
-                    prev = preceding_leaf(leaf)
-                    if prev and prev.type not in WHITESPACE:
+            if comment.value not in FMT_PASS:
+                previous_consumed = comment.consumed
+                continue
+            # We only want standalone comments. If there's no previous leaf or
+            # the previous leaf is indentation, it's a standalone comment in
+            # disguise.
+            if comment.value in FMT_PASS and comment.type != STANDALONE_COMMENT:
+                prev = preceding_leaf(leaf)
+                if prev:
+                    if comment.value in FMT_OFF and prev.type not in WHITESPACE:
+                        continue
+                    if comment.value in FMT_SKIP and prev.type in WHITESPACE:
                         continue
 
-                ignored_nodes = list(generate_ignored_nodes(leaf))
-                if not ignored_nodes:
-                    continue
-
-                first = ignored_nodes[0]  # Can be a container node with the `leaf`.
-                parent = first.parent
-                prefix = first.prefix
-                first.prefix = prefix[comment.consumed :]
-                hidden_value = (
-                    comment.value + "\n" + "".join(str(n) for n in ignored_nodes)
-                )
-                if hidden_value.endswith("\n"):
-                    # That happens when one of the `ignored_nodes` ended with a NEWLINE
-                    # leaf (possibly followed by a DEDENT).
-                    hidden_value = hidden_value[:-1]
-                first_idx: Optional[int] = None
-                for ignored in ignored_nodes:
-                    index = ignored.remove()
-                    if first_idx is None:
-                        first_idx = index
-                assert parent is not None, "INTERNAL ERROR: fmt: on/off handling (1)"
-                assert first_idx is not None, "INTERNAL ERROR: fmt: on/off handling (2)"
-                parent.insert_child(
-                    first_idx,
-                    Leaf(
-                        STANDALONE_COMMENT,
-                        hidden_value,
-                        prefix=prefix[:previous_consumed] + "\n" * comment.newlines,
-                    ),
-                )
-                return True
+            ignored_nodes = list(generate_ignored_nodes(leaf, comment))
+            if not ignored_nodes:
+                continue
 
-            previous_consumed = comment.consumed
+            first = ignored_nodes[0]  # Can be a container node with the `leaf`.
+            parent = first.parent
+            prefix = first.prefix
+            first.prefix = prefix[comment.consumed :]
+            hidden_value = "".join(str(n) for n in ignored_nodes)
+            if comment.value in FMT_OFF:
+                hidden_value = comment.value + "\n" + hidden_value
+            if comment.value in FMT_SKIP:
+                hidden_value += "  " + comment.value
+            if hidden_value.endswith("\n"):
+                # That happens when one of the `ignored_nodes` ended with a NEWLINE
+                # leaf (possibly followed by a DEDENT).
+                hidden_value = hidden_value[:-1]
+            first_idx: Optional[int] = None
+            for ignored in ignored_nodes:
+                index = ignored.remove()
+                if first_idx is None:
+                    first_idx = index
+            assert parent is not None, "INTERNAL ERROR: fmt: on/off handling (1)"
+            assert first_idx is not None, "INTERNAL ERROR: fmt: on/off handling (2)"
+            parent.insert_child(
+                first_idx,
+                Leaf(
+                    STANDALONE_COMMENT,
+                    hidden_value,
+                    prefix=prefix[:previous_consumed] + "\n" * comment.newlines,
+                ),
+            )
+            return True
 
     return False
 
 
-def generate_ignored_nodes(leaf: Leaf) -> Iterator[LN]:
+def generate_ignored_nodes(leaf: Leaf, comment: ProtoComment) -> Iterator[LN]:
     """Starting from the container of `leaf`, generate all leaves until `# fmt: on`.
 
+    If comment is skip, returns leaf only.
     Stops at the end of the block.
     """
     container: Optional[LN] = container_of(leaf)
+    if comment.value in FMT_SKIP:
+        prev_sibling = leaf.prev_sibling
+        if comment.value in leaf.prefix and prev_sibling is not None:
+            leaf.prefix = leaf.prefix.replace(comment.value, "")
+            siblings = [prev_sibling]
+            while (
+                "\n" not in prev_sibling.prefix
+                and prev_sibling.prev_sibling is not None
+            ):
+                prev_sibling = prev_sibling.prev_sibling
+                siblings.insert(0, prev_sibling)
+            for sibling in siblings:
+                yield sibling
+        elif leaf.parent is not None:
+            yield leaf.parent
+        return
     while container is not None and container.type != token.ENDMARKER:
         if is_fmt_on(container):
             return
@@ -5333,6 +5608,7 @@ def maybe_make_parens_invisible_in_atom(node: LN, parent: LN) -> bool:
     Returns whether the node should itself be wrapped in invisible parentheses.
 
     """
+
     if (
         node.type != syms.atom
         or is_empty_tuple(node)
@@ -5342,6 +5618,15 @@ def maybe_make_parens_invisible_in_atom(node: LN, parent: LN) -> bool:
     ):
         return False
 
+    if is_walrus_assignment(node):
+        if parent.type in [
+            syms.annassign,
+            syms.expr_stmt,
+            syms.assert_stmt,
+            syms.return_stmt,
+        ]:
+            return False
+
     first = node.children[0]
     last = node.children[-1]
     if first.type == token.LPAR and last.type == token.RPAR:
@@ -5403,6 +5688,17 @@ def unwrap_singleton_parenthesis(node: LN) -> Optional[LN]:
     return wrapped
 
 
+def first_child_is_arith(node: Node) -> bool:
+    """Whether first child is an arithmetic or a binary arithmetic expression"""
+    expr_types = {
+        syms.arith_expr,
+        syms.shift_expr,
+        syms.xor_expr,
+        syms.and_expr,
+    }
+    return bool(node.children and node.children[0].type in expr_types)
+
+
 def wrap_in_parentheses(parent: Node, child: LN, *, visible: bool = True) -> None:
     """Wrap `child` in parentheses.
 
@@ -5443,6 +5739,49 @@ def is_walrus_assignment(node: LN) -> bool:
     return inner is not None and inner.type == syms.namedexpr_test
 
 
+def is_simple_decorator_trailer(node: LN, last: bool = False) -> bool:
+    """Return True iff `node` is a trailer valid in a simple decorator"""
+    return node.type == syms.trailer and (
+        (
+            len(node.children) == 2
+            and node.children[0].type == token.DOT
+            and node.children[1].type == token.NAME
+        )
+        # last trailer can be arguments
+        or (
+            last
+            and len(node.children) == 3
+            and node.children[0].type == token.LPAR
+            # and node.children[1].type == syms.argument
+            and node.children[2].type == token.RPAR
+        )
+    )
+
+
+def is_simple_decorator_expression(node: LN) -> bool:
+    """Return True iff `node` could be a 'dotted name' decorator
+
+    This function takes the node of the 'namedexpr_test' of the new decorator
+    grammar and test if it would be valid under the old decorator grammar.
+
+    The old grammar was: decorator: @ dotted_name [arguments] NEWLINE
+    The new grammar is : decorator: @ namedexpr_test NEWLINE
+    """
+    if node.type == token.NAME:
+        return True
+    if node.type == syms.power:
+        if node.children:
+            return (
+                node.children[0].type == token.NAME
+                and all(map(is_simple_decorator_trailer, node.children[1:-1]))
+                and (
+                    len(node.children) < 2
+                    or is_simple_decorator_trailer(node.children[-1], last=True)
+                )
+            )
+    return False
+
+
 def is_yield(node: LN) -> bool:
     """Return True if `node` holds a `yield` or `yield from` expression."""
     if node.type == syms.yield_expr:
@@ -5561,24 +5900,63 @@ def ensure_visible(leaf: Leaf) -> None:
         leaf.value = ")"
 
 
-def should_explode(line: Line, opening_bracket: Leaf) -> bool:
-    """Should `line` immediately be split with `delimiter_split()` after RHS?"""
+def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
+    """Should `line` be immediately split with `delimiter_split()` after RHS?"""
 
-    if not (
-        opening_bracket.parent
-        and opening_bracket.parent.type in {syms.atom, syms.import_from}
-        and opening_bracket.value in "[{("
-    ):
+    if not (opening_bracket.parent and opening_bracket.value in "[{("):
         return False
 
+    # We're essentially checking if the body is delimited by commas and there's more
+    # than one of them (we're excluding the trailing comma and if the delimiter priority
+    # is still commas, that means there's more).
+    exclude = set()
+    trailing_comma = False
     try:
         last_leaf = line.leaves[-1]
-        exclude = {id(last_leaf)} if last_leaf.type == token.COMMA else set()
+        if last_leaf.type == token.COMMA:
+            trailing_comma = True
+            exclude.add(id(last_leaf))
         max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
     except (IndexError, ValueError):
         return False
 
-    return max_priority == COMMA_PRIORITY
+    return max_priority == COMMA_PRIORITY and (
+        (line.mode.magic_trailing_comma and trailing_comma)
+        # always explode imports
+        or opening_bracket.parent.type in {syms.atom, syms.import_from}
+    )
+
+
+def is_one_tuple_between(opening: Leaf, closing: Leaf, leaves: List[Leaf]) -> bool:
+    """Return True if content between `opening` and `closing` looks like a one-tuple."""
+    if opening.type != token.LPAR and closing.type != token.RPAR:
+        return False
+
+    depth = closing.bracket_depth + 1
+    for _opening_index, leaf in enumerate(leaves):
+        if leaf is opening:
+            break
+
+    else:
+        raise LookupError("Opening paren not found in `leaves`")
+
+    commas = 0
+    _opening_index += 1
+    for leaf in leaves[_opening_index:]:
+        if leaf is closing:
+            break
+
+        bracket_depth = leaf.bracket_depth
+        if bracket_depth == depth and leaf.type == token.COMMA:
+            commas += 1
+            if leaf.parent and leaf.parent.type in {
+                syms.arglist,
+                syms.typedargslist,
+            }:
+                commas += 1
+                break
+
+    return commas < 2
 
 
 def get_features_used(node: Node) -> Set[Feature]:
@@ -5589,6 +5967,8 @@ def get_features_used(node: Node) -> Set[Feature]:
     - underscores in numeric literals;
     - trailing commas after * or ** in function signatures and calls;
     - positional only arguments in function signatures and lambdas;
+    - assignment expression;
+    - relaxed decorator syntax;
     """
     features: Set[Feature] = set()
     for n in node.pre_order():
@@ -5608,6 +5988,12 @@ def get_features_used(node: Node) -> Set[Feature]:
         elif n.type == token.COLONEQUAL:
             features.add(Feature.ASSIGNMENT_EXPRESSIONS)
 
+        elif n.type == syms.decorator:
+            if len(n.children) > 1 and not is_simple_decorator_expression(
+                n.children[1]
+            ):
+                features.add(Feature.RELAXED_DECORATORS)
+
         elif (
             n.type in {syms.typedargslist, syms.arglist}
             and n.children
@@ -5645,11 +6031,13 @@ def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[Leaf
     a preceding closing bracket fits in one line.
 
     Yielded sets are cumulative (contain results of previous yields, too).  First
-    set is empty.
+    set is empty, unless the line should explode, in which case bracket pairs until
+    the one that needs to explode are omitted.
     """
 
     omit: Set[LeafID] = set()
-    yield omit
+    if not line.magic_trailing_comma:
+        yield omit
 
     length = 4 * line.depth
     opening_bracket: Optional[Leaf] = None
@@ -5668,9 +6056,22 @@ def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[Leaf
             if leaf is opening_bracket:
                 opening_bracket = None
             elif leaf.type in CLOSING_BRACKETS:
+                prev = line.leaves[index - 1] if index > 0 else None
+                if (
+                    prev
+                    and prev.type == token.COMMA
+                    and not is_one_tuple_between(
+                        leaf.opening_bracket, leaf, line.leaves
+                    )
+                ):
+                    # Never omit bracket pairs with trailing commas.
+                    # We need to explode on those.
+                    break
+
                 inner_brackets.add(id(leaf))
         elif leaf.type in CLOSING_BRACKETS:
-            if index > 0 and line.leaves[index - 1].type in OPENING_BRACKETS:
+            prev = line.leaves[index - 1] if index > 0 else None
+            if prev and prev.type in OPENING_BRACKETS:
                 # Empty brackets would fail a split so treat them as "inner"
                 # brackets (e.g. only add them to the `omit` set if another
                 # pair of brackets was good enough.
@@ -5683,6 +6084,15 @@ def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[Leaf
                 inner_brackets.clear()
                 yield omit
 
+            if (
+                prev
+                and prev.type == token.COMMA
+                and not is_one_tuple_between(leaf.opening_bracket, leaf, line.leaves)
+            ):
+                # Never omit bracket pairs with trailing commas.
+                # We need to explode on those.
+                break
+
             if leaf.value:
                 opening_bracket = leaf.opening_bracket
                 closing_bracket = leaf
@@ -5740,7 +6150,7 @@ def get_future_imports(node: Node) -> Set[str]:
 
 @lru_cache()
 def get_gitignore(root: Path) -> PathSpec:
-    """ Return a PathSpec matching gitignore content if present."""
+    """Return a PathSpec matching gitignore content if present."""
     gitignore = root / ".gitignore"
     lines: List[str] = []
     if gitignore.is_file():
@@ -5749,16 +6159,51 @@ def get_gitignore(root: Path) -> PathSpec:
     return PathSpec.from_lines("gitwildmatch", lines)
 
 
+def normalize_path_maybe_ignore(
+    path: Path, root: Path, report: "Report"
+) -> Optional[str]:
+    """Normalize `path`. May return `None` if `path` was ignored.
+
+    `report` is where "path ignored" output goes.
+    """
+    try:
+        abspath = path if path.is_absolute() else Path.cwd() / path
+        normalized_path = abspath.resolve().relative_to(root).as_posix()
+    except OSError as e:
+        report.path_ignored(path, f"cannot be read because {e}")
+        return None
+
+    except ValueError:
+        if path.is_symlink():
+            report.path_ignored(path, f"is a symbolic link that points outside {root}")
+            return None
+
+        raise
+
+    return normalized_path
+
+
+def path_is_excluded(
+    normalized_path: str,
+    pattern: Optional[Pattern[str]],
+) -> bool:
+    match = pattern.search(normalized_path) if pattern else None
+    return bool(match and match.group(0))
+
+
 def gen_python_files(
     paths: Iterable[Path],
     root: Path,
     include: Optional[Pattern[str]],
-    exclude_regexes: Iterable[Pattern[str]],
+    exclude: Pattern[str],
+    extend_exclude: Optional[Pattern[str]],
+    force_exclude: Optional[Pattern[str]],
     report: "Report",
     gitignore: PathSpec,
 ) -> Iterator[Path]:
     """Generate all files under `path` whose paths are not excluded by the
-    `exclude` regex, but are included by the `include` regex.
+    `exclude_regex`, `extend_exclude`, or `force_exclude` regexes,
+    but are included by the `include` regex.
 
     Symbolic links pointing outside of the `root` directory are ignored.
 
@@ -5766,43 +6211,44 @@ def gen_python_files(
     """
     assert root.is_absolute(), f"INTERNAL ERROR: `root` must be absolute but is {root}"
     for child in paths:
-        # Then ignore with `exclude` option.
-        try:
-            normalized_path = child.resolve().relative_to(root).as_posix()
-        except OSError as e:
-            report.path_ignored(child, f"cannot be read because {e}")
+        normalized_path = normalize_path_maybe_ignore(child, root, report)
+        if normalized_path is None:
             continue
-        except ValueError:
-            if child.is_symlink():
-                report.path_ignored(
-                    child, f"is a symbolic link that points outside {root}"
-                )
-                continue
-
-            raise
 
         # First ignore files matching .gitignore
         if gitignore.match_file(normalized_path):
             report.path_ignored(child, "matches the .gitignore file content")
             continue
 
+        # Then ignore with `--exclude` `--extend-exclude` and `--force-exclude` options.
         normalized_path = "/" + normalized_path
         if child.is_dir():
             normalized_path += "/"
 
-        is_excluded = False
-        for exclude in exclude_regexes:
-            exclude_match = exclude.search(normalized_path) if exclude else None
-            if exclude_match and exclude_match.group(0):
-                report.path_ignored(child, "matches the --exclude regular expression")
-                is_excluded = True
-                break
-        if is_excluded:
+        if path_is_excluded(normalized_path, exclude):
+            report.path_ignored(child, "matches the --exclude regular expression")
+            continue
+
+        if path_is_excluded(normalized_path, extend_exclude):
+            report.path_ignored(
+                child, "matches the --extend-exclude regular expression"
+            )
+            continue
+
+        if path_is_excluded(normalized_path, force_exclude):
+            report.path_ignored(child, "matches the --force-exclude regular expression")
             continue
 
         if child.is_dir():
             yield from gen_python_files(
-                child.iterdir(), root, include, exclude_regexes, report, gitignore
+                child.iterdir(),
+                root,
+                include,
+                exclude,
+                extend_exclude,
+                force_exclude,
+                report,
+                gitignore,
             )
 
         elif child.is_file():
@@ -5812,11 +6258,11 @@ def gen_python_files(
 
 
 @lru_cache()
-def find_project_root(srcs: Iterable[str]) -> Path:
+def find_project_root(srcs: Tuple[str, ...]) -> Path:
     """Return a directory containing .git, .hg, or pyproject.toml.
 
-    That directory can be one of the directories passed in `srcs` or their
-    common parent.
+    That directory will be a common parent of all files and directories
+    passed in `srcs`.
 
     If no directory in the tree contains a marker that would specify it's the
     project root, the root of the file system is returned.
@@ -5824,11 +6270,20 @@ def find_project_root(srcs: Iterable[str]) -> Path:
     if not srcs:
         return Path("/").resolve()
 
-    common_base = min(Path(src).resolve() for src in srcs)
-    if common_base.is_dir():
-        # Append a fake file so `parents` below returns `common_base_dir`, too.
-        common_base /= "fake-file"
-    for directory in common_base.parents:
+    path_srcs = [Path(Path.cwd(), src).resolve() for src in srcs]
+
+    # A list of lists of parents for each 'src'. 'src' is included as a
+    # "parent" of itself if it is a directory
+    src_parents = [
+        list(path.parents) + ([path] if path.is_dir() else []) for path in path_srcs
+    ]
+
+    common_base = max(
+        set.intersection(*(set(parents) for parents in src_parents)),
+        key=lambda path: path.parts,
+    )
+
+    for directory in (common_base, *common_base.parents):
         if (directory / ".git").exists():
             return directory
 
@@ -5841,6 +6296,22 @@ def find_project_root(srcs: Iterable[str]) -> Path:
     return directory
 
 
+@lru_cache()
+def find_user_pyproject_toml() -> Path:
+    r"""Return the path to the top-level user configuration for black.
+
+    This looks for ~\.black on Windows and ~/.config/black on Linux and other
+    Unix systems.
+    """
+    if sys.platform == "win32":
+        # Windows
+        user_config_path = Path.home() / ".black"
+    else:
+        config_root = os.environ.get("XDG_CONFIG_HOME", "~/.config")
+        user_config_path = Path(config_root).expanduser() / "black"
+    return user_config_path.resolve()
+
+
 @dataclass
 class Report:
     """Provides a reformatting counter. Can be rendered with `str(report)`."""
@@ -5942,7 +6413,12 @@ def parse_ast(src: str) -> Union[ast.AST, ast3.AST, ast27.AST]:
                 return ast3.parse(src, filename, feature_version=feature_version)
             except SyntaxError:
                 continue
-
+    if ast27.__name__ == "ast":
+        raise SyntaxError(
+            "The requested source code has invalid Python 3 syntax.\n"
+            "If you are trying to format Python 2 files please reinstall Black"
+            " with the 'python2' extra: `python3 -m pip install black[python2]`."
+        )
     return ast27.parse(src)
 
 
@@ -6006,13 +6482,24 @@ def _stringify_ast(
 
         else:
             # Constant strings may be indented across newlines, if they are
-            # docstrings; fold spaces after newlines when comparing
+            # docstrings; fold spaces after newlines when comparing. Similarly,
+            # trailing and leading space may be removed.
+            # Note that when formatting Python 2 code, at least with Windows
+            # line-endings, docstrings can end up here as bytes instead of
+            # str so make sure that we handle both cases.
             if (
                 isinstance(node, ast.Constant)
                 and field == "value"
-                and isinstance(value, str)
+                and isinstance(value, (str, bytes))
             ):
-                normalized = re.sub(r"\n[ \t]+", "\n ", value)
+                lineend = "\n" if isinstance(value, str) else b"\n"
+                # To normalize, we strip any leading and trailing space from
+                # each line...
+                stripped = [line.strip() for line in value.splitlines()]
+                normalized = lineend.join(stripped)  # type: ignore[attr-defined]
+                # ...and remove any blank lines at the beginning and end of
+                # the whole string
+                normalized = normalized.strip()
             else:
                 normalized = value
             yield f"{'  ' * (depth+2)}{normalized!r},  # {value.__class__.__name__}"
@@ -6020,7 +6507,7 @@ def _stringify_ast(
     yield f"{'  ' * depth})  # /{node.__class__.__name__}"
 
 
-def assert_equivalent(src: str, dst: str) -> None:
+def assert_equivalent(src: str, dst: str, *, pass_num: int = 1) -> None:
     """Raise AssertionError if `src` and `dst` aren't equivalent."""
     try:
         src_ast = parse_ast(src)
@@ -6035,9 +6522,9 @@ def assert_equivalent(src: str, dst: str) -> None:
     except Exception as exc:
         log = dump_to_file("".join(traceback.format_tb(exc.__traceback__)), dst)
         raise AssertionError(
-            f"INTERNAL ERROR: Black produced invalid code: {exc}. Please report a bug"
-            " on https://github.com/psf/black/issues.  This invalid output might be"
-            f" helpful: {log}"
+            f"INTERNAL ERROR: Black produced invalid code on pass {pass_num}: {exc}. "
+            "Please report a bug on https://github.com/psf/black/issues.  "
+            f"This invalid output might be helpful: {log}"
         ) from None
 
     src_ast_str = "\n".join(_stringify_ast(src_ast))
@@ -6046,8 +6533,8 @@ def assert_equivalent(src: str, dst: str) -> None:
         log = dump_to_file(diff(src_ast_str, dst_ast_str, "src", "dst"))
         raise AssertionError(
             "INTERNAL ERROR: Black produced code that is not equivalent to the"
-            " source.  Please report a bug on https://github.com/psf/black/issues. "
-            f" This diff might be helpful: {log}"
+            f" source on pass {pass_num}.  Please report a bug on "
+            f"https://github.com/psf/black/issues.  This diff might be helpful: {log}"
         ) from None
 
 
@@ -6056,6 +6543,7 @@ def assert_stable(src: str, dst: str, mode: Mode) -> None:
     newdst = format_str(dst, mode=mode)
     if dst != newdst:
         log = dump_to_file(
+            str(mode),
             diff(src, dst, "source", "first pass"),
             diff(dst, newdst, "first pass", "second pass"),
         )
@@ -6067,14 +6555,14 @@ def assert_stable(src: str, dst: str, mode: Mode) -> None:
 
 
 @mypyc_attr(patchable=True)
-def dump_to_file(*output: str) -> str:
+def dump_to_file(*output: str, ensure_final_newline: bool = True) -> str:
     """Dump `output` to a temporary file. Return path to the file."""
     with tempfile.NamedTemporaryFile(
         mode="w", prefix="blk_", suffix=".log", delete=False, encoding="utf8"
     ) as f:
         for lines in output:
             f.write(lines)
-            if lines and lines[-1] != "\n":
+            if ensure_final_newline and lines and lines[-1] != "\n":
                 f.write("\n")
     return f.name
 
@@ -6092,11 +6580,20 @@ def diff(a: str, b: str, a_name: str, b_name: str) -> str:
     """Return a unified diff string between strings `a` and `b`."""
     import difflib
 
-    a_lines = [line + "\n" for line in a.splitlines()]
-    b_lines = [line + "\n" for line in b.splitlines()]
-    return "".join(
-        difflib.unified_diff(a_lines, b_lines, fromfile=a_name, tofile=b_name, n=5)
-    )
+    a_lines = [line for line in a.splitlines(keepends=True)]
+    b_lines = [line for line in b.splitlines(keepends=True)]
+    diff_lines = []
+    for line in difflib.unified_diff(
+        a_lines, b_lines, fromfile=a_name, tofile=b_name, n=5
+    ):
+        # Work around https://bugs.python.org/issue2142
+        # See https://www.gnu.org/software/diffutils/manual/html_node/Incomplete-Lines.html
+        if line[-1] == "\n":
+            diff_lines.append(line)
+        else:
+            diff_lines.append(line + "\n")
+            diff_lines.append("\\ No newline at end of file\n")
+    return "".join(diff_lines)
 
 
 def cancel(tasks: Iterable["asyncio.Task[Any]"]) -> None:
@@ -6232,7 +6729,11 @@ def can_be_split(line: Line) -> bool:
     return True
 
 
-def can_omit_invisible_parens(line: Line, line_length: int) -> bool:
+def can_omit_invisible_parens(
+    line: Line,
+    line_length: int,
+    omit_on_explode: Collection[LeafID] = (),
+) -> bool:
     """Does `line` have a shape safe to reformat without optional parens around it?
 
     Returns True for only a subset of potentially nice looking formattings but
@@ -6255,37 +6756,27 @@ def can_omit_invisible_parens(line: Line, line_length: int) -> bool:
 
     assert len(line.leaves) >= 2, "Stranded delimiter"
 
-    first = line.leaves[0]
-    second = line.leaves[1]
-    penultimate = line.leaves[-2]
-    last = line.leaves[-1]
-
     # With a single delimiter, omit if the expression starts or ends with
     # a bracket.
+    first = line.leaves[0]
+    second = line.leaves[1]
     if first.type in OPENING_BRACKETS and second.type not in CLOSING_BRACKETS:
-        remainder = False
-        length = 4 * line.depth
-        for _index, leaf, leaf_length in enumerate_with_length(line):
-            if leaf.type in CLOSING_BRACKETS and leaf.opening_bracket is first:
-                remainder = True
-            if remainder:
-                length += leaf_length
-                if length > line_length:
-                    break
-
-                if leaf.type in OPENING_BRACKETS:
-                    # There are brackets we can further split on.
-                    remainder = False
-
-        else:
-            # checked the entire string and line length wasn't exceeded
-            if len(line.leaves) == _index + 1:
-                return True
+        if _can_omit_opening_paren(line, first=first, line_length=line_length):
+            return True
 
         # Note: we are not returning False here because a line might have *both*
         # a leading opening bracket and a trailing closing bracket.  If the
         # opening bracket doesn't match our rule, maybe the closing will.
 
+    penultimate = line.leaves[-2]
+    last = line.leaves[-1]
+    if line.magic_trailing_comma:
+        try:
+            penultimate, last = last_two_except(line.leaves, omit=omit_on_explode)
+        except LookupError:
+            # Turns out we'd omit everything.  We cannot skip the optional parentheses.
+            return False
+
     if (
         last.type == token.RPAR
         or last.type == token.RBRACE
@@ -6306,21 +6797,120 @@ def can_omit_invisible_parens(line: Line, line_length: int) -> bool:
             # unnecessary.
             return True
 
-        length = 4 * line.depth
-        seen_other_brackets = False
-        for _index, leaf, leaf_length in enumerate_with_length(line):
+        if line.magic_trailing_comma and penultimate.type == token.COMMA:
+            # The rightmost non-omitted bracket pair is the one we want to explode on.
+            return True
+
+        if _can_omit_closing_paren(line, last=last, line_length=line_length):
+            return True
+
+    return False
+
+
+def _can_omit_opening_paren(line: Line, *, first: Leaf, line_length: int) -> bool:
+    """See `can_omit_invisible_parens`."""
+    remainder = False
+    length = 4 * line.depth
+    _index = -1
+    for _index, leaf, leaf_length in enumerate_with_length(line):
+        if leaf.type in CLOSING_BRACKETS and leaf.opening_bracket is first:
+            remainder = True
+        if remainder:
             length += leaf_length
-            if leaf is last.opening_bracket:
-                if seen_other_brackets or length <= line_length:
-                    return True
+            if length > line_length:
+                break
 
-            elif leaf.type in OPENING_BRACKETS:
+            if leaf.type in OPENING_BRACKETS:
                 # There are brackets we can further split on.
-                seen_other_brackets = True
+                remainder = False
+
+    else:
+        # checked the entire string and line length wasn't exceeded
+        if len(line.leaves) == _index + 1:
+            return True
+
+    return False
+
+
+def _can_omit_closing_paren(line: Line, *, last: Leaf, line_length: int) -> bool:
+    """See `can_omit_invisible_parens`."""
+    length = 4 * line.depth
+    seen_other_brackets = False
+    for _index, leaf, leaf_length in enumerate_with_length(line):
+        length += leaf_length
+        if leaf is last.opening_bracket:
+            if seen_other_brackets or length <= line_length:
+                return True
+
+        elif leaf.type in OPENING_BRACKETS:
+            # There are brackets we can further split on.
+            seen_other_brackets = True
 
     return False
 
 
+def last_two_except(leaves: List[Leaf], omit: Collection[LeafID]) -> Tuple[Leaf, Leaf]:
+    """Return (penultimate, last) leaves skipping brackets in `omit` and contents."""
+    stop_after = None
+    last = None
+    for leaf in reversed(leaves):
+        if stop_after:
+            if leaf is stop_after:
+                stop_after = None
+            continue
+
+        if last:
+            return leaf, last
+
+        if id(leaf) in omit:
+            stop_after = leaf.opening_bracket
+        else:
+            last = leaf
+    else:
+        raise LookupError("Last two leaves were also skipped")
+
+
+def run_transformer(
+    line: Line,
+    transform: Transformer,
+    mode: Mode,
+    features: Collection[Feature],
+    *,
+    line_str: str = "",
+) -> List[Line]:
+    if not line_str:
+        line_str = line_to_string(line)
+    result: List[Line] = []
+    for transformed_line in transform(line, features):
+        if str(transformed_line).strip("\n") == line_str:
+            raise CannotTransform("Line transformer returned an unchanged result")
+
+        result.extend(transform_line(transformed_line, mode=mode, features=features))
+
+    if not (
+        transform.__name__ == "rhs"
+        and line.bracket_tracker.invisible
+        and not any(bracket.value for bracket in line.bracket_tracker.invisible)
+        and not line.contains_multiline_strings()
+        and not result[0].contains_uncollapsable_type_comments()
+        and not result[0].contains_unsplittable_type_ignore()
+        and not is_line_short_enough(result[0], line_length=mode.line_length)
+    ):
+        return result
+
+    line_copy = line.clone()
+    append_leaves(line_copy, line, line.leaves)
+    features_fop = set(features) | {Feature.FORCE_OPTIONAL_PARENTHESES}
+    second_opinion = run_transformer(
+        line_copy, transform, mode, features_fop, line_str=line_str
+    )
+    if all(
+        is_line_short_enough(ln, line_length=mode.line_length) for ln in second_opinion
+    ):
+        result = second_opinion
+    return result
+
+
 def get_cache_file(mode: Mode) -> Path:
     return CACHE_DIR / f"cache.{mode.get_cache_key()}.pickle"
 
@@ -6357,8 +6947,8 @@ def filter_cached(cache: Cache, sources: Iterable[Path]) -> Tuple[Set[Path], Set
     """
     todo, done = set(), set()
     for src in sources:
-        src = src.resolve()
-        if cache.get(src) != get_cache_info(src):
+        res_src = src.resolve()
+        if cache.get(str(res_src)) != get_cache_info(res_src):
             todo.add(src)
         else:
             done.add(src)
@@ -6370,7 +6960,10 @@ def write_cache(cache: Cache, sources: Iterable[Path], mode: Mode) -> None:
     cache_file = get_cache_file(mode)
     try:
         CACHE_DIR.mkdir(parents=True, exist_ok=True)
-        new_cache = {**cache, **{src.resolve(): get_cache_info(src) for src in sources}}
+        new_cache = {
+            **cache,
+            **{str(src.resolve()): get_cache_info(src) for src in sources},
+        }
         with tempfile.NamedTemporaryFile(dir=str(cache_file.parent), delete=False) as f:
             pickle.dump(new_cache, f, protocol=4)
         os.replace(f.name, cache_file)
@@ -6406,13 +6999,48 @@ def patched_main() -> None:
     main()
 
 
+def is_docstring(leaf: Leaf) -> bool:
+    if prev_siblings_are(
+        leaf.parent, [None, token.NEWLINE, token.INDENT, syms.simple_stmt]
+    ):
+        return True
+
+    # Multiline docstring on the same line as the `def`.
+    if prev_siblings_are(leaf.parent, [syms.parameters, token.COLON, syms.simple_stmt]):
+        # `syms.parameters` is only used in funcdefs and async_funcdefs in the Python
+        # grammar. We're safe to return True without further checks.
+        return True
+
+    return False
+
+
+def lines_with_leading_tabs_expanded(s: str) -> List[str]:
+    """
+    Splits string into lines and expands only leading tabs (following the normal
+    Python rules)
+    """
+    lines = []
+    for line in s.splitlines():
+        # Find the index of the first non-whitespace character after a string of
+        # whitespace that includes at least one tab
+        match = re.match(r"\s*\t+\s*(\S)", line)
+        if match:
+            first_non_whitespace_idx = match.start(1)
+
+            lines.append(
+                line[:first_non_whitespace_idx].expandtabs()
+                + line[first_non_whitespace_idx:]
+            )
+        else:
+            lines.append(line)
+    return lines
+
+
 def fix_docstring(docstring: str, prefix: str) -> str:
     # https://www.python.org/dev/peps/pep-0257/#handling-docstring-indentation
     if not docstring:
         return ""
-    # Convert tabs to spaces (following the normal Python rules)
-    # and split into a list of lines:
-    lines = docstring.expandtabs().splitlines()
+    lines = lines_with_leading_tabs_expanded(docstring)
     # Determine minimum indentation (first line doesn't count):
     indent = sys.maxsize
     for line in lines[1:]:
@@ -6429,7 +7057,6 @@ def fix_docstring(docstring: str, prefix: str) -> str:
                 trimmed.append(prefix + stripped_line)
             else:
                 trimmed.append("")
-    # Return a single string:
     return "\n".join(trimmed)