""" jsonlines implementation """ import builtins import codecs import enum import io import json import os import types import typing from typing import ( Any, Callable, Dict, Iterable, Iterator, List, Literal, Optional, Tuple, Type, TypeVar, Union, cast, overload, ) import attr orjson: Optional[types.ModuleType] try: import orjson except ImportError: orjson = None ujson: Optional[types.ModuleType] try: import ujson except ImportError: ujson = None VALID_TYPES = { bool, dict, float, int, list, str, } # Characters to skip at the beginning of a line. Note: at most one such # character is skipped per line. SKIPPABLE_SINGLE_INITIAL_CHARS = ( "\x1e", # RFC7464 text sequence codecs.BOM_UTF8.decode(), ) class DumpsResultConversion(enum.Enum): LeaveAsIs = enum.auto() EncodeToBytes = enum.auto() DecodeToString = enum.auto() # https://docs.python.org/3/library/functions.html#open Openable = Union[str, bytes, int, os.PathLike] LoadsCallable = Callable[[Union[str, bytes]], Any] DumpsCallable = Callable[[Any], Union[str, bytes]] # Currently, JSON structures cannot be typed properly: # - https://github.com/python/typing/issues/182 # - https://github.com/python/mypy/issues/731 JSONCollection = Union[Dict[str, Any], List[Any]] JSONScalar = Union[bool, float, int, str] JSONValue = Union[JSONCollection, JSONScalar] TJSONValue = TypeVar("TJSONValue", bound=JSONValue) TRW = TypeVar("TRW", bound="ReaderWriterBase") # Default to using the fastest JSON library for reading, falling back to the # standard library (always available) if none are installed. if orjson is not None: default_loads = orjson.loads elif ujson is not None: default_loads = ujson.loads else: default_loads = json.loads # For writing, use the stdlib. Other packages may be faster but their behaviour # (supported types etc.) and output (whitespace etc.) are not the same as the # stdlib json module, so this should be opt-in via the ‘dumps=’ arg. def default_dumps(obj: Any) -> str: """ Fake ``dumps()`` function to use as a default marker. """ raise NotImplementedError # pragma: no cover @attr.s(auto_exc=True, auto_attribs=True) class Error(Exception): """ Base error class. """ message: str @attr.s(auto_exc=True, auto_attribs=True, init=False) class InvalidLineError(Error, ValueError): """ Error raised when an invalid line is encountered. This happens when the line does not contain valid JSON, or if a specific data type has been requested, and the line contained a different data type. The original line itself is stored on the exception instance as the ``.line`` attribute, and the line number as ``.lineno``. This class subclasses both ``jsonlines.Error`` and the built-in ``ValueError``. """ #: The invalid line line: Union[str, bytes] #: The line number lineno: int def __init__(self, message: str, line: Union[str, bytes], lineno: int) -> None: self.line = line.rstrip() self.lineno = lineno super().__init__(f"{message} (line {lineno})") @attr.s(auto_attribs=True, repr=False) class ReaderWriterBase: """ Base class with shared behaviour for both the reader and writer. """ _fp: Union[typing.IO[str], typing.IO[bytes], None] = attr.ib( default=None, init=False ) _closed: bool = attr.ib(default=False, init=False) _should_close_fp: bool = attr.ib(default=False, init=False) def close(self) -> None: """ Close this reader/writer. This closes the underlying file if that file has been opened by this reader/writer. When an already opened file-like object was provided, the caller is responsible for closing it. """ if self._closed: return self._closed = True if self._fp is not None and self._should_close_fp: self._fp.close() def __repr__(self) -> str: cls_name = type(self).__name__ wrapped = self._repr_for_wrapped() return f"" def _repr_for_wrapped(self) -> str: raise NotImplementedError # pragma: no cover def __enter__(self: TRW) -> TRW: return self def __exit__( self, exc_type: Optional[Type[BaseException]], exc_val: Optional[BaseException], exc_tb: Optional[types.TracebackType], ) -> None: self.close() @attr.s(auto_attribs=True, repr=False) class Reader(ReaderWriterBase): """ Reader for the jsonlines format. The first argument must be an iterable that yields JSON encoded strings. Usually this will be a readable file-like object, such as an open file or an ``io.TextIO`` instance, but it can also be something else as long as it yields strings when iterated over. Instances are iterable and can be used as a context manager. The `loads` argument can be used to replace the standard json decoder. If specified, it must be a callable that accepts a (unicode) string and returns the decoded object. :param file_or_iterable: file-like object or iterable yielding lines as strings :param loads: custom json decoder callable """ _file_or_iterable: Union[ typing.IO[str], typing.IO[bytes], Iterable[Union[str, bytes]] ] _line_iter: Iterator[Tuple[int, Union[bytes, str]]] = attr.ib(init=False) _loads: LoadsCallable = attr.ib(default=default_loads, kw_only=True) def __attrs_post_init__(self) -> None: if isinstance(self._file_or_iterable, io.IOBase): self._fp = cast( Union[typing.IO[str], typing.IO[bytes]], self._file_or_iterable, ) self._line_iter = enumerate(self._file_or_iterable, 1) # No type specified, None not allowed @overload def read( self, *, type: Literal[None] = ..., allow_none: Literal[False] = ..., skip_empty: bool = ..., ) -> JSONValue: ... # pragma: no cover # No type specified, None allowed @overload def read( self, *, type: Literal[None] = ..., allow_none: Literal[True], skip_empty: bool = ..., ) -> Optional[JSONValue]: ... # pragma: no cover # Type specified, None not allowed @overload def read( self, *, type: Type[TJSONValue], allow_none: Literal[False] = ..., skip_empty: bool = ..., ) -> TJSONValue: ... # pragma: no cover # Type specified, None allowed @overload def read( self, *, type: Type[TJSONValue], allow_none: Literal[True], skip_empty: bool = ..., ) -> Optional[TJSONValue]: ... # pragma: no cover # Generic definition @overload def read( self, *, type: Optional[Type[Any]] = ..., allow_none: bool = ..., skip_empty: bool = ..., ) -> Optional[JSONValue]: ... # pragma: no cover def read( self, *, type: Optional[Type[Any]] = None, allow_none: bool = False, skip_empty: bool = False, ) -> Optional[JSONValue]: """ Read and decode a line. The optional `type` argument specifies the expected data type. Supported types are ``dict``, ``list``, ``str``, ``int``, ``float``, and ``bool``. When specified, non-conforming lines result in :py:exc:`InvalidLineError`. By default, input lines containing ``null`` (in JSON) are considered invalid, and will cause :py:exc:`InvalidLineError`. The `allow_none` argument can be used to change this behaviour, in which case ``None`` will be returned instead. If `skip_empty` is set to ``True``, empty lines and lines containing only whitespace are silently skipped. """ if self._closed: raise RuntimeError("reader is closed") if type is not None and type not in VALID_TYPES: raise ValueError("invalid type specified") try: lineno, line = next(self._line_iter) while skip_empty and not line.rstrip(): lineno, line = next(self._line_iter) except StopIteration: raise EOFError from None if isinstance(line, bytes): try: line = line.decode("utf-8") except UnicodeDecodeError as orig_exc: exc = InvalidLineError( f"line is not valid utf-8: {orig_exc}", line, lineno ) raise exc from orig_exc if line.startswith(SKIPPABLE_SINGLE_INITIAL_CHARS): line = line[1:] try: value: JSONValue = self._loads(line) except ValueError as orig_exc: exc = InvalidLineError( f"line contains invalid json: {orig_exc}", line, lineno ) raise exc from orig_exc if value is None: if allow_none: return None raise InvalidLineError("line contains null value", line, lineno) if type is not None: valid = isinstance(value, type) if type is int and isinstance(value, bool): # isinstance() is not sufficient, since bool is an int subclass valid = False if not valid: raise InvalidLineError( "line does not match requested type", line, lineno ) return value # No type specified, None not allowed @overload def iter( self, *, type: Literal[None] = ..., allow_none: Literal[False] = ..., skip_empty: bool = ..., skip_invalid: bool = ..., ) -> Iterator[JSONValue]: ... # pragma: no cover # No type specified, None allowed @overload def iter( self, *, type: Literal[None] = ..., allow_none: Literal[True], skip_empty: bool = ..., skip_invalid: bool = ..., ) -> Iterator[JSONValue]: ... # pragma: no cover # Type specified, None not allowed @overload def iter( self, *, type: Type[TJSONValue], allow_none: Literal[False] = ..., skip_empty: bool = ..., skip_invalid: bool = ..., ) -> Iterator[TJSONValue]: ... # pragma: no cover # Type specified, None allowed @overload def iter( self, *, type: Type[TJSONValue], allow_none: Literal[True], skip_empty: bool = ..., skip_invalid: bool = ..., ) -> Iterator[Optional[TJSONValue]]: ... # pragma: no cover # Generic definition @overload def iter( self, *, type: Optional[Type[TJSONValue]] = ..., allow_none: bool = ..., skip_empty: bool = ..., skip_invalid: bool = ..., ) -> Iterator[Optional[TJSONValue]]: ... # pragma: no cover def iter( self, type: Optional[Type[Any]] = None, allow_none: bool = False, skip_empty: bool = False, skip_invalid: bool = False, ) -> Iterator[Optional[JSONValue]]: """ Iterate over all lines. This is the iterator equivalent to repeatedly calling :py:meth:`~Reader.read()`. If no arguments are specified, this is the same as directly iterating over this :py:class:`Reader` instance. When `skip_invalid` is set to ``True``, invalid lines will be silently ignored. See :py:meth:`~Reader.read()` for a description of the other arguments. """ try: while True: try: yield self.read( type=type, allow_none=allow_none, skip_empty=skip_empty ) except InvalidLineError: if not skip_invalid: raise except EOFError: pass def __iter__(self) -> Iterator[Any]: """ See :py:meth:`~Reader.iter()`. """ return self.iter() def _repr_for_wrapped(self) -> str: if self._fp is not None: return repr_for_fp(self._fp) class_name = type(self._file_or_iterable).__name__ return f"<{class_name} at 0x{id(self._file_or_iterable):x}>" @attr.s(auto_attribs=True, repr=False) class Writer(ReaderWriterBase): """ Writer for the jsonlines format. Instances can be used as a context manager. The `fp` argument must be a file-like object with a ``.write()`` method accepting either text (unicode) or bytes. The `compact` argument can be used to to produce smaller output. The `sort_keys` argument can be used to sort keys in json objects, and will produce deterministic output. For more control, provide a a custom encoder callable using the `dumps` argument. The callable must produce (unicode) string output. If specified, the `compact` and `sort` arguments will be ignored. When the `flush` argument is set to ``True``, the writer will call ``fp.flush()`` after each written line. :param fp: writable file-like object :param compact: whether to use a compact output format :param sort_keys: whether to sort object keys :param dumps: custom encoder callable :param flush: whether to flush the file-like object after writing each line """ _fp: Union[typing.IO[str], typing.IO[bytes]] = attr.ib(default=None) _fp_is_binary: bool = attr.ib(default=False, init=False) _compact: bool = attr.ib(default=False, kw_only=True) _sort_keys: bool = attr.ib(default=False, kw_only=True) _flush: bool = attr.ib(default=False, kw_only=True) _dumps: DumpsCallable = attr.ib(default=default_dumps, kw_only=True) _dumps_result_conversion: DumpsResultConversion = attr.ib( default=DumpsResultConversion.LeaveAsIs, init=False ) def __attrs_post_init__(self) -> None: if isinstance(self._fp, io.TextIOBase): self._fp_is_binary = False elif isinstance(self._fp, io.IOBase): self._fp_is_binary = True else: try: self._fp.write("") # type: ignore[call-overload] except TypeError: self._fp_is_binary = True else: self._fp_is_binary = False if self._dumps is default_dumps: self._dumps = json.JSONEncoder( ensure_ascii=False, separators=(",", ":") if self._compact else (", ", ": "), sort_keys=self._sort_keys, ).encode # Detect if str-to-bytes conversion (or vice versa) is needed for the # combination of this file-like object and the used dumps() callable. # This avoids checking this for each .write(). Note that this # deliberately does not support ‘dynamic’ return types that depend on # input and dump options, like simplejson on Python 2 in some cases. sample_dumps_result = self._dumps({}) if isinstance(sample_dumps_result, str) and self._fp_is_binary: self._dumps_result_conversion = DumpsResultConversion.EncodeToBytes elif isinstance(sample_dumps_result, bytes) and not self._fp_is_binary: self._dumps_result_conversion = DumpsResultConversion.DecodeToString def write(self, obj: Any) -> int: """ Encode and write a single object. :param obj: the object to encode and write :return: number of characters or bytes written """ if self._closed: raise RuntimeError("writer is closed") line = self._dumps(obj) # This handles either str or bytes, but the type checker does not know # that this code always passes the right type of arguments. if self._dumps_result_conversion == DumpsResultConversion.EncodeToBytes: line = line.encode() # type: ignore[union-attr] elif self._dumps_result_conversion == DumpsResultConversion.DecodeToString: line = line.decode() # type: ignore[union-attr] fp = self._fp fp.write(line) # type: ignore[arg-type] fp.write(b"\n" if self._fp_is_binary else "\n") # type: ignore[call-overload] if self._flush: fp.flush() return len(line) + 1 # including newline def write_all(self, iterable: Iterable[Any]) -> int: """ Encode and write multiple objects. :param iterable: an iterable of objects :return: number of characters or bytes written """ return sum(self.write(obj) for obj in iterable) def _repr_for_wrapped(self) -> str: return repr_for_fp(self._fp) @overload def open( file: Openable, mode: Literal["r"] = ..., *, loads: Optional[LoadsCallable] = ..., ) -> Reader: ... # pragma: no cover @overload def open( file: Openable, mode: Literal["w", "a", "x"], *, dumps: Optional[DumpsCallable] = ..., compact: Optional[bool] = ..., sort_keys: Optional[bool] = ..., flush: Optional[bool] = ..., ) -> Writer: ... # pragma: no cover @overload def open( file: Openable, mode: str = ..., *, loads: Optional[LoadsCallable] = ..., dumps: Optional[DumpsCallable] = ..., compact: Optional[bool] = ..., sort_keys: Optional[bool] = ..., flush: Optional[bool] = ..., ) -> Union[Reader, Writer]: ... # pragma: no cover def open( file: Openable, mode: str = "r", *, loads: Optional[LoadsCallable] = None, dumps: Optional[DumpsCallable] = None, compact: Optional[bool] = None, sort_keys: Optional[bool] = None, flush: Optional[bool] = None, ) -> Union[Reader, Writer]: """ Open a jsonlines file for reading or writing. This is a convenience function to open a file and wrap it in either a :py:class:`Reader` or :py:class:`Writer` instance, depending on the specified `mode`. Additional keyword arguments will be passed on to the reader and writer; see their documentation for available options. The resulting reader or writer must be closed after use by the caller, which will also close the opened file. This can be done by calling ``.close()``, but the easiest way to ensure proper resource finalisation is to use a ``with`` block (context manager), e.g. :: with jsonlines.open('out.jsonl', mode='w') as writer: writer.write(...) :param file: name or ‘path-like object’ of the file to open :param mode: whether to open the file for reading (``r``), writing (``w``), appending (``a``), or exclusive creation (``x``). """ if mode not in {"r", "w", "a", "x"}: raise ValueError("'mode' must be either 'r', 'w', 'a', or 'x'") cls = Reader if mode == "r" else Writer encoding = "utf-8-sig" if mode == "r" else "utf-8" fp = builtins.open(file, mode=mode + "t", encoding=encoding) kwargs = dict( loads=loads, dumps=dumps, compact=compact, sort_keys=sort_keys, flush=flush, ) kwargs = {key: value for key, value in kwargs.items() if value is not None} instance: Union[Reader, Writer] = cls(fp, **kwargs) instance._should_close_fp = True return instance def repr_for_fp(fp: typing.IO[Any]) -> str: """ Helper to make a useful repr() for a file-like object. """ name = getattr(fp, "name", None) if name is not None: return repr(name) else: return repr(fp)