"""Color stripping, delta coalescing, and stdout/stderr redirection.
The redirection machinery (`WrappingIO`, `StreamWrapper`, and the
module-level `streams` singleton constructed at the bottom of this
module) is what lets `print()` calls and logging output appear as
normal lines above a redrawing progress bar instead of corrupting its
line. See the "Print while a bar is running" how-to for the full
picture.
"""
from __future__ import annotations
import atexit
import contextlib
import datetime
import io
import logging
import os
import re
import sys
import typing
from collections.abc import Callable, Iterable, Iterator, Mapping
from types import TracebackType
from python_utils import types
from python_utils.converters import scale_1024
from python_utils.terminal import get_terminal_size
from python_utils.time import epoch, format_time, timedelta_to_seconds
from progressbar import base, env, terminal
# Make sure these are available for import
assert timedelta_to_seconds is not None
assert get_terminal_size is not None
assert format_time is not None
assert scale_1024 is not None
assert epoch is not None
StringT = typing.TypeVar('StringT', bound=types.StringTypes)
T = typing.TypeVar('T')
logger: logging.Logger = logging.getLogger(__name__)
class _ProgressListener(typing.Protocol):
"""Structural type for the bars the stream wrapper notifies.
Defined locally instead of importing ``ProgressBarMixinBase`` from
``bar`` so ``utils`` has no dependency on ``bar``, not even a
type-checking-only one, which CodeQL still reports as a ``bar`` <->
``utils`` module-level import cycle.
"""
def update(self) -> None:
"""Redraw in response to redirected output being written."""
# Precompiled ANSI CSI escape-sequence patterns (str and bytes). Compiled once
# at import instead of per no_color() call, which runs for every widget on
# every redraw.
_ANSI_COLOR_RE: re.Pattern[str] = re.compile('\x1b\\[.*?[@-~]')
_ANSI_COLOR_RE_BYTES: re.Pattern[bytes] = re.compile(
bytes(terminal.ESC, 'ascii') + b'\\[.*?[@-~]',
)
@typing.overload
def deltas_to_seconds(
*deltas: datetime.timedelta | float | int | None,
default: type[ValueError] = ...,
) -> float:
"""Coalesce to seconds, raising ``ValueError`` if no delta is valid."""
@typing.overload
def deltas_to_seconds(
*deltas: datetime.timedelta | float | int | None,
default: T,
) -> float | T:
"""Coalesce to seconds, returning ``default`` if no delta is valid."""
[docs]
def deltas_to_seconds(
*deltas: datetime.timedelta | float | int | None,
default: typing.Any = ValueError,
) -> typing.Any:
"""Coalesce timedeltas and second counts to a single seconds float.
Returns the first argument in `deltas` that isn't `None`, converted
to seconds as a `float`. Raises (or returns `default`, if given) only
when every argument is `None`.
>>> deltas_to_seconds(datetime.timedelta(seconds=1, milliseconds=234))
1.234
>>> deltas_to_seconds(123)
123.0
>>> deltas_to_seconds(1.234)
1.234
>>> deltas_to_seconds(None, 1.234)
1.234
>>> deltas_to_seconds(0, 1.234)
0.0
>>> deltas_to_seconds()
Traceback (most recent call last):
...
ValueError: No valid deltas passed to `deltas_to_seconds`
>>> deltas_to_seconds(None)
Traceback (most recent call last):
...
ValueError: No valid deltas passed to `deltas_to_seconds`
>>> deltas_to_seconds(default=0.0)
0.0
"""
for delta in deltas:
if delta is None:
continue
if isinstance(delta, datetime.timedelta):
return timedelta_to_seconds(delta)
elif not isinstance(delta, float):
return float(delta)
else:
return delta
if default is ValueError:
raise ValueError('No valid deltas passed to `deltas_to_seconds`')
else:
return default
[docs]
def no_color(value: StringT) -> StringT:
"""Return the `value` without ANSI escape codes.
>>> no_color(b'\u001b[1234]abc')
b'abc'
>>> str(no_color('\u001b[1234]abc'))
'abc'
>>> str(no_color('\u001b[1234]abc'))
'abc'
>>> no_color(123)
Traceback (most recent call last):
...
TypeError: `value` must be a string or bytes, got 123
"""
if isinstance(value, bytes):
# Fast path: with no ESC byte there is nothing to strip, so the regex
# would return the value unchanged anyway. Skipping it avoids a
# substitution on the common plain-text case, which dominates the
# per-redraw render cost (len_color is called for every widget).
if b'\x1b' not in value:
return value # type: ignore
return _ANSI_COLOR_RE_BYTES.sub(b'', value) # type: ignore
elif isinstance(value, str):
if '\x1b' not in value:
return value # type: ignore
return _ANSI_COLOR_RE.sub('', value) # type: ignore
else:
raise TypeError(f'`value` must be a string or bytes, got {value!r}')
[docs]
def len_color(value: types.StringTypes) -> int:
"""Return the length of `value` without ANSI escape codes.
>>> len_color(b'\u001b[1234]abc')
3
>>> len_color('\u001b[1234]abc')
3
>>> len_color('\u001b[1234]abc')
3
"""
return len(no_color(value))
[docs]
class WrappingIO:
"""`sys.stdout`/`sys.stderr` replacement installed while capturing.
Buffers writes in memory instead of passing them straight through
while `capturing` is on, so a bar can erase its own line, flush the
buffer above it, and redraw. See `StreamWrapper.wrap_stdout`/
`wrap_stderr` for how one gets installed.
"""
buffer: io.StringIO
target: base.IO
capturing: bool
listeners: set[_ProgressListener]
needs_clear: bool = False
def __init__(
self,
target: base.IO,
capturing: bool = False,
listeners: set[_ProgressListener] | None = None,
) -> None:
"""Wrap `target` so writes can be buffered while `capturing`.
Args:
target: The real stream writes are buffered for, and
eventually flushed through to.
capturing: Start in buffering mode immediately instead of
passing writes straight through.
listeners: Bars to notify (`update()`) whenever a buffered
write completes a line. Typically shared with
`StreamWrapper.listeners` by the caller, so every
wrapped stream notifies the same bars.
"""
self.buffer = io.StringIO()
self.target = target
self.capturing = capturing
self.listeners = listeners or set()
self.needs_clear = False
[docs]
def write(self, value: str) -> int:
"""Write `value`, buffering it in memory while `capturing`.
While `capturing` is on, `value` is appended to `buffer` instead
of reaching `target`. If the buffered text now contains a
newline, `needs_clear` is set and every listener's `update()` is
called, so a capturing bar redraws promptly instead of waiting
for its next scheduled update. While not `capturing`, `value` is
written straight through to `target` and `target` is flushed on
every newline, so unbuffered output still appears live.
Returns:
The number of characters written (buffered or passed
through), mirroring the return value of a normal file
`write()`.
"""
ret = 0
if self.capturing:
ret += self.buffer.write(value)
if '\n' in value: # pragma: no branch
self.needs_clear = True
for listener in self.listeners: # pragma: no branch
listener.update()
else:
ret += self.target.write(value)
if '\n' in value: # pragma: no branch
self.flush_target()
return ret
[docs]
def flush(self) -> None:
"""Flush the in-memory buffer, a no-op for `io.StringIO`.
Only satisfies the file-like `flush()` protocol. It does not
write buffered text through to `target`. Use `_flush()` (or
`StreamWrapper.flush()`, which calls it) for that.
"""
self.buffer.flush()
def _flush(self) -> None:
"""Write buffered output through to `target`, then flush it.
The buffer is drained (`seek`/`truncate`) *before*
`target.write()` is called, not after, so if that write
raises, the already-buffered text isn't written a second time
by the next call. `target` is flushed unconditionally at the
end, even when the buffer was empty, since this runs on every
bar redraw, not just when there's something to flush.
"""
if value := self.buffer.getvalue():
self.flush()
# Clear the buffer before writing so a failed write cannot
# cause the same data to be written again by the next flush
self.buffer.seek(0)
self.buffer.truncate(0)
self.needs_clear = False
if not self.target.closed:
self.target.write(value)
# when explicitly flushing, always flush the target as well
self.flush_target()
[docs]
def flush_target(self) -> None: # pragma: no cover
"""Flush `target` itself, if it's open and flushable."""
if not self.target.closed and getattr(self.target, 'flush', None):
self.target.flush()
def __enter__(self) -> WrappingIO:
"""Return `self` for use as a context manager."""
return self
[docs]
def fileno(self) -> int:
"""Return `target`'s file descriptor."""
return self.target.fileno()
[docs]
def isatty(self) -> bool:
"""Return whether `target` is a tty."""
return self.target.isatty()
[docs]
def read(self, n: int = -1) -> str:
"""Read up to `n` characters from `target`."""
return self.target.read(n)
[docs]
def readable(self) -> bool:
"""Return whether `target` supports reading."""
return self.target.readable()
[docs]
def readline(self, limit: int = -1) -> str:
"""Read a single line (up to `limit` characters) from `target`."""
return self.target.readline(limit)
[docs]
def readlines(self, hint: int = -1) -> list[str]:
"""Read all lines from `target`."""
return self.target.readlines(hint)
[docs]
def seek(self, offset: int, whence: int = os.SEEK_SET) -> int:
"""Seek `target` to `offset`, relative to `whence`."""
return self.target.seek(offset, whence)
[docs]
def seekable(self) -> bool:
"""Return whether `target` supports seeking."""
return self.target.seekable()
[docs]
def tell(self) -> int:
"""Return `target`'s current stream position."""
return self.target.tell()
[docs]
def truncate(self, size: int | None = None) -> int:
"""Truncate `target` to `size`."""
return self.target.truncate(size)
[docs]
def writable(self) -> bool:
"""Return whether `target` supports writing."""
return self.target.writable()
[docs]
def writelines(self, lines: Iterable[str]) -> None:
"""Write `lines` to `target`."""
return self.target.writelines(lines)
[docs]
def close(self) -> None:
"""Flush the buffer and close `target`."""
self.flush()
self.target.close()
def __next__(self) -> str:
"""Return the next line read from `target`."""
return self.target.__next__()
def __iter__(self) -> Iterator[str]:
"""Return an iterator over `target`'s lines."""
return self.target.__iter__()
def __exit__(
self,
__t: type[BaseException] | None,
__value: BaseException | None,
__traceback: TracebackType | None,
) -> None:
"""Close on context-manager exit, regardless of `__t`."""
self.close()
[docs]
class StreamWrapper:
"""Wrap `sys.stdout`/`sys.stderr` for output, logging, and a bar to share.
Almost always used via the module-level `streams` singleton
(constructed once at the bottom of this module) rather than
instantiated directly. Each `wrap_*`/`unwrap_*` pair is refcounted,
so nested or concurrent bars that both request redirection share one
wrapper and it's only undone once the last one finishes.
"""
stdout: base.TextIO | WrappingIO
stderr: base.TextIO | WrappingIO
original_excepthook: Callable[
[
type[BaseException],
BaseException,
TracebackType | None,
],
None,
]
wrapped_stdout: int = 0
wrapped_stderr: int = 0
wrapped_logging: int = 0
wrapped_excepthook: int = 0
logging_handlers: list[tuple[logging.StreamHandler[base.IO], base.IO]]
capturing: int = 0
listeners: set[_ProgressListener]
def __init__(self) -> None:
"""Capture the *current* `sys.stdout`/`sys.stderr` as "real".
Note:
This runs once, at construction, and `streams` (below) is
constructed at import time. Anything that reassigns
`sys.stdout`/`sys.stderr` after `progressbar.utils` is first
imported will not be picked up: `original_stdout`/
`original_stderr` keep pointing at whatever was installed at
that moment, not whatever is live later. This has been a
repeat source of bugs.
"""
self.stdout = self.original_stdout = sys.stdout
self.stderr = self.original_stderr = sys.stderr
self.original_excepthook = sys.excepthook
self.wrapped_stdout = 0
self.wrapped_stderr = 0
self.wrapped_logging = 0
self.wrapped_excepthook = 0
self.logging_handlers = []
self.capturing = 0
self.listeners = set()
if env.env_flag('WRAP_STDOUT', default=False): # pragma: no cover
self.wrap_stdout()
if env.env_flag('WRAP_STDERR', default=False): # pragma: no cover
self.wrap_stderr()
[docs]
def start_capturing(self, bar: _ProgressListener | None = None) -> None:
"""Turn capturing on for `bar` and bump the shared refcount.
Args:
bar: Registered as a listener so it's notified (`update()`)
when captured output completes a line. Omit to just
bump the refcount without listening.
"""
if bar: # pragma: no branch
self.listeners.add(bar)
self.capturing += 1
self.update_capturing()
[docs]
def stop_capturing(self, bar: _ProgressListener | None = None) -> None:
"""Unregister `bar` and drop the shared capturing refcount.
Args:
bar: The listener to remove, if it was registered via
`start_capturing`.
"""
if bar: # pragma: no branch
with contextlib.suppress(KeyError):
self.listeners.remove(bar)
self.capturing -= 1
self.update_capturing()
[docs]
def update_capturing(self) -> None: # pragma: no cover
"""Propagate the capturing refcount to the wrapped streams.
Flushes immediately once the refcount drops to zero or below,
so whatever's left in the buffer reaches the terminal as soon
as the last bar stops capturing, rather than sitting there
until something else happens to trigger a flush.
"""
if isinstance(self.stdout, WrappingIO):
self.stdout.capturing = self.capturing > 0
if isinstance(self.stderr, WrappingIO):
self.stderr.capturing = self.capturing > 0
if self.capturing <= 0:
self.flush()
[docs]
def wrap(self, stdout: bool = False, stderr: bool = False) -> None:
"""Wrap `stdout` and/or `stderr`, per the given flags."""
if stdout:
self.wrap_stdout()
if stderr:
self.wrap_stderr()
[docs]
def wrap_stdout(self) -> WrappingIO:
"""Install a `WrappingIO` over `sys.stdout`, or share it.
Refcounted: only the first call actually replaces `sys.stdout`;
later calls just bump `wrapped_stdout` so nested/concurrent bars
share one wrapper, and it takes a matching number of
`unwrap_stdout()` calls to restore the original stream. Also
wraps `sys.excepthook`, since a traceback printed while
capturing would otherwise bypass the buffer.
Returns:
The installed `WrappingIO`, the same instance on every
call until it's fully unwrapped.
"""
self.wrap_excepthook()
if not self.wrapped_stdout:
self.stdout = sys.stdout = WrappingIO( # type: ignore
self.original_stdout,
listeners=self.listeners,
)
self.wrapped_stdout += 1
return sys.stdout # type: ignore
[docs]
def wrap_stderr(self) -> WrappingIO:
"""Install a `WrappingIO` over `sys.stderr`, or share it.
See `wrap_stdout` for the refcounting behavior, mirrored here
for `sys.stderr`.
Returns:
The installed `WrappingIO`, the same instance on every
call until it's fully unwrapped.
"""
self.wrap_excepthook()
if not self.wrapped_stderr:
self.stderr = sys.stderr = WrappingIO( # type: ignore
self.original_stderr,
listeners=self.listeners,
)
self.wrapped_stderr += 1
return sys.stderr # type: ignore
[docs]
def wrap_logging(self) -> None:
"""Retarget every `StreamHandler` in the logger tree to the wrapper.
Refcounted like `wrap_stdout`/`wrap_stderr`: only the first call
actually walks the logger tree and rewrites handlers, so
nested/concurrent redirection doesn't fight over the same
handlers or lose track of what to restore. Each handler is
visited once (deduplicated by `id()`, via `_wrap_logging_handler`)
because the same handler object can be attached to more than one
logger in the tree, and each retargeted handler is recorded in
`logging_handlers` so `unwrap_logging` can put it back later.
"""
self.wrapped_logging += 1
if self.wrapped_logging > 1:
return
wrapped_streams = {
self.original_stdout: self.stdout,
self.original_stderr: self.stderr,
sys.stdout: self.stdout,
sys.stderr: self.stderr,
}
restore_streams: dict[object, base.IO] = {}
if isinstance(self.stdout, WrappingIO):
restore_streams[self.stdout] = self.original_stdout
if isinstance(self.stderr, WrappingIO):
restore_streams[self.stderr] = self.original_stderr
seen: set[int] = set()
for logger_ in self._iter_loggers():
for handler in tuple(logger_.handlers):
if id(handler) in seen:
continue
seen.add(id(handler))
if not isinstance(handler, logging.StreamHandler):
continue
self._wrap_logging_handler(
handler,
wrapped_streams,
restore_streams,
)
def _wrap_logging_handler(
self,
handler: logging.StreamHandler[base.IO],
wrapped_streams: Mapping[types.Any, types.Any],
restore_streams: Mapping[types.Any, base.IO],
) -> None:
"""Retarget one handler's stream, or note it for restoration.
`wrapped_streams` maps the real/current stdout and stderr
objects to their wrapper. If `handler.stream` is one of those,
it's repointed at the wrapper (`_set_handler_stream`) and the
stream it used to point at is saved in `logging_handlers` so
`unwrap_logging` can restore it.
Otherwise, if `handler.stream` is itself already a wrapper (per
`restore_streams`, e.g. `wrap_stdout`/`wrap_stderr` ran before
`wrap_logging`, so the handler already points at the wrapper),
it's left untouched but still recorded for restoration, so
`unwrap_logging` still puts it back once this session ends.
"""
stream = handler.stream
replacement = wrapped_streams.get(stream)
if replacement is not None and replacement is not stream:
if self._set_handler_stream(handler, replacement):
self.logging_handlers.append((handler, stream))
elif (restore_stream := restore_streams.get(stream)) is not None:
self.logging_handlers.append((handler, restore_stream))
[docs]
def unwrap_logging(self) -> None:
"""Undo one `wrap_logging()` call, restoring at refcount zero.
Only the call that brings `wrapped_logging` to zero actually
restores anything: it pops every entry `wrap_logging` recorded
in `logging_handlers` and puts each handler's original stream
back.
"""
if self.wrapped_logging > 1:
self.wrapped_logging -= 1
return
if not self.wrapped_logging:
return
while self.logging_handlers:
handler, stream = self.logging_handlers.pop()
self._set_handler_stream(handler, stream)
self.wrapped_logging = 0
def _set_handler_stream(
self,
handler: logging.StreamHandler[base.IO],
stream: types.Any,
) -> bool:
"""Point `handler` at `stream`, tolerating a closed old stream.
`StreamHandler.setStream()` flushes the handler's *current*
stream as part of switching, which raises `ValueError` if that
stream is already closed. `AttributeError` guards against
`handler` not actually exposing `setStream` (defensive, given
the type hint promises it does). Either way, the handler is left
untouched rather than raising.
Returns:
Whether `handler.stream` was actually changed.
"""
with contextlib.suppress(AttributeError, ValueError):
handler.setStream(stream)
return True
return False
def _iter_loggers(self) -> types.Iterator[logging.Logger]:
"""Yield the root logger, then every other registered logger.
Snapshots `logging.Logger.manager.loggerDict` with `tuple()`
first, since new loggers can be created while a caller is still
draining this generator. Entries that are placeholders for a
not-yet-created parent logger (not real `Logger` instances) are
skipped.
"""
yield logging.getLogger()
for logger_ in tuple(logging.Logger.manager.loggerDict.values()):
if isinstance(logger_, logging.Logger):
yield logger_
[docs]
def unwrap_excepthook(self) -> None:
"""Restore the original `sys.excepthook`, if currently wrapped."""
if self.wrapped_excepthook:
self.wrapped_excepthook -= 1
sys.excepthook = self.original_excepthook
[docs]
def wrap_excepthook(self) -> None:
"""Install the shared excepthook that flushes buffered output.
A no-op if already wrapped: `wrap_stdout()` and `wrap_stderr()`
both call this unconditionally, and either one may already have
wrapped it.
"""
if not self.wrapped_excepthook:
logger.debug('wrapping excepthook')
self.wrapped_excepthook += 1
sys.excepthook = self.excepthook
[docs]
def unwrap(self, stdout: bool = False, stderr: bool = False) -> None:
"""Unwrap `stdout` and/or `stderr`, per the given flags."""
if stdout:
self.unwrap_stdout()
if stderr:
self.unwrap_stderr()
[docs]
def unwrap_stdout(self) -> None:
"""Undo one `wrap_stdout()` call, restoring at refcount zero.
Only the call that brings `wrapped_stdout` to zero actually
restores `sys.stdout`, and `self.stdout` alongside it, so
`needs_clear()` and `update_capturing()` don't keep reading a
wrapper that's no longer installed. Also unwraps the shared
excepthook once `stderr` is back to its original too, since
it's shared between the two.
"""
if self.wrapped_stdout > 1:
self.wrapped_stdout -= 1
else:
# Also reset our own reference so needs_clear() and
# update_capturing() don't act on a stale wrapper
self.stdout = sys.stdout = self.original_stdout
self.wrapped_stdout = 0
if not self.wrapped_stderr:
self.unwrap_excepthook()
[docs]
def unwrap_stderr(self) -> None:
"""Undo one `wrap_stderr()` call, restoring at refcount zero.
Mirrors `unwrap_stdout()`: only the call that brings
`wrapped_stderr` to zero restores `sys.stderr`, and the shared
excepthook is unwrapped once `stdout` is back to its original
too.
"""
if self.wrapped_stderr > 1:
self.wrapped_stderr -= 1
else:
# Also reset our own reference so needs_clear() and
# update_capturing() don't act on a stale wrapper
self.stderr = sys.stderr = self.original_stderr
self.wrapped_stderr = 0
if not self.wrapped_stdout:
self.unwrap_excepthook()
[docs]
def needs_clear(self) -> bool: # pragma: no cover
"""Return whether either wrapped stream has buffered output.
Uses `getattr` with a `False` default so this is safe to call
whether or not `stdout`/`stderr` are currently wrapped, since a
plain, unwrapped stream has no `needs_clear` attribute.
Returns:
Whether a bar's next redraw should erase its line first, so
buffered `print()`/logging output can be flushed above it.
"""
stdout_needs_clear = getattr(self.stdout, 'needs_clear', False)
stderr_needs_clear = getattr(self.stderr, 'needs_clear', False)
return stderr_needs_clear or stdout_needs_clear
[docs]
def flush(self) -> None:
"""Flush buffered captured output on both wrapped streams.
If writing the buffered text to a stream's target raises
`io.UnsupportedOperation` (as happens for some non-seekable
streams), that stream's redirection disables itself:
`wrapped_stdout`/`wrapped_stderr` is reset to 0 so this method
stops attempting to flush it on future calls, and a warning is
logged. `sys.stdout`/`sys.stderr` are left installed as-is:
only further flush attempts are skipped, not the wrapping
itself.
"""
if self.wrapped_stdout and isinstance(self.stdout, WrappingIO):
try:
self.stdout._flush()
except io.UnsupportedOperation:
self.wrapped_stdout = 0
logger.warning(
'Disabling stdout redirection, %r is not seekable',
sys.stdout,
)
if self.wrapped_stderr and isinstance(self.stderr, WrappingIO):
try:
self.stderr._flush()
except io.UnsupportedOperation: # pragma: no cover
self.wrapped_stderr = 0
logger.warning(
'Disabling stderr redirection, %r is not seekable',
sys.stderr,
)
[docs]
def excepthook(
self,
exc_type: type[BaseException],
exc_value: BaseException,
exc_traceback: TracebackType | None,
) -> None:
"""Run the original excepthook, then flush buffered output.
Installed as `sys.excepthook` while stdout or stderr is wrapped
(see `wrap_excepthook`), so an uncaught exception's traceback,
written via the original hook, is followed by whatever output
was still buffered, instead of that text getting lost or
appearing in the wrong order relative to the traceback.
"""
self.original_excepthook(exc_type, exc_value, exc_traceback)
self.flush()
[docs]
class AttributeDict(dict[str, T], typing.Generic[T]):
"""A dict that can be accessed with .attribute.
Note:
Double-underscore names (e.g. ``__orig_class__``, set by
``typing.Generic`` on subscripted instances) are routed to real
instance attributes instead of dict entries, keeping runtime
metadata like that out of the mapping's contents. See
``__setattr__``/``__delattr__``.
>>> attrs = AttributeDict(spam=123)
# Reading
>>> attrs['spam']
123
>>> attrs.spam
123
# Read after update using attribute
>>> attrs.spam = 456
>>> attrs['spam']
456
>>> attrs.spam
456
# Read after update using dict access
>>> attrs['spam'] = 123
>>> attrs['spam']
123
>>> attrs.spam
123
# Read after update using dict access
>>> del attrs.spam
>>> attrs['spam']
Traceback (most recent call last):
...
KeyError: 'spam'
>>> attrs.spam
Traceback (most recent call last):
...
AttributeError: No such attribute: spam
>>> del attrs.spam
Traceback (most recent call last):
...
AttributeError: No such attribute: spam
"""
def __getattr__(self, name: str) -> T:
"""Return `self[name]`, so dict keys are readable as attributes.
Raises:
AttributeError: `name` is not a key in this dict.
"""
if name in self:
return self[name]
else:
raise AttributeError(f'No such attribute: {name}')
def __setattr__(self, name: str, value: T) -> None:
"""Store `name`/`value` as a dict entry, unless `name` is a dunder.
Dunder names are set as a real instance attribute instead (via
`object.__setattr__`), per the class `Note:` above.
"""
if name.startswith('__') and name.endswith('__'):
object.__setattr__(self, name, value)
return
self[name] = value
def __delattr__(self, name: str) -> None:
"""Delete the `name` dict entry, unless `name` is a dunder.
Mirrors `__setattr__`: dunder names are deleted via
`object.__delattr__` instead of as dict entries.
Raises:
AttributeError: `name` is neither a dunder nor an existing
key.
"""
if name.startswith('__') and name.endswith('__'):
object.__delattr__(self, name)
return
if name in self:
del self[name]
else:
raise AttributeError(f'No such attribute: {name}')
#: Process-global, constructed once at import. The only place that knows
#: the *real* ``sys.stdout``/``sys.stderr`` versus whatever is currently
#: installed in their place. Every bar that redirects goes through this
#: one shared instance rather than each keeping its own. Mutating it
#: (wrapping/unwrapping) affects every bar and every plain ``print()``
#: in the process. ``bar.py``'s ``StdRedirectMixin`` is its only real
#: consumer.
streams = StreamWrapper()
atexit.register(streams.flush)