big is a Python package of small functions and classes that aren't big enough to get a package of their own. It's zillions of useful little bits of Python code I always want to have handy.
For years, I've copied-and-pasted all my little helper functions between projects--we've all done it. But now I've finally taken the time to consolidate all those useful little functions into one big package--no more copy-and-paste, I just install one package and I'm ready to go. And, since it's a public package, you can use 'em too!
Not only that, but I've taken my time and re-thought and retooled a lot of this code. All the difficult-to-use, overspecialized, cheap hacks I've lived with for years have been upgraded with elegant, intuitive APIs and dazzling functionality. big is chock full of the sort of little functions and classes we've all hacked together a million times--only with all the API gotchas fixed, and thoroughly tested with 100% coverage. It's the code you would have written... if only you had the time. It's a real pleasure to use!
big requires Python 3.6 or newer. It has no required dependencies (unless you want to run the test suite). The current version is 0.12.7.
Think big!
It's true that much of the code in big is short, and one might reasonably have the reaction "that's so short, it's easier to write it from scratch every time I need it than remember where it is and how to call it". I still see value in these short functions in big because:
For example, consider
StateManager.
If you remove the comments and documentation, it's actually pretty
short--easily less than a hundred lines. I myself have written
state machines from scratch using a similar approach many times.
They're easy to write. So why bother using StateManager?
Why not roll your own each time?
Because StateManager not only supports all the features you need--consider
accessor
and
dispatch--its
API is carefully designed to help prevent bugs and logical errors.
In considering the predecessor of StateManager for inclusion in big,
I realized that if an "observer" initiated a state transition, it would produce
a blurry mess of observer callbacks and entered and exited states,
executed in a confusing order. So StateManager in big simply
prevents you from executing state transitions in observers.
To use big, just install the big package (and its dependencies) from PyPI using your favorite Python package manager.
Once big is installed, you can simply import it. However, the top-level big package doesn't contain anything but a version number. Internally big is broken up into submodules, aggregated together loosely by problem domain, and you can selectively import just the functions you want. For example, if you only want to use the text functions, just import the text submodule:
import big.text
If you'd prefer to import everything all at once, simply import the big.all module. This one module imports all the other modules, and imports all their symbols too. So, one convenient way to work with big is this:
import big.all as big
That will make every symbol defined in big accessible from the big
object. For example, if you want to use
multisplit,
you can access it with just big.multisplit.
You can also use big.all with import *:
from big.all import *
but that's up to you.
big is licensed using the MIT license. You're free to use it and even ship it in your own programs, as long as you leave my copyright notice on the source code.
Although big is crammed full of fabulous code, a few of its subsystems rise above the rest. If you're curious what big might do for you, here are the five things in big I'm proudest of:
multi- family of string functionsbig.stateTopologicalSorterlines and lines modifier functionsAnd here are five little functions/classes I use all the time:
accessor(attribute='state', state_manager='state_manager')
combine_splits(s, *split_arrays)
datetime_ensure_timezone(d, timezone)
datetime_set_timezone(d, timezone)
decode_python_script(script, *, newline=None, use_bom=True, use_source_code_encoding=True)
Delimiter(close, *, escape='', multiline=True, quoting=False)
dispatch(state_manager='state_manager', *, prefix='', suffix='')
encode_strings(o, *, encoding='ascii')
Event(scheduler, event, time, priority, sequence)
fgrep(path, text, *, encoding=None, enumerate=False, case_insensitive=False)
gently_title(s, *, apostrophes=None, double_quotes=None)
get_float(o, default=_sentinel)
get_int_or_float(o, default=_sentinel)
grep(path, pattern, *, encoding=None, enumerate=False, flags=0)
int_to_words(i, *, flowery=True, ordinal=False)
LineInfo.clip_leading(line, s)
LineInfo.clip_trailing(line, s)
lines_convert_tabs_to_spaces(li)
lines_filter_line_comment_lines(li, comment_markers)
lines_containing(li, s, *, invert=False)
lines_grep(li, pattern, *, invert=False, flags=0, match='match')
lines_rstrip(li, separators=None)
lines_sort(li, *, key=None, reverse=False)
lines_strip(li, separators=None)
lines_strip_line_comments(li, line_comment_markers, *, quotes=(), escape='\\', multiline_quotes=())
multipartition(s, separators, count=1, *, reverse=False, separate=True)
multisplit(s, separators, *, keep=False, maxsplit=-1, reverse=False, separate=False, strip=False)
multistrip(s, separators, left=True, right=True)
normalize_whitespace(s, separators=None, replacement=None)
parse_timestamp_3339Z(s, *, timezone=None)
PushbackIterator(iterable=None)
PushbackIterator.next(default=None)
read_python_file(path, *, newline=None, use_bom=True, use_source_code_encoding=True)
re_partition(text, pattern, count=1, *, flags=0, reverse=False)
re_rpartition(text, pattern, count=1, *, flags=0)
reversed_re_finditer(pattern, string, flags=0)
Scheduler(regulator=default_regulator)
Scheduler.schedule(o, time, *, absolute=False, priority=DEFAULT_PRIORITY)
split_delimiters(s, delimiters={...}, *, state=(), yields=None)
split_quoted_strings(s, quotes=('"', "'"), *, escape='\\', multiline_quotes=(), state='')
split_text_with_code(s, *, tab_width=8, allow_code=True, code_indent=4, convert_tabs_to_spaces=True)
split_title_case(s, *, split_allcaps=True)
StateManager(state, *, on_enter='on_enter', on_exit='on_exit', state_class=None)
timestamp_3339Z(t=None, want_microseconds=None)
timestamp_human(t=None, want_microseconds=None)
TopologicalSorter.remove(node)
TopologicalSorter.View.close()
TopologicalSorter.View.done(*nodes)
TopologicalSorter.View.print(print=print)
TopologicalSorter.View.ready()
TopologicalSorter.View.reset()
translate_filename_to_exfat(s)
unicode_linebreaks_without_crlf
The multi- family of string functions
Whitespace and line-breaking characters in Python and big
big.allThis submodule doesn't define any of its own symbols. Instead, it
imports every other submodule in big, and uses import * to
import every symbol from every other submodule, too. Every
public symbol in big is available in big.all.
big.boundinnerclassClass decorators that implement bound inner classes. See the Bound inner classes deep-dive for more information.
BoundInnerClass(cls)Class decorator for an inner class. When accessing the inner class
through an instance of the outer class, "binds" the inner class to
the instance. This changes the signature of the inner class's __init__
from
def __init__(self, *args, **kwargs):`
to
def __init__(self, outer, *args, **kwargs):
where outer is the instance of the outer class.
Compare this to functions:
__init__,
after self.Note that this has an implication for all subclasses.
If class B is decorated with BoundInnerClass,
and class S is a subclass of B, such that
issubclass(S, B),
class S must be decorated with either
BoundInnerClass or UnboundInnerClass.
UnboundInnerClass(cls)Class decorator for an inner class that prevents binding the inner class to an instance of the outer class.
If class B is decorated with BoundInnerClass,
and class S is a subclass of B, such that
issubclass(S, B) returns True,
class S must be decorated with either
BoundInnerClass or UnboundInnerClass.
big.builtinFunctions for working with builtins. (Named builtin to avoid a
name collision with the builtins module.)
In general, the idea with these functions is a principle I first read about in either Code Complete or Writing Solid Code:
Don't associate with losers.
The intent here is, try to design APIs where it's impossible to call them the wrong way. Restrict the inputs to your functions to values you can always handle, and you won't ever have to return an error.
The functions in this sub-module are designed to always work. None of them should ever raise an exception--no matter what nonsense you pass in. (But don't take that as a challenge!)
get_float(o, default=_sentinel)Returns float(o), unless that conversion fails,
in which case returns the default value. If
you don't pass in an explicit default value,
the default value is o.
get_int(o, default=_sentinel)Returns int(o), unless that conversion fails,
in which case returns the default value. If
you don't pass in an explicit default value,
the default value is o.
get_int_or_float(o, default=_sentinel)Converts o into a number, preferring an int to a float.
If o is already an int or float, returns o unchanged. Otherwise,
tries int(o). If that conversion succeeds, returns the result.
Otherwise, tries float(o). If that conversion succeeds, returns
the result. Otherwise returns the default value. If you don't
pass in an explicit default value, the default value is o.
pure_virtual()A decorator for class methods. When you have a method in
a base class that's "pure virtual"--that must not be called,
but must be overridden in child classes--decorate it with
@pure_virtual(). Calling that method will throw a
NotImplementedError.
Note that the body of any function decorated with
@pure_virtual() is ignored. By convention the body
of these methods should contain only a single ellipsis,
literally like this:
class BaseClass:
@big.pure_virtual()
def on_reset(self):
...
try_float(o)Returns True if o can be converted into a float,
and False if it can't.
try_int(o)Returns True if o can be converted into an int,
and False if it can't.
big.deprecatedOld versions of functions (and classes) from big. These versions are deprecated, either because the name was changed, or the semantics were changed, or both.
Unlike the other modules, the contents of big.deprecated aren't
automatically imported into big.all. (big.all does import
the deprecated submodule, it just doesn't from deprected import *
all the symbols.)
big.fileFunctions for working with files, directories, and I/O.
fgrep(path, text, *, encoding=None, enumerate=False, case_insensitive=False)Find the lines of a file that match some text, like the UNIX fgrep utility
program.
path should be an object representing a path to an existing file, one of:
pathlib.Path object.text should be either string or bytes.
encoding is used as the file encoding when opening the file.
text is a str, the file is opened in text mode.text is a bytes object, the file is opened in binary mode.
encoding must be None when the file is opened in binary mode.If case_insensitive is true, perform the search in a case-insensitive
manner.
Returns a list of lines in the file containing text. The lines are either
strings or bytes objects, depending on the type of pattern. The lines
have their newlines stripped but preserve all other whitespace.
If enumerate is true, returns a list of tuples of (line_number, line).
The first line of the file is line number 1.
For simplicity of implementation, the entire file is read in to memory
at one time. If case_insensitive is true, fgrep also makes a lowercased
copy.
file_mtime(path)Returns the modification time of path, in seconds since the epoch.
Note that seconds is a float, indicating the sub-second with some
precision.
file_mtime_ns(path)Returns the modification time of path, in nanoseconds since the epoch.
file_size(path)Returns the size of the file at path, as an integer representing the
number of bytes.
grep(path, pattern, *, encoding=None, enumerate=False, flags=0)Look for matches to a regular expression pattern in the lines of a file,
similarly to the UNIX grep utility program.
path should be an object representing a path to an existing file, one of:
pathlib.Path object.pattern should be an object containing a regular expression, one of:
re.Pattern, initialized with either str or bytes.encoding is used as the file encoding when opening the file.
If pattern uses a str, the file is opened in text mode.
If pattern uses a bytes object, the file is opened in binary mode.
encoding must be None when the file is opened in binary mode.
flags is passed in as the flags argument to re.compile if pattern
is a string or bytes. (It's ignored if pattern is an re.Pattern object.)
Returns a list of lines in the file matching the pattern. The lines
are either strings or bytes objects, depending on the type of text.
The lines have their newlines stripped but preserve all other whitespace.
If enumerate is true, returns a list of tuples of (line_number, line).
The first line of the file is line number 1.
For simplicity of implementation, the entire file is read in to memory at one time.
Tip: to perform a case-insensitive pattern match, pass in the
re.IGNORECASE flag into flags for this function (if pattern is a string
or bytes) or when creating your regular expression object (if pattern is
an re.Pattern object.
(In older versions of Python, re.Pattern was a private type called
re._pattern_type.)
pushd(directory)A context manager that temporarily changes the directory. Example:
with big.pushd('x'):
pass
This would change into the 'x' subdirectory before
executing the nested block, then change back to
the original directory after the nested block.
You can change directories in the nested block; this won't affect pushd restoring the original current working directory upon exiting the nested block.
You can safely nest with pushd blocks.
read_python_file(path, *, newline=None, use_bom=True, use_source_code_encoding=True)Opens, reads, and correctly decodes a Python script from a file.
path should specify the filesystem path to the file; it can
be any object accepted by builtins.open (a "path-like object").
Returns a str containing the decoded Python script.
Opens the file using builtins.open.
Decodes the script using big's decode_python_script function.
The newline, use_bom and use_source_code_encoding
parameters are passed through to that function.
safe_mkdir(path)Ensures that a directory exists at path.
If this function returns and doesn't raise,
it guarantees that a directory exists at path.
If a directory already exists at path,
safe_mkdir does nothing.
If a file exists at path, safe_mkdir
unlinks path then creates the directory.
If the parent directory doesn't exist,
safe_mkdir creates that directory,
then creates path.
This function can still fail:
path could be on a read-only filesystem.path.x/y
and x is a file (not a directory).safe_unlink(path)Unlinks path, if path exists and is a file.
search_path(paths, extensions=('',), *, case_sensitive=None, preserve_extension=True, want_directories=False, want_files=True)Search a list of directories for a file. Given a sequence of directories, an optional list of file extensions, and a filename, searches those directories for a file with that name and possibly one of those file extensions.
search_path accepts the paths and extensions as parameters and
returns a search function. The search function accepts one
filename parameter and performs the search, returning either the
path to the file it found (as a pathlib.Path object) or None.
You can reuse the search function to perform as many searches
as you like.
paths should be an iterable of str or pathlib.Path objects
representing directories. These may be relative or absolute
paths; relative paths will be relative to the current directory
at the time the search function is run. Specifying a directory
that doesn't exist is not an error.
extensions should be an iterable of str objects representing
extensions. Every non-empty extension specified should start
with a period ('.') character (technically os.extsep). You
may specify at most one empty string in extensions, which
represents testing the filename without an additional
extension. By default extensions is the tuple `('',)``.
Extension strings may contain additional period characters
after the initial one.
Shell-style "globbing" isn't supported for any parameter. Both
the filename and the extension strings may contain filesystem
globbing characters, but they will only match those literal
characters themselves. ('*' won't match any character, it'll
only match a literal '*' in the filename or extension.)
case_sensitive works like the parameter to pathlib.Path.glob.
If case_sensitive is true, files found while searching must
match the filename and extension exactly. If case_sensitive
is false, the comparison is done in a case-insensitive manner.
If case_sensitive is None (the default), case sensitivity obeys
the platform default (as per os.path.normcase). In practice,
only Windows platforms are case-insensitive by convention;
all other platforms that support Python are case-sensitive
by convention.
If preserve_extension is true (the default), the search function
checks the filename to see if it already ends with one of the
extensions. If it does, the search is restricted to only files
with that extension--the other extensions are ignored. This
check obeys the case_sensitive flag; if case_sensitive is None,
this comparison is case-insensitive only on Windows.
want_files and want_directories are boolean values; the
search function will only return that type of file if the
corresponding want_ parameter is true. You can request files,
directories, or both. (want_files and want_directories
can't both be false.) By default, want_files is true and
want_directories is false.
paths and extensions are both tried in order, and the search
function returns the first match it finds. All extensions are
tried in a path entry before considering the next path.
Returns a function:
search(filename)
which returns either a pathlib.Path object on success or None
on failure.
touch(path)Ensures that path exists, and its modification time is the current time.
If path does not exist, creates an empty file.
If path exists, updates its modification time to the current time.
translate_filename_to_exfat(s)Ensures that all characters in s are legal for a FAT filesystem.
Returns a copy of s where every character not allowed in a FAT
filesystem filename has been replaced with a character (or characters)
that are permitted.
translate_filename_to_unix(s)Ensures that all characters in s are legal for a UNIX filesystem.
Returns a copy of s where every character not allowed in a UNIX
filesystem filename has been replaced with a character (or characters)
that are permitted.
big.graphA drop-in replacement for Python's
graphlib.TopologicalSorter
with an enhanced API. This version of TopologicalSorter allows modifying the
graph at any time, and supports multiple simultaneous views, allowing
iteration over the graph more than once.
See the Enhanced TopologicalSorter deep-dive for more information.
CycleError()Exception thrown by TopologicalSorter when it detects a cycle.
TopologicalSorter(graph=None)An object representing a directed graph of nodes. See Python's
graphlib.TopologicalSorter
for concepts and the basic API.
New methods on TopologicalSorter:
TopologicalSorter.copy()Returns a shallow copy of the graph. The copy also duplicates
the state of get_ready and done.
TopologicalSorter.cycle()Checks the graph for cycles. If no cycles exist, returns None. If at least one cycle exists, returns a tuple containing nodes that constitute a cycle.
TopologicalSorter.print(print=print)Prints the internal state of the graph. Used for debugging.
print is the function used for printing;
it should behave identically to the builtin print function.
TopologicalSorter.remove(node)Removes node from the graph.
If any node P depends on a node N, and N is removed,
this dependency is also removed, but P is not
removed from the graph.
Note that, while remove() works, it's slow. (It's O(N).)
TopologicalSorter is optimized for fast adds and fast views.
TopologicalSorter.reset()Resets get_ready and done to their initial state.
TopologicalSorter.view()Returns a new View object on this graph.
TopologicalSorter.ViewA view on a TopologicalSorter graph object.
Allows iterating over the nodes of the graph
in dependency order.
Methods on a View object:
TopologicalSorter.View.__bool__()Returns True if more work can be done in the
view--if there are nodes waiting to be yielded by
get_ready, or waiting to be returned by done.
Aliased to TopologicalSorter.is_active for compatibility
with graphlib.
TopologicalSorter.View.close()Closes the view. A closed view can no longer be used.
TopologicalSorter.View.copy()Returns a shallow copy of the view, duplicating its current state.
TopologicalSorter.View.done(*nodes)Marks nodes returned by ready as "done",
possibly allowing additional nodes to be available
from ready.
TopologicalSorter.View.print(print=print)Prints the internal state of the view, and its graph. Used for debugging.
print is the function used for printing;
it should behave identically to the builtin print function.
TopologicalSorter.View.ready()Returns a tuple of "ready" nodes--nodes with no predecessors, or nodes whose predecessors have all been marked "done".
Aliased to TopologicalSorter.get_ready for
compatibility with graphlib.
TopologicalSorter.View.reset()Resets the view to its initial state, forgetting all "ready" and "done" state.
big.heapFunctions for working with heap objects. Well, just one heap object really.
Heap(i=None)An object-oriented wrapper around the heapq library, designed to be
easy to use--and easy to remember how to use. The heapq library
implements a binary heap,
a data structure used for sorting;
you add objects to the heap, and you can then remove
objects in sorted order. Heaps are useful because they have are efficient
both in space and in time; they're also inflexible, in that iterating over
the sorted items is destructive.
The Heap API in big mimics the list and collections.deque objects;
this way, all you need to remember is "it works kinda like a list object".
You append new items to the heap, then popleft them off in sorted order.
By default Heap creates an empty heap. If you pass in an iterable i
to the constructor, this is equivalent to calling the extend(i) on the
freshly-constructed Heap.
In addition to the below methods, Heap objects support iteration,
len, the in operator, and use as a boolean expression. You can
also index or slice into a Heap object, which behaves as if the
heap is a list of objects in sorted order. Getting the first item
(Heap[0], aka peek) is cheap, the other operations can get very
expensive.
Methods on a Heap object:
Heap.append(o)Adds object o to the heap.
Heap.clear()Removes all objects from the heap, resetting it to empty.
Heap.copy()Returns a shallow copy of the heap. Only duplicates the heap data structures itself; does not duplicate the objects in the heap.
Heap.extend(i)Adds all the objects from the iterable i to the heap.
Heap.remove(o)If object o is in the heap, removes it. If o is not
in the heap, raises ValueError.
Heap.popleft()If the heap is not empty, returns the first item in the
heap in sorted order. If the heap is empty, raises IndexError.
Heap.append_and_popleft(o)Equivalent to calling Heap.append(o) immediately followed
by Heap.popleft(). If o is smaller than any other object
in the heap at the time it's added, this will return o.
Heap.popleft_and_append(o)Equivalent to calling Heap.popleft() immediately followed
by Heap.append(o). This method will never return o,
unless o was already in the heap before the method was called.
Heap.queueNot a method, a property. Returns a copy of the contents of the heap, in sorted order.
big.itertoolsFunctions and classes for working with iteration. Only one entry so far.
PushbackIterator(iterable=None)Wraps any iterator, allowing you to push items back on the iterator. This allows you to "peek" at the next item (or items); you can get the next item, examine it, and then push it back. If any objects have been pushed onto the iterator, they are yielded first, before attempting to yield from the wrapped iterator.
The constructor accepts one argument, an iterable, with a default of None.
If iterable is None, the PushbackIterator is created in an exhausted state.
When the wrapped iterable is exhausted (or if you passed in None
to the constructor) you can still call the push method to add items,
at which point the PushBackIterator can be iterated over again.
In addition to the following methods, PushbackIterator supports
the iterator protocol and testing for truth. A PushbackIterator
is true if iterating over it will yield at least one value.
PushbackIterator.next(default=None)Equivalent to next(PushbackIterator), but won't raise StopIteration.
If the iterator is exhausted, returns the default argument.
PushbackIterator.push(o)Pushes a value into the iterator's internal stack.
When a PushbackIterator is iterated over, and there are
any pushed values, the top value on the stack will be popped
and yielded. PushbackIterator only yields from the
iterator it wraps when this internal stack is empty.
big.logA simple and lightweight logging class, useful for performance analysis.
Not intended as a full-fledged logging facility like Python's
logging module.
default_clock()The default clock function used by the Log class.
This function returns elapsed time in nanoseconds,
expressed as an integer.
In Python 3.7+, this is
time.monotonic_ns;
in Python 3.6 this is
a custom function that calls
time.perf_counter,
then converts that time to an integer number of nanoseconds.
Log(*, clock=None)A simple and lightweight logging class, useful for performance analysis.
Not intended as a full-fledged logging facility like Python's
logging module.
Allows nesting, which is literally just a presentation thing.
The clock named parameter specifies the function the Log object
should call to get the time. This function should return an int,
representing elapsed time in nanoseconds.
To use: first, create your Log object.
log = Log()
Then log events by calling your Log object, passing in
a string describing the event.
log('text')
Enter a nested subsystem containing events with log.enter:
log.enter('subsystem')
Then later exit that subsystem with log.exit:
log.exit()
And finally print the log:
log.print()
You can also iterate over the log events using iter(log).
This yields 4-tuples:
(start_time, elapsed_time, event, depth)
start_time and elapsed_time are times, expressed as
an integer number of nanoseconds. The first event
is at start_time 0, and all subsequent start times are
relative to that time.
event is the event string you
passed in to log() (or "<subsystem> start" or
"<subsystem> end").
depth is an integer indicating how many subsystems
the event is nested in; larger numbers indicate deeper nesting.
Log.enter(subsystem)Notifies the log that you've entered a subsystem.
The subsystem parameter should be a string describing the subsystem.
This is really just a presentation
thing; all subsequent logged entries will be indented
until you make the corresponding log.exit() call.
You may nest subsystems as deeply as you like.
Log.exit()Exits a logged subsystem. See Log.enter.
Log.print(*, print=None, title="[event log]", headings=True, indent=2, seconds_width=2, fractional_width=9)Prints the log.
Keyword-only parameters:
print specifies the print function to use, default is builtins.print.
title specifies the title to print at the beginning.
Default is "[event log]". To suppress, pass in None.
headings is a boolean; if True (the default),
prints column headings for the log.
indent is the number of spaces to indent in front of log entries,
and also how many spaces to indent each time we enter a subsystem.
seconds_width is how wide to make the seconds column, default 2.
fractional_width is how wide to make the fractional column, default 9.
Log.reset()Resets the log to its initial state.
After resetting the log, the log is
empty except for the initial "log start"
message, the elapsed time is zero, and
the log has not "entered" any subsystems.
big.metadataContains metadata about big itself.
metadata.versionA
Version
object representing the current version of big.
big.schedulerA replacement for Python's sched.scheduler object,
adding full threading support and a modern Python interface.
Python's sched.scheduler object was added way back in 1991,
and it was full of clever ideas. It abstracted away the
concept of time from its interface, allowing it to be adapted
to new schemes of measuring time--including mock time, making
testing easy and repeatable. Very nice!
Unfortunately, sched.scheduler predates multithreading
becoming common, much less multicore computers. It certainly
predates threading support in Python. And its API isn't
flexible enough to correctly handle some common scenarios in
multithreaded programs:
sched.scheduler.run,
and the next scheduled event will occur at time T,
and a second thread schedules a new event which
occurs at a time < T, sched.scheduler.run won't
return any events to the first thread until time T.sched.scheduler.run,
and the next scheduled event will occur at time T,
and a second thread cancels all events,
sched.scheduler.run won't exit until time T.big's Scheduler object fixes both these problems.
Also, sched.scheduler is thirty years behind the times in
Python API design--its design predates many common modern
Python conventions. Its events are callbacks, which it
calls directly. Scheduler fixes this: its events are
objects, and you iterate over the Scheduler object to
see events as they occur.
Scheduler also benefits from thirty years of experience
with sched.scheduler. In particular, big reimplements the
relevant parts of the sched.scheduler test suite, ensuring
Scheduler will never trip over the problems discovered
by sched.scheduler over its lifetime.
Event(scheduler, event, time, priority, sequence)An object representing a scheduled event in a Scheduler.
You shouldn't need to create them manually; Event objects
are created automatically when you add events to a Scheduler.
Supports one method:
Event.cancel()Cancels this event. If this event has already been canceled,
raises ValueError.
Regulator()An abstract base class for Scheduler regulators.
A "regulator" handles all the details about time
for a Scheduler. Scheduler objects don't actually
understand time; it's all abstracted away by the
Regulator.
You can implement your own Regulator and use it
with Scheduler. Your Regulator subclass must
implement three methods: now, sleep, and wake.
It must also provide a lock attribute.
Normally a Regulator represents time using
a floating-point number, representing a fractional
number of seconds since some epoch. But this
isn't strictly necessary. Any Python object that
fulfills these requirements will work:
__le__, __eq__, __add__,
and __sub__, and these operations must be consistent in the
same way they are for number objects.a and b are instances of the time class,
and a.__le__(b) is true, then a must either be
an earlier time, or a smaller interval of time.0 must
represent both the earliest time and a zero-length
interval of time.Regulator.lockA lock
object. The Scheduler uses this lock
to protect its internal data structures.
Must support the "context manager" protocol
(__enter__ and __exit__). Entering the
object must acquire the lock; exiting must
release the lock.
This lock does not need to be recursive.
Regulator.now()Returns the current time in local units. Must be monotonically increasing; for any two calls to now during the course of the program, the later call must never have a lower value than the earlier call.
A Scheduler will only call this method while
holding this regulator's lock.
Regulator.sleep(t)Sleeps for some amount of time, in local units.
Must support an interval of 0, which should
represent not sleeping. (Though it's preferable
that an interval of 0 yields the rest of the
current thread's remaining time slice back to
the operating system.)
If wake is called on this Regulator object while a
different thread has called this function to sleep,
sleep must abandon the rest of the sleep interval
and return immediately.
A Scheduler will only call this method while
not holding this regulator's lock.
Regulator.wake()Aborts all current calls to sleep on this
Regulator, across all threads.
A Scheduler will only call this method while
holding this regulator's lock.
Scheduler(regulator=default_regulator)Implements a scheduler. The only argument is the
"regulator" object to use; the regulator abstracts away all
time-related details for the scheduler. By default Scheduler
uses an instance of SingleThreadedRegulator,
which is not thread-safe.
(If you need the scheduler to be thread-safe, pass in an
instance of a thread-safe Regulator class like
ThreadSafeRegulator.)
In addition to the below methods, Scheduler objects support
being evaluated in a boolean context (they are true if they
contain any events), and they support being iterated over.
Iterating over a Scheduler object blocks until the next
event comes due, at which point the Scheduler yields that
event. An empty Scheduler that is iterated over raises
StopIteration. You can reuse Scheduler objects, iterating
over them until empty, then adding more objects and iterating
over them again.
Scheduler.schedule(o, time, *, absolute=False, priority=DEFAULT_PRIORITY)Schedules an object o to be yielded as an event by this schedule
object at some time in the future.
By default the time value is a relative time value,
and is added to the current time; using a time value of 0
should schedule this event to be yielded immediately.
If absolute is true, time is regarded as an absolute time value.
If multiple events are scheduled for the same time, they will
be yielded by order of priority. Lowever values of
priority represent higher priorities. The default value
is Scheduler.DEFAULT_PRIORITY, which is 100. If two events
are scheduled for the same time, and have the same priority,
Scheduler will yield the events in the order they were added.
Returns an Event object, which can be used to cancel the event.
Scheduler.cancel(event)Cancels a scheduled event. event must be an object
returned by this Scheduler object. If event is not
currently scheduled in this Scheduler object,
raises ValueError.
Scheduler.queueA list of the currently scheduled Event objects,
in the order they will be yielded.
Scheduler.non_blocking()Returns an iterator for the events in the
Scheduler that only yields the events that
are currently due. Never blocks; if the next
event is not due yet, raises StopIteration.
SingleThreadedRegulator()An implementation of Regulator designed for
use in single-threaded programs. It doesn't support
multiple threads, and in particular is not thread-safe.
But it's much higher performance
than thread-safe Regulator implementations.
ThreadSafeRegulator()A thread-safe implementation of Regulator
designed for use in multithreaded programs.
big.stateLibrary code for working with simple state machines.
There are lots of popular Python libraries for implementing
state machines. But they all seem to be designed for
large-scale state machines. These libraries are
sophisticated and data-driven, with expansive APIs.
And, as a rule, they require the state to be
a passive object (e.g. an Enum), and require you to explicitly
describe every possible state transition.
This approach is great for massive, super-complex state machines--you need the features of a sophisticated library to manage all that complexity. It also enables clever features like automatically generating diagrams of your state machine, which is great!
But most of the time this level of sophistication is
unnecessary. There are lots of use cases for small scale,
simple state machines, where this data-driven approach
and expansive, complex API only gets in the way. I prefer
writing my state machines with active objects--where states
are implemented as classes, events are implemented as method
calls on those classes, and you transition to a new state by
simply overwriting a state attribute with a different state
instance.
big.state makes this style of state machine easy. It has
a deliberately minimal, simple interface--the constructor for
the main StateManager class only has four parameters,
and it only exposes three attributes. The module also has
two decorators to make your life easier. And that's it!
But even this small API surface area makes it effortless to
write large scale state machines.
(But you can also write tiny data-driven state machines too.
Although big.state makes state machines with active states
easy to write, it's agnostic about how you actually implement
your state machine. big.state makes it easy to write any
kind of state machine you like!)
big.state provides features like:
The main class in big.state is StateManager. This class
maintains the current "state" of your state machine, and
manages transitions to new states. The constructor takes
one required parameter, the initial state.
Here are my recommendations for best practices when working
with StateManager for medium-sized and larger state machines:
StateManager as an attribute of that class,
preferably called state_manager. (Your state machine
should have a "has-a" relationship with StateManager,
not an "is-a" relationship where it inherits from StateManager.)accessor
decorator--this will save you a lot of boilerplate.
If your state machine is stored in o, decorating with
accessor lets you can access the current state using
o.state instead of o.state_manager.state.BoundInnerClass
so they automatically get references to the state machine
they're a part of.dispatch
decorator to save you more boilerplate when calling the event
method. Similarly to
accessor,
this creates a new method for you that calls the equivalent
method on the current state, passing in all the arguments
it received.Here's a simple example demonstrating all this functionality.
It's a state machine with two states, On and Off, and
one event method toggle. Calling toggle transitions
the state machine from the Off state to the On state,
and vice-versa.
from big.all import accessor, BoundInnerClass, dispatch, pure_virtual, StateManager
@accessor()
class StateMachine:
def __init__(self):
self.state_manager = StateManager(self.Off())
@dispatch()
def toggle(self):
...
@BoundInnerClass
class State:
def __init__(self, state_machine):
self.state_machine = state_machine
def __repr__(self):
return f"<{type(self).__name__}>"
@pure_virtual()
def toggle(self):
...
@BoundInnerClass
class Off(State.cls):
def on_enter(self):
print("off!")
def toggle(self):
sm = self.state_machine
sm.state = sm.On() # sm.state is the accessor
@BoundInnerClass
class On(State.cls):
def on_enter(self):
print("on!")
def toggle(self):
sm = self.state_machine
sm.state = sm.Off()
sm = StateMachine()
print(sm.state)
for _ in range(3):
sm.toggle()
print(sm.state)
This code demonstrates both
accessor
and
dispatch.
accessor lets us reference the current state with sm.state
instead of sm.state_manager.state, and dispatch lets us call
sm.toggle() instead of sm.state_manager.state.toggle().
For a more complete example of working with StateManager,
see the test_vending_machine test code in tests/test_state.py
in the big source tree.
accessor(attribute='state', state_manager='state_manager')Class decorator. Adds a convenient state accessor attribute to your class.
When you have a state machine class containing a
StateManager
object, it can be wordy and inconvenient to access the state through
the state machine attribute:
class StateMachine:
def __init__(self):
self.state_manager = StateManager(self.InitialState)
...
sm = StateMachine()
# vvvvvvvvvvvvvvvvvvvv that's a lot!
sm.state_manager.state = NextState()
The accessor class decorator creates a property for you--a
shortcut that directly accesses the state attribute of
your state manager. Just decorate your state machine class
with @accessor():
@accessor()
class StateMachine:
def __init__(self):
self.state_manager = StateManager(self.InitialState)
...
sm = StateMachine()
# vvvvvv that's a lot shorter!
sm.state = NextState()
The state attribute evaluates to the same value:
sm.state == sm.state_manager.state
And setting it sets the state on your StateManager instance.
These two statements now do the same thing:
sm.state_manager.state = new_state
sm.state = new_state
By default, this decorator assumes your StateManager instance
is in the state_manager attribute, and you want to name the new
accessor attribute state. You can override these defaults;
the decorator's first parameter, attribute, should be the string used
for the new accessor attribute, and the second parameter,
state_manager, should be the name of the attribute where your
StateManager instance is stored.
For example, if your state manager is stored in an attribute called sm,
and you want the short-cut to be called st, you'd decorate your
state machine class with
@accessor(attribute='st', state_manager='sm')
dispatch(state_manager='state_manager', *, prefix='', suffix='')Decorator for state machine event methods, dispatching the event from the state machine object to its current state.
dispatch helps with the following scenario:
StateManager object.Simply create a method in your state machine class
with the correct name and parameters but a no-op body,
and decorate it with @dispatch. The dispatch
decorator will rewrite your method so it calls the
equivalent method on the current state, passing through
all the arguments.
For example, instead of writing this:
class StateMachine:
def __init__(self):
self.state_manager = StateManager(self.InitialState)
def on_sunrise(self, time, *, verbose=False):
return self.state_manager.state.on_sunrise(time, verbose=verbose)
you can literally write this, which does the same thing:
class StateMachine:
def __init__(self):
self.state_manager = StateManager(self.InitialState)
@dispatch()
def on_sunrise(self, time, *, verbose=False):
...
Here, the on_sunrise function you wrote is actually thrown away.
(That's why the body is simply one "..." statement.) Your function
is replaced with a function that gets the state_manager attribute
from self, then gets the state attribute from that StateManager
instance, then calls a method with the same name as the decorated
function, passing in using *args and **kwargs.
Note that, as a stylistic convention, you're encouraged to literally
use a single ellipsis as the body of these functions, as in the
example above. This is a visual cue to readers that the body of the
function doesn't matter. (In fact, the original on_sunrise method
above is thrown away inside the decorator, and replaced with a customized
method dispatch function.)
The state_manager argument to the decorator should be the name of
the attribute where the StateManager instance is stored in self.
The default is 'state_manager', but you can specify a different
string if you've stored your StateManager in another attribute.
For example, if your state manager is in the attribute smedley,
you'd decorate with:
@dispatch('smedley')
The prefix and suffix arguments are strings added to the
beginning and end of the method call we call on the current state.
For example, if you want the method you call to have an active verb
form (e.g. reset), but you want it to directly call an event
handler that starts with on_ by convention (e.g. on_reset),
you could do this:
@dispatch(prefix='on_')
def reset(self):
...
This is equivalent to:
def reset(self):
return self.state_manager.state.on_reset()
If you have more than one event method, instead of decorating
every event method with the same copy-and-pasted dispatch
call, it's better to call dispatch once, cache the
function it returns, and decorate with that. Like so:
my_dispatch = dispatch('smedley', prefix='on_')
@my_dispatch
def reset(self):
...
@my_dispatch
def sunrise(self):
...
State()Base class for state machine state implementation classes.
Use of this base class is optional; states can be instances
of any type except types.NoneType.
StateManager(state, *, on_enter='on_enter', on_exit='on_exit', state_class=None)Simple, Pythonic state machine manager.
Has three public attributes:
state
The current state. You transition from one state to another by assigning to this attribute.
next
The state the StateManager is transitioning to,
if it's currently in the process of transitioning to a
new state. If the StateManager isn't currently
transitioning to a new state, its next attribute is None.
And if the StateManager is currently
transitioning to a new state, its next attribute will
not be None.
During the time the manager is currently transitioning to
a new state, it's illegal to start a second transition. (In
other words: you can't assign to state while next is not
None.)
observers
A list of callables that get called during every state transition. It's initially empty; you may add and remove observers to the list as needed.
StateManager uses a copy of the observer list.)The constructor takes the following parameters:
state
The initial state. It can be any valid state object;
by default, any Python value can be a state except None.
(But also see the state_class parameter below.)
on_enter
on_enter represents a method call on states called when
entering that state. The value itself is a string used
to look up an attribute on state objects; by default
on_enter is the string 'on_enter', but it can be any legal
Python identifier string, or any false value.
If on_enter is a valid identifier string, and this
StateManager
object transitions to a state object O, and O has an attribute
with this name,
StateManager
will call that attribute (with no
arguments) immediately after transitioning to that state.
Passing in a false value for on_enter disables this behavior.
on_enter is called immediately after the transition is complete,
which means you're expressly permitted to make a state transition
inside an on_enter call.
If defined, on_exit will be called on
the initial state object, from inside the StateManager constructor.
on_exit
on_exit is similar to on_enter, except the attribute is
called when transitioning away from a state object.
Its default value is 'on_exit'.
on_exit is called
during the state transition, which means you're expressly
forbidden from making a state transition inside an on_exit call.
state_class
state_class is used to enforce that this StateManager
only ever transitions to valid state objects.
It should be either None or a class. If it's a class,
the StateManager object will require every value assigned
to its state attribute to be an instance of that class.
If it's None, states can be any object (except None).
To transition to a new state, simply assign to the 'state' attribute.
state_class is None, you may use any value as a state
except None.state while currently
transitioning to a new state. (Or, in other words,
at any time self.next is not None.)on_exit method,
it will be called (with zero arguments) during
the the transition to the next state. This means it's
illegal to initiate a state transition inside an on_exit
call.state that has an on_enter
attribute, that method will be called (with zero
arguments) immediately after we have transitioned to that
state. This means it's permitted to initiate a state
transition inside an on_enter call.state_manager.state is already foo,
state_manager.state = foo will raise an exception.If you have an StateManager instance called state_manager,
and you transition it to new_state:
state_manager.state = new_state
StateManager will execute the following sequence of events:
state_manager.next to new_state.
state_manager is "transitioning"
to the new state.state_manager.state has an on_exit attribute,
call state_manager.state.on_exit().o in the state_manager.observer list,
call o(self).state_manager.next to None.state_manager.state to new_state.
state_manager is now "in" the new state.state_manager.state has an on_enter attribute,
call state_manager.state.on_enter().TransitionError()There are only two types of illegal state transitions:
An attempted state transition while we're in the process
of transitioning to another state. In other words,
if state_manager is your StateManager object, you can't
set state_manager.state when state_manager.next is not None.
An attempt to transition to the current state. This is illegal:
state_manager = StateManager()
state_manager.state = foo
state_manager.state = foo # <-- this statement raises TransitionError
Note that transitioning to a different but identical object is expressly permitted.
big.textFunctions for working with text strings. There are
several families of functions inside the text module;
for a higher-level view of those families, read the
following deep-dives:
All the functions in big.text will work with either
str or bytes objects, except the three
Word wrapping and formatting
functions. When working with bytes,
by default the functions will only work with ASCII
characters.
The big text functions all support both str and bytes.
The functions all automatically detect whether you passed in
str or bytes using an
intentionally simple and predictable process, as follows:
At the start of each function, it'll test its first "string"
argument to see if it's a bytes object.
is_bytes = isinstance(<argument>, bytes)
If isinstance returns True, the function assumes all arguments are
bytes objects. Otherwise the function assumes all arguments
are str objects.
As a rule, no further testing, casting, or catching exceptions is done.
Functions that take multiple string-like parameters require all such arguments to be the same type. These functions will check that all such arguments are of the same type.
Subclasses of str and bytes will also work; anywhere you
should pass in a str, you can also pass in a subclass of
str, and likewise for bytes.
ascii_linebreaksA tuple of str objects, representing every line-breaking whitespace
character defined by ASCII.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'.
If you don't want to include this string, use ascii_linebreaks_without_crlf instead.
See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
ascii_linebreaks_without_crlfEquivalent to ascii_linebreaks without '\r\n'.
ascii_whitespaceA tuple of str objects, representing every whitespace
character defined by ASCII.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'.
If you don't want to include this string, use ascii_whitespace_without_crlf instead.
See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
ascii_whitespace_without_crlfEquivalent to ascii_whitespace without '\r\n'.
bytes_linebreaksA tuple of bytes objects, representing every line-breaking whitespace
character recognized by the Python bytes object.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains b'\r\n'.
If you don't want to include this string, use bytes_linebreaks_without_crlf instead.
See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
bytes_linebreaks_without_crlfEquivalent to bytes_linebreaks with '\r\n' removed.
bytes_whitespaceA tuple of bytes objects, representing every line-breaking whitespace
character recognized by the Python bytes object. (bytes.isspace,
bytes.split, etc will tell you which characters are considered whitespace...)
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains b'\r\n'.
If you don't want to include this string, use bytes_whitespace_without_crlf instead.
See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
bytes_whitespace_without_crlfEquivalent to bytes_whitespace without '\r\n'.
combine_splits(s, *split_arrays)Takes a string s, and one or more "split arrays",
and applies all the splits to s. Returns
an iterator of the resulting string segments.
A "split array" is an array containing the original
string, but split into multiple pieces. For example,
the string "a b c d e" could be split into the
split array ["a ", "b ", "c ", "d ", "e"].
For example,
combine_splits('abcde', ['abcd', 'e'], ['a', 'bcde'])
returns ['a', 'bcd', 'e'].
Note that the split arrays must contain all the
characters from s. ''.join(split_array) must recreate s.
combine_splits only examines the lengths of the strings
in the split arrays, and makes no attempt to infer
stripped characters. (So, don't use the string's .split
method if you want to use combine_splits. Instead, consider
big's
multisplit
with keep=True or keep=ALTERNATING.)
decode_python_script(script, *, newline=None, use_bom=True, use_source_code_encoding=True)Correctly decodes a Python script from a bytes string.
script should be a bytes object containing an encoded Python script.
Returns a str containing the decoded Python script.
By default, Python 3 scripts must be encoded using UTF-8.
(This was established by PEP 3120.)
Python scripts are allowed to use other encodings, but when they do so
they must explicitly specify what encoding they used. Python defines
two methods for scripts to specify their encoding; decode_python_script
supports both.
The first method uses a "byte order mark", aka "BOM". This is a sequence of bytes at the beginning of the file that indicate the file's encoding.
If use_bom is true (the default), decode_python_script will
recognize a BOM if present, and decode the file using the encoding
specified by the BOM. Note that decode_python_script removes the BOM
when it decodes the file.
The second method is called a "source code encoding", and it was defined in PEP 263. This is a "magic comment" that must be one of the first two lines of the file.
If use_source_code_encoding is true (the default), decode_python_script
will recognize a source code encoding magic comment, and use that to decode
the file. (decode_python_script leaves the magic comment in place.)
If both these "use_" keyword-only parameters are true (the default),
decode_python_script can handle either, both, or neither. In this case,
if script contains both a BOM and a source code encoding magic comment,
the script will be decoded using the encoding specified by the BOM, and the
source code encoding must agree with the BOM.
The newline parameter supports Python's "universal newlines" convention.
This behaves identically to the newline parameter for Python's
open()
function.
Delimiter(close, *, escape='', multiline=True, quoting=False)Class representing a delimiter for
split_delimiters.
close is the closing delimiter character. It must be a valid
string or bytes object, and cannot be a backslash ('"\"' or b"\\").
If escape is true, it should be a string;
when inside this delimiter, you can escape the trailing
delimiter with this string. If escape is false,
there is no escape string for this delimiter.
quoting is a boolean: does this set of delimiters "quote" the text inside?
When an open delimiter enables quoting, split_delimiters will ignore all
other delimiters in the text until it encounters the matching close delimiter.
(Single- and double-quotes set this to True.)
If escape is true, quoting must also be true.
If multiline is true, the closing delimiter may be on the current line
or any subsequent line. If multiline is false, the closing delimiter
must appear on the current line.
encode_strings(o, *, encoding='ascii')Converts an object o from str to bytes.
If o is a container, recursively converts
all objects and containers inside.
o and all objects inside o must be either
bytes, str, dict, set, list, tuple, or a subclass
of one of those.
Encodes every string inside using the encoding
specified in the encoding parameter, default
is 'ascii'.
Handles nested containers.
If o is of, or contains, a type not listed above,
raises TypeError.
format_map(s, mapping)Unlike str.format_map, big's format_map allows you to perform
string replacements inside of other string replacements:
big.format_map("{{extension} size}",
{'extension': 'mp3', 'mp3 size': 8555})
returns the string '8555'.
Another difference between str.format_map and big's format_map
is how you escape curly braces. To produce a '{' or '}' in the
output string, add '\{' or '\}' respectively. (To produce a
backslash, '\\', you must put four backslashes, '\\\\'.)
See the documentation for str.format_map for more.
gently_title(s, *, apostrophes=None, double_quotes=None)Uppercases the first character of every word in s,
leaving the other letters alone. s should be str or bytes.
(For the purposes of this algorithm, words are any contiguous run of non-whitespace characters.)
This function will also capitalize the letter after an apostrophe if the apostrophe:
'('), orIn this last case, the O or D will also be capitalized.
Finally, this function will capitalize the letter after a quote mark if the quote mark:
(A run of consecutive apostrophes and/or quote marks is considered one quote mark for the purposes of capitalization.)
All these rules mean gently_title correctly handles
internally quoted strings:
He Said 'No I Did Not'
and contractions that start with an apostrophe:
'Twas The Night Before Christmas
as well as certain Irish, French, and Italian names:
Peter O'Toole
Lord D'Arcy
If specified, apostrophes should be a str
or bytes object containing characters that
should be considered apostrophes. If apostrophes
is false, and s is bytes, apostrophes is set to
a bytes object containing the only ASCII apostrophe character:
'
If apostrophes is false and s is str, apostrophes
is set to a string containing these Unicode apostrophe code points:
'‘’‚‛
Note that neither of these strings contains the "back-tick" character:
`
This is a diacritical used for modifying letters, and isn't used as an apostrophe.
If specified, double_quotes should be a str
or bytes object containing characters that
should be considered double-quote characters.
If double_quotes is false, and s is bytes,
double_quotes is set to a bytes object containing
the only ASCII double-quote character:
"
If double_quotes is false and s is str, double_quotes
is set to a string containing these Unicode double-quote code points:
"“”„‟«»‹›
int_to_words(i, *, flowery=True, ordinal=False)Converts an integer into the equivalent English string.
int_to_words(2) -> "two"
int_to_words(35) -> "thirty-five"
If the keyword-only parameter flowery is true (the default),
you also get commas and the word and where you'd expect them.
(When flowery is true, int_to_words(i) produces identical
output to inflect.engine().number_to_words(i), except for
negative numbers: inflect starts negative numbers with
"minus", big starts them with "negative".)
If the keyword-only parameter ordinal is true,
the string produced describes that ordinal number
(instead of that cardinal number). Ordinal numbers
describe position, e.g. where a competitor placed in
a competition. In other words, int_to_words(1)
returns the string 'one', but
int_to_words(1, ordinal=True) returns the
string 'first'.
Numbers >= 10**66 (one thousand vigintillion)
are only converted using str(i). Sorry!
linebreaksA tuple of str objects, representing every line-breaking
whitespace character recognized by the Python str object.
Identical to str_linebreaks.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'. See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
linebreaks_without_crlfEquivalent to linebreaks without '\r\n'.
LineInfo(lines, line, line_number, column_number, *, leading=None, trailing=None, end=None, indent=0, match=None, **kwargs)The first object in the 2-tuple yielded by a
lines
iterator, containing metadata about the line.
Every parameter to the constructor is stored as
an attribute of the new LineInfo object using
the same identifier.
line is the original unmodified line, split
from the original s input to lines. Note
that line includes the trailing newline character,
if any.
line_number is the line number of this line.
column_numberis the starting column of the accompanyinglinestring (the second entry in the 2-tuple yielded bylines`).
leading and trailing are strings that have
been stripped from the beginning or end of the
original line, if any. (Not counting the
line-terminating linebreak character.)
end is the linebreak character that terminated
the current line, if any. If the s passed in to
lines is an iterator yielding strings, end
will always be an empty string.
indent is the indent level of the current line,
represented as an integer. See lines_strip_indent.
If the indent level hasn't been measured yet this
should be 0.
match is the re.Match object that matched this
line, if any. See lines_grep.
You can add your own fields by passing them in
via **kwargs; you can also add new attributes
or modify existing attributes as needed from
inside a "lines modifier" function.
For more information, see the deep-dive on
lines and lines modifier functions.
LineInfo.clip_leading(line, s)Clip the leading substring s from line.
s may be either a string (str or bytes) or an int.
If s is a string, it must match the leading substring
of line you wish clipped. If s is an int, it should
representing the number of characters you want clipped
from the beginning of s.
Returns line with s clipped; also appends
the clipped portion to self.leading, and updates
self.column_number to represent the column number
where line now starts. (If the clipped portion of
line contains tabs, it's detabbed using lines.tab_width
and the detab method on the clipped substring before it
is measured.)
LineInfo.clip_trailing(line, s)Clip the trailing substring s from line.
s may be either a string (str or bytes) or an int.
If s is a string, it must match the trailing substring
of line you wish clipped. If s is an int, it should
representing the number of characters you want clipped
from the end of s.
Returns line with s clipped; also appends
the clipped portion to self.trailing.
LineInfo.copy()Returns a copy of the LineInfo object,
including all current state (e.g. leading, trailing).
lines(s, separators=None, *, clip_linebreaks=True, line_number=1, column_number=1, tab_width=8, **kwargs)A "lines iterator" object. Splits s into lines, and iterates yielding those lines.
When iterated over, yields 2-tuples:
(info, line)
where info is a
LineInfo
object, and line is a str or bytes object.
s can be str, bytes, or an iterable.
If s is neither str nor bytes, s must be an iterable.
The iterable should either yield individual strings, which is the
line, or it should yield a tuple containing two strings, in which case
the strings should be the line and the line-terminating newline respectively.
All "string" objects yielded by this iterable should be homogeneous,
either str or bytes.
separators should either be None or an iterable of separator strings,
as per the separators argument to multisplit. If s is str or bytes,
it will be split using multisplit, using these separators. If
separators is None--which is the default value--and s is str or bytes,
s will be split at linebreak characters. (If s is neither str nor bytes,
separators must be None.)
line_number is the starting line number given to the first
LineInfo
object. This number is then incremented for every subsequent line.
column_number is the starting column number given to every
LineInfo
object. This number represents the leftmost column of every line.
tab_width isn't used by lines itself, but is stored internally and
may be used by other lines modifier functions (e.g. lines_strip_indent,
lines_convert_tabs_to_spaces). Similarly, all keyword arguments passed
in via kwargs are stored internally and can be accessed by user-defined
lines modifier functions.
lines copies the line-breaking character (usually \n) from each line
to info.end. If clip_linebreaks is true (the default), lines will clip
the line-breaking character off the end of each line. If clip_linebreaks
is false, lines will leave the line-breaking character in place.
You can pass in an instance of a subclass of bytes or str
for s and elements of separators, but the base class
for both must be the same (str or bytes). lines will
only yield str or bytes objects for line.
Composable with all the lines_ modifier functions in the big.text module.
For more information, see the deep-dive on
lines and lines modifier functions.
lines_convert_tabs_to_spaces(li)A lines modifier function. Converts tabs to spaces for the lines
of a "lines iterator", using the tab_width passed in to
lines.
For more information, see the deep-dive on
lines and lines modifier functions.
lines_filter_empty_lines(li)A lines modifier function. Filters out the empty lines of a "lines iterator".
Preserves the line numbers. If lines 0 through 2 are empty,
line 3 is 'a'``, line 4 is empty, and line 5 is 'b'``, this will yield:
(LineInfo(line='a', line_number=3), 'a')
(LineInfo(line='b', line_number=5), 'b')
For more information, see the deep-dive on
lines and lines modifier functions.
lines_filter_line_comment_lines(li, comment_markers)A lines modifier function. Filters out comment lines from the
lines of a "lines iterator". Comment lines are lines whose
first non-whitespace characters appear in the iterable of
comment_separators strings passed in.
What's the difference between
lines_strip_line_comments
and
lines_filter_line_comment_lines?
lines_filter_line_comment_lines
only recognizes lines that start with a comment separator
(ignoring leading whitespace). Also, it filters out those
lines completely, rather than modifying the line.lines_strip_line_comments
handles comment markers anywhere in the line, and it can also ignore
comments inside quoted strings. It truncates the line but still always
yields the line.For more information, see the deep-dive on
lines and lines modifier functions.
lines_containing(li, s, *, invert=False)A lines modifier function. Only yields lines
that contain s. (Filters out lines that
don't contain s.)
If invert is true, returns the opposite--filters
out lines that contain s.
For more information, see the deep-dive on
lines and lines modifier functions.
lines_grep(li, pattern, *, invert=False, flags=0, match='match')A lines modifier function. Only yields lines
that match the regular expression pattern.
(Filters out lines that don't match pattern.)
Stores the resulting re.Match object in info.match.
pattern can be str, bytes, or an re.Pattern object.
If pattern is not an re.Pattern object, it's compiled
with re.compile(pattern, flags=flags).
If invert is true, returns the opposite--filters
out lines that match pattern.
The match parameter specifies the
LineInfo
attribute name to
write to. By default it writes to info.match; you can specify
any valid identifier, and it will instead write the re.Match
object (or None) to the identifier you specify.
For more information, see the deep-dive on
lines and lines modifier functions.
(In older versions of Python, re.Pattern was a private type called
re._pattern_type.)
lines_rstrip(li, separators=None)A lines modifier function. Strips trailing whitespace from the lines of a "lines iterator".
separators is an iterable of separators, like the argument
to multistrip. The default value is None, which means
lines_rstrip strips all trailing whitespace characters.
All characters removed are clipped to info.trailing
as appropriate. If the line is non-empty before stripping, and
empty after stripping, the entire line is clipped to info.trailing.
For more information, see the deep-dive on
lines and lines modifier functions.
lines_sort(li, *, key=None, reverse=False)A lines modifier function. Sorts all input lines before yielding them.
If key is specified, it's used as the key parameter to list.sort.
The key function will be called with the (info, line)`` tuple yielded by the *lines iterator.* If keyis a false value,lines_sort`
sorts the lines lexicographically, from lowest to highest.
If reverse is true, lines are sorted from highest to lowest.
For more information, see the deep-dive on
lines and lines modifier functions.
lines_strip(li, separators=None)A lines modifier function. Strips leading and trailing strings from the lines of a "lines iterator".
separators is an iterable of separators, like the argument
to multistrip. The default value is None, which means
lines_strip strips all leading and trailing whitespace characters.
All characters are clipped to info.leading and info.trailing
as appropriate. If the line is non-empty before stripping, and
empty after stripping, the entire line is clipped to info.trailing.
For more information, see the deep-dive on
lines and lines modifier functions.
lines_strip_indent(li)A lines modifier function. Strips leading whitespace and tracks the indent level.
The indent level is stored in the
LineInfo
object's attribute
indent. indent is an integer, the ordinal number of the current
indent; if the text has been indented three times, indent will be 3.
Strips any leading whitespace from the line, updating the
LineInfo
attributes leading and column_number as needed.
Uses an intentionally simple algorithm.
Only understands tab and space characters as indent characters.
Internally detabs to spaces first for consistency, using the
tab_width passed in to lines.
You can only dedent out to a previous indent.
Raises IndentationError if there's an illegal dedent.
For more information, see the deep-dive on
lines and lines modifier functions.
lines_strip_line_comments(li, line_comment_markers, *, quotes=(), escape='\\', multiline_quotes=())A lines modifier function. Strips comments from the lines
of a "lines iterator". Comments are substrings that indicate
the rest of the line should be ignored; lines_strip_line_comments
truncates the line at the beginning of the leftmost comment
separator.
line_comment_markers should be an iterable of line comment
marker strings. These are strings that denote a "line comment",
which is to say, a comment that starts at that marker and extends
to the end of the line.
By default, quotes and multiline_quotes are both false,
in which case lines_strip_line_comments will truncate each
line, starting at the leftmost comment marker, and yield
the resulting line. If the line doesn't contain any comment
markers, lines_strip_line_comments will yield it unchanged.
However, the syntax of the text you're parsing might support
quoted strings, and if so, comment marks in those quoted strings
should be ignored. lines_strip_quoted_strings supports this
too, with its escape, quotes, and multiline_quotes parameters.
If quotes is true, it must be an iterable of quote characters.
lines_strip_line_comments will parse the line using big's
split_quoted_strings
function and ignore comment
markers inside quoted strings. Quote marks must be balanced; if you
open a quoted string, you must close it. If a line ends with an
quoted string still open, lines_strip_line_comments will raise
SyntaxError.
multiline_quotes is similar to quotes, except quoted strings are
permitted to span lines. If the iterator stops iteration with a
multiline quoted string still open, lines_strip_line_comments
will raise SyntaxError.
escape specifies an escape string to allow having the closing quote
marker inside a quoted string without closing ending the string.
If false, there is no escape string.
What's the difference between
lines_strip_line_comments
and
lines_filter_line_comment_lines?
lines_filter_line_comment_lines
only recognizes lines that start with a comment separator
(ignoring leading whitespace). Also, it filters out those lines
completely, rather than modifying the line.lines_strip_line_comments
handles comment markers anywhere in the line, and it can even ignore
comments inside quoted strings. It always yields the line, whether or
not it's truncated the line.For more information, see the deep-dive on
lines and lines modifier functions.
merge_columns(*columns, column_separator=" ", overflow_response=OverflowResponse.RAISE, overflow_before=0, overflow_after=0)Merge an arbitrary number of separate text strings into columns. Returns a single formatted string.
columns should be an iterable of "column tuples".
Each column tuple should contain three items:
(text, min_width, max_width)
text should be a single string, either str or bytes,
with newline characters separating lines. min_width
and max_width are the minimum and maximum permissible widths
for that column, not including the column separator (if any).
Note that this function does not text-wrap the text of the
columns. The text in the columns should already be broken
into lines and separated by newline characters. (Lines in
that are longer than that column tuple's max_width are
handled with the overflow_strategy, described below.)
column_separator is printed between every column.
overflow_strategy tells merge_columns how to handle a column
with one or more lines that are wider than that column's max_width.
The supported values are:
OverflowStrategy.RAISE: Raise an OverflowError. The default.OverflowStrategy.INTRUDE_ALL: Intrude into all subsequent columns
on all lines where the overflowed column is wider than its max_width.OverflowStrategy.DELAY_ALL: Delay all columns after the overflowed
column, not beginning any until after the last overflowed line
in the overflowed column.When overflow_strategy is INTRUDE_ALL or DELAY_ALL, and
either overflow_before or overflow_after is nonzero, these
specify the number of extra lines before or after
the overflowed lines in a column.
For more information, see the deep-dive on Word wrapping and formatting.
multipartition(s, separators, count=1, *, reverse=False, separate=True)Like str.partition, but supports partitioning based on multiple
separator strings, and can partition more than once.
s can be either str or bytes.
separators should be an iterable of objects of the same type as s.
By default, if any of the strings in separators are found in s,
returns a tuple of three strings: the portion of s leading up to
the earliest separator, the separator, and the portion of s after
that separator. Example:
multipartition('aXbYz', ('X', 'Y')) => ('a', 'X', 'bYz')
If none of the separators are found in the string, returns
a tuple containing s unchanged followed by two empty strings.
multipartition is greedy: if two or more separators appear at
the leftmost location in s, multipartition partitions using
the longest matching separator. For example:
multipartition('wxabcyz', ('a', 'abc')) => `('wx', 'abc', 'yz')`
Passing in an explicit count lets you control how many times
multipartition partitions the string. multipartition will always
return a tuple containing (2*count)+1 elements.
Passing in a count of 0 will always return a tuple containing s.
If separate is true, multiple adjacent separator strings behave
like one separator. Example:
big.text.multipartition('aXYbYXc', ('X', 'Y',), count=2, separate=False) => ('a', 'XY', 'b', 'YX', 'c')
big.text.multipartition('aXYbYXc', ('X', 'Y',), count=2, separate=True ) => ('a', 'X', '', 'Y', 'bYXc')
If reverse is true, multipartition behaves like str.rpartition.
It partitions starting on the right, scanning backwards through s
looking for separators.
For more information, see the deep-dive on
The multi- family of string functions.
multisplit(s, separators=None, *, keep=False, maxsplit=-1, reverse=False, separate=False, strip=False)Splits strings like str.split, but with multiple separators and options.
s can be str or bytes.
separators should either be None (the default),
or an iterable of str or bytes, matching s.
If separators is None and s is str,
multisplit will use big.whitespace
as separators.
If separators is None and s is bytes,
multisplit will use big.ascii_whitespace
as separators.
Returns an iterator yielding the strings split from s. If keep
is true (or ALTERNATING), and strip is false, joining these strings
together will recreate s.
multisplit is greedy: if two or more separators start at the same
location in s, multisplit splits using the longest matching separator.
For example:
big.multisplit('wxabcyz', ('a', 'abc'))
yields 'wx' then 'yz'.
keep indicates whether or not multisplit should preserve the separator
strings in the strings it yields. It supports four values:
false (the default)
Discard the separators.
true (apart from ALTERNATING and AS_PAIRS)
Append the separators to the end of the split strings.
You can recreate the original string by using "".join
to join the strings yielded by multisplit.
ALTERNATING
Yield alternating strings in the output: strings consisting
of separators, alternating with strings consisting of
non-separators. If "separate" is true, separator strings
will contain exactly one separator, and non-separator strings
may be empty; if "separate" is false, separator strings will
contain one or more separators, and non-separator strings
will never be empty, unless "s" was empty.
You can recreate the original string by using "".join
to join the strings yielded by multisplit.
AS_PAIRS
Yield 2-tuples containing a non-separator string and its subsequent separator string. Either string may be empty; the separator string in the last 2-tuple will always be empty, and if "s" ends with a separator string, both strings in the final 2-tuple will be empty.
separate indicates whether multisplit should consider adjacent
separator strings in s as one separator or as multiple separators
each separated by a zero-length string. It supports two values:
false (the default)
Group separators together. Multiple adjacent separators behave as if they're one big separator.
true
Don't group separators together. Each separator should
split the string individually, even if there are no
characters between two separators. (multisplit will
behave as if there's a zero-character-wide string between
adjacent separators.)
strip indicates whether multisplit should strip separators from
the beginning and/or end of s. It supports five values:
false (the default)
true (apart from LEFT, RIGHT, and PROGRESSIVE)
LEFT
RIGHT
PROGRESSIVE
maxsplit should be either an integer or None. If maxsplit is an
integer greater than -1, multisplit will split text no more than
maxsplit times.
reverse changes where multisplit starts splitting the string, and
what direction it moves through the string when parsing.
false (the default)
true
Splitting starting from the end of the string and parsing moving
left has two effects. First, if maxsplit is a number
greater than 0, the splits will start at the end of the string
rather than the beginning. Second, if there are overlapping
instances of separators in the string, multisplit will prefer
the rightmost separator rather than the leftmost. Consider this
example, where reverse is false:
multisplit("A x x Z", (" x ",), keep=big.ALTERNATING) => "A", " x ", "x Z"
If you pass in a true value for reverse, multisplit will prefer
the rightmost overlapping separator:
multisplit("A x x Z", (" x ",), keep=big.ALTERNATING, reverse=True) => "A x", " x ", "Z"
For more information, see the deep-dive on
The multi- family of string functions.
multistrip(s, separators, left=True, right=True)Like str.strip, but supports stripping multiple substrings from s.
Strips from the string s all leading and trailing instances of strings
found in separators.
s should be str or bytes.
separators should be an iterable of either str or bytes
objects matching the type of s.
If left is a true value, strips all leading separators
from s.
If right is a true value, strips all trailing separators
from s.
Processing always stops at the first character that doesn't match one of the separators.
Returns a copy of s with the leading and/or trailing
separators stripped. (If left and right are both
false, returns s unchanged.)
For more information, see the deep-dive on
The multi- family of string functions.
normalize_whitespace(s, separators=None, replacement=None)Returns s, but with every run of consecutive
separator characters turned into a replacement string.
By default turns all runs of consecutive whitespace
characters into a single space character.
s may be str or bytes.
separators should be an iterable of either str or bytes
objects, matching s.
replacement should be either a str or bytes object,
also matching s, or None (the default).
If replacement is None, normalize_whitespace will use
a replacement string consisting of a single space character.
Leading or trailing runs of separator characters will be replaced with the replacement string, e.g.:
normalize_whitespace(" a b c") == " a b c"
python_delimitersA delimiters mapping suitable for use as the delimiters
argument for split_delimiters.
python_delimiters defines all the delimiters for Python, and
is able to correctly split any modern Python text at its delimiter boundaries.
python_delimiters changes the rules a little bit for split_delimiters:
When you use split_delimiters with python_delimiters, it yields four
values, not three. The fourth value is change. See split_delimiters
for more information.
If you make a copy of python_delimiters and modify it, you will break
its semantics. Internally python_delimiters is really just a symbolic
token, and split_delimiters uses a secret, internal-only, manually
modified set of delimiters. This was necessary because the Delimiters
object isn't sophisticated enough (yet) to express all the semantics
needed for python_delimiters.
When you call split_delimiters and pass in python_delimiters,
you must include the linebreak characters in the text string(s)
you pass in. This is necessary to support the comment delimiter
correctly, and to enforce the no-linebreaks-inside-single-quoted-strings rule.
If you're using big.lines to pre-process a script before passing
it in to split_delimiters, consider calling it with clip_linebreaks=False.
Here's a list of all the delimiters recognized by python_delimiters:
(), {}, and [].', ", ''', and """.b, f, r, and u, in both lower and upper case.{{ and }} are passed through in text
unmodified.!) and format spec (:) inside the curly braces
inside an f-string. These two delimiters are the only two that
use the new change value yielded by split_delimiters.# and "close" with either a
linebreak (\n) or a carriage return (\r). (Python's
"universal newlines" support should mean you won't normally
see carriage returns here... unless you specifically permit them.)See also python_delimiters_version.
python_delimiters_versionA dictionary mapping strings containing a Python major and minor version to
python_delimiters objects.
By default, python_delimiters parses the version of the Python language
matching the version it's being run under. If you run Python 3.12, and
call big.split_delimiters and pass in python_delimiters, it will split
delimiters based on Python 3.12. If you instead wanted to parse using the
semantics from Python 3.8, you would instead pass in python_delimiters_version['3.8']
as the delimiters argument to split_delimiters.
There are entries in python_split_delimiters for every version of
Python supported by big (currently 3.6 to 3.13).
re_partition(text, pattern, count=1, *, flags=0, reverse=False)Like str.partition, but pattern is matched as a regular expression.
text can be a string or a bytes object.
pattern can be a string, bytes, or re.Pattern object.
text and pattern (or pattern.pattern) must be the same type.
If pattern is found in text, returns a tuple
(before, match, after)
where before is the text before the matched text,
match is the re.Match object resulting from the match, and
after is the text after the matched text.
If pattern appears in text multiple times,
re_partition will match against the first (leftmost)
appearance.
If pattern is not found in text, returns a tuple
(text, None, '')
where the empty string is str or bytes as appropriate.
Passing in an explicit count lets you control how many times
re_partition partitions the string. re_partition will always
return a tuple containing (2*count)+1 elements, and
odd-numbered elements will be either re.Match objects or None.
Passing in a count of 0 will always return a tuple containing s.
If pattern is a string or bytes object, flags is passed in
as the flags argument to re.compile.
If reverse is true, partitions starting at the right,
like re_rpartition.
Note:
re_partitionsupports partitioning on subclasses ofstrorbytes, and thebeforeandafterobjects in the tuple returned will be slices of thetextobject. However, thematchobject doesn't honor this this; the objects it returns from e.g.match.groupwill always be of the base type, eitherstrorbytes. This isn't fixable, as you can't createre.Matchobjects in Python, nor can you subclass it.
(In older versions of Python, re.Pattern was a private type called
re._pattern_type.)
re_rpartition(text, pattern, count=1, *, flags=0)Like str.rpartition, but pattern is matched as a regular expression.
text can be a str or bytes object.
pattern can be a str, bytes, or re.Pattern object.
text and pattern (or pattern.pattern) must be the same type.
If pattern is found in text, returns a tuple
(before, match, after)
where before is the text before the matched text,
match is the re.Match object resulting from the match, and
after is the text after the matched text.
If pattern appears in text multiple times,
re_partition will match against the last (rightmost)
appearance.
If pattern is not found in text, returns a tuple
('', None, text)
where the empty string is str or bytes as appropriate.
Passing in an explicit count lets you control how many times
re_rpartition partitions the string. re_rpartition will always
return a tuple containing (2*count)+1 elements, and
odd-numbered elements will be either re.Match objects or None.
Passing in a count of 0 will always return a tuple containing s.
If pattern is a string, flags is passed in
as the flags argument to re.compile.
Note:
re_rpartitionsupports partitioning on subclasses ofstrorbytes, and thebeforeandafterobjects in the tuple returned will be slices of thetextobject. However, thematchobject doesn't honor this this; the objects it returns from e.g.match.groupwill always be of the base type, eitherstrorbytes. This isn't fixable, as you can't createre.Matchobjects in Python, nor can you subclass it.
(In older versions of Python, re.Pattern was a private type called
re._pattern_type.)
reversed_re_finditer(pattern, string, flags=0)An iterator. Behaves almost identically to the Python
standard library function re.finditer, yielding
non-overlapping matches of pattern in string. The difference
is, reversed_re_finditer searches string from right to left.
pattern can be str, bytes, or a precompiled re.Pattern object.
If it's str or bytes, it'll be compiled
with re.compile using the flags you passed in.
string should be the same type as pattern (or pattern.pattern).
split_delimiters(s, delimiters={...}, *, state=(), yields=None)Splits a string s at delimiter substrings.
s may be str or bytes.
delimiters may be either None or a mapping of open delimiter
strings to Delimiter objects. The open delimiter strings,
close delimiter strings, and escape strings must match the type
of s (either str or bytes).
If delimiters is None, split_delimiters uses a default
value matching these pairs of delimiters:
() [] {} "" ''
The first three delimiters allow multiline, disable
quoting, and have no escape string. The last two
(the quote mark delimiters) enable quoting, disallow
multiline, and specify their escape string as a
single backslash. (This default value automatically
supports both str and bytes.)
state specifies the initial state of parsing. It's an iterable
of open delimiter strings specifying the initial nested state of
the parser, with the innermost nesting level on the right.
If you wanted split_delimiters to behave as if it'd already seen
a '(' and a '[', in that order, pass in ['(', '[']
to state.
(Tip: Use a list as a stack to track the state of split_delimiters.
Push open delimiters with .append, and pop them off using .pop
whenever you see a close delimiter. Since split_delimiters ensures
that open and close delimiters match, you don't need to check them
yourself!)
Yields a object of type SplitDelimitersValue. This object
contains five fields:
text
A string, the text before the next opening, closing, or changing delimiter.
open
A string, the trailing opening delimiter.
close
A string, the trailing closing delimiter.
change
A string, the trailing change delimiter.
yields
An integer, either 3 or 4.
At least one of the four strings will always be non-empty.
(Only one of open, close, and change will ever be non-empty in
a single SplitDelimitersValue object.) If s doesn't end with
an opening or closing delimiter, the final value yielded will
have empty strings for open, close, and change.
The yields parameter to split_delimiters affects iteration over
a SplitDelimitersValue object. yields may be None, 3, or 4:
yields is 3, when iterating over a SplitDelimitersValue
object, it will yield text, open, and close in that order.yields is 4, when iterating over a SplitDelimitersValue
object, it will yield text, open, close, and change in that order.None (the default), split_delimiters will use
a value of 4 if its delimiters argument is python_delimiters,
and a value of 3 otherwise.(The yields parameter exists because previously split_delimiters
always yielded an tuple containing three string values. python_delimiters
required adding the fourth string value, change. Eventually
split_delimiters will always yield an object yielding four values,
but big is allowing for a transition period to minimize code breakage.
See the release notes for big version 0.12.5 for more information.)
You may not specify backslash ('\\') as an open delimiter.
Multiple Delimiter objects specified in delimiters may use the same close delimiter string.
split_delimiters doesn't react if the string ends with
unterminated delimiters.
See the Delimiter object for how delimiters are defined, and how
you can define your own delimiters.
split_quoted_strings(s, quotes=('"', "'"), *, escape='\\', multiline_quotes=(), state='')Splits s into quoted and unquoted segments.
Returns an iterator yielding 3-tuples:
(leading_quote, segment, trailing_quote)
where leading_quote and trailing_quote are either
empty strings or quote delimiters from quotes,
and segment is a substring of s. Joining together
all strings yielded recreates s.
s can be either str or bytes.
quotes is an iterable of unique quote delimiters.
Quote delimiters may be any non-empty string.
They must be the same type as s, either str or bytes.
By default, quotes is ('"', "'"). (If s is bytes,
quotes defaults to (b'"', b"'").) If a newline character
appears inside a quoted string, split_quoted_strings will
raise SyntaxError.
multiline_quotes is like quotes, except quoted strings
using multiline quotes are permitted to contain newlines.
By default split_quoted_strings doesn't define any
multiline quote marks.
escape is a string of any length. If escape is not
an empty string, the string will "escape" (quote)
quote delimiters inside a quoted string, like the
backslash ('\') character inside strings in Python.
By default, escape is '\\'. (If s is bytes,
escape defaults to b'\\'.)
state is a string. It sets the initial state of
the function. The default is an empty string (str
or bytes, matching s); this means the parser starts
parsing the string in an unquoted state. If you
want parsing to start as if it had already encountered
a quote delimiter--for example, if you were parsing
multiple lines individually, and you wanted to begin
a new line continuing the state from the previous line--
pass in the appropriate quote delimiter from quotes
into state. Note that when a non-empty string is
passed in to state, the leading_quote in the first
3-tuple yielded by split_quoted_strings will be an
empty string:
list(split_quoted_strings("a b c'", state="'"))
evaluates to
[('', 'a b c', "'")]
Note:
split_quoted_strings is agnostic about the length
of quoted strings. If you're using split_quoted_strings
to parse a C-like language, and you want to enforce
C's requirement that single-quoted strings only contain
one character, you'll have to do that yourself.split_quoted_strings doesn't raise an error
if s ends with an unterminated quoted string. In
that case, the last tuple yielded will have a non-empty
leading_quote and an empty trailing_quote. (If you
consider this an error, you'll need to raise SyntaxError
in your own code.)split_quoted_strings only supports the opening and
closing markers for a string being the same string.
If you need the opening and closing markers to be
different strings, use split_delimiters.split_text_with_code(s, *, tab_width=8, allow_code=True, code_indent=4, convert_tabs_to_spaces=True)Splits s into individual words,
suitable for feeding into
wrap_words.
s may be either str or bytes.
Paragraphs indented by less than code_indent will be
broken up into individual words.
If allow_code is true, paragraphs indented by at least
code_indent spaces will preserve their whitespace:
internal whitespace is preserved, and the newline is
preserved. (This will preserve the formatting of code
examples when these words are rejoined into lines by
wrap_words.)
For more information, see the deep-dive on Word wrapping and formatting.
split_title_case(s, *, split_allcaps=True)Splits s into words, assuming that
upper-case characters start new words.
Returns an iterator yielding the split words.
Example:
list(split_title_case('ThisIsATitleCaseString'))
is equal to
['This', 'Is', 'A', 'Title', 'Case', 'String']
If split_allcaps is a true value (the default),
runs of multiple uppercase characters will also
be split before the last character. This is
needed to handle splitting single-letter words.
Consider:
list(split_title_case('WhenIWasATeapot', split_allcaps=True))
returns
['When', 'I', 'Was', 'A', 'Teapot']
but
list(split_title_case('WhenIWasATeapot', split_allcaps=False))
returns
['When', 'IWas', 'ATeapot']
Note: uses the isupper and islower methods
to determine what are upper- and lower-case
characters. This means it only recognizes the ASCII
upper- and lower-case letters for bytes strings.
str_linebreaksA tuple of str objects, representing every line-breaking
whitespace character recognized by the Python str object.
Identical to linebreaks.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'. See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
str_linebreaks_without_crlfEquivalent to str_linebreaks without '\r\n'.
str_whitespaceA tuple of str objects, representing every whitespace
character recognized by the Python str object.
Identical to whitespace.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'. See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
str_whitespace_without_crlfEquivalent to str_whitespace without '\r\n'.
unicode_linebreaksA tuple of str objects, representing every line-breaking
whitespace character defined by Unicode.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'. See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
unicode_linebreaks_without_crlfEquivalent to unicode_linebreaks without '\r\n'.
unicode_whitespaceA tuple of str objects, representing every whitespace
character defined by Unicode.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'. See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
unicode_whitespace_without_crlfEquivalent to unicode_whitespace without '\r\n'.
whitespaceA tuple of str objects, representing every whitespace
character recognized by the Python str object.
Identical to str_whitespace.
Useful as a separator argument for big functions that accept one,
e.g. the big "multi-" family of functions,
or the lines and lines modifier functions.
Also contains '\r\n'. See the deep-dive section on
The Unix, Mac, and DOS linebreak conventions
for more.
For more information, please see the Whitespace and line-breaking characters in Python and big deep-dive.
whitespace_without_crlfEquivalent to whitespace without '\r\n'.
wrap_words(words, margin=79, *, two_spaces=True)Combines words into lines and returns the result as a string.
Similar to textwrap.wrap.
words should be an iterator yielding str or bytes strings,
and these strings should already be split at word boundaries.
Here's an example of a valid argument for words:
"this is an example of text split at word boundaries".split()
A single '\n' indicates a line break.
If you want a paragraph break, embed two '\n' characters in a row.
margin specifies the maximum length of each line. The length of
every line will be less than or equal to margin, unless the length
of an individual element inside words is greater than margin.
If two_spaces is true, elements from words that end in
sentence-ending punctuation ('.', '?', and '!')
will be followed by two spaces, not one.
Elements in words are not modified; any leading or trailing
whitespace will be preserved. You can use this to preserve
whitespace where necessary, like in code examples.
For more information, see the deep-dive on Word wrapping and formatting.
big.timeFunctions for working with time. Currently deals specifically with timestamps. The time functions in big are designed to make it easy to use best practices.
date_ensure_timezone(d, timezone)Ensures that a datetime.date object has a timezone set.
If d has a timezone set, returns d.
Otherwise, returns a new datetime.date
object equivalent to d with its tzinfo set
to timezone.
date_set_timezone(d, timezone)Returns a new datetime.date object identical
to d but with its tzinfo set to timezone.
datetime_ensure_timezone(d, timezone)Ensures that a datetime.datetime object has
a timezone set.
If d has a timezone set, returns d.
Otherwise, creates a new datetime.datetime
object equivalent to d with its tzinfo set
to timezone.
datetime_set_timezone(d, timezone)Returns a new datetime.datetime object identical
to d but with its tzinfo set to timezone.
parse_timestamp_3339Z(s, *, timezone=None)Parses a timestamp string returned by timestamp_3339Z.
Returns a datetime.datetime object.
timezone is an optional default timezone, and should
be a datetime.tzinfo object (or None). If provided,
and the time represented in the string doesn't specify
a timezone, the tzinfo attribute of the returned object
will be explicitly set to timezone.
parse_timestamp_3339Z depends on the
python-dateutil
package. If python-dateutil is unavailable,
parse_timestamp_3339Z will also be unavailable.
timestamp_3339Z(t=None, want_microseconds=None)Return a timestamp string in RFC 3339 format, in the UTC
time zone. This format is intended for computer-parsable
timestamps; for human-readable timestamps, use timestamp_human().
Example timestamp: '2021-05-25T06:46:35.425327Z'
t may be one of several types:
t is None, timestamp_3339Z uses the current time in UTC.t is an int or a float, it's interpreted as seconds
since the epoch in the UTC time zone.t is a time.struct_time object or datetime.datetime
object, and it's not in UTC, it's converted to UTC.
(Technically, time.struct_time objects are converted to GMT,
using time.gmtime. Sorry, pedants!)If want_microseconds is true, the timestamp ends with
microseconds, represented as a period and six digits between
the seconds and the 'Z'. If want_microseconds
is false, the timestamp will not include this text.
If want_microseconds is None (the default), the timestamp
ends with microseconds if the type of t can represent
fractional seconds: a float, a datetime object, or the
value None.
timestamp_human(t=None, want_microseconds=None)Return a timestamp string formatted in a pleasing way
using the currently-set local timezone. This format
is intended for human readability; for computer-parsable
time, use timestamp_3339Z().
Example timestamp: "2021/05/24 23:42:49.099437"
t can be one of several types:
t is None, timestamp_human uses the current local time.t is an int or float, it's interpreted as seconds since the epoch.t is a time.struct_time or datetime.datetime object,
it's converted to the local timezone.If want_microseconds is true, the timestamp will end with
the microseconds, represented as ".######". If want_microseconds
is false, the timestamp will not include the microseconds.
If want_microseconds is None (the default), the timestamp
ends with microseconds if the type of t can represent
fractional seconds: a float, a datetime object, or the
value None.
big.versionSupport for version metadata objects.
Version(s=None, *, epoch=None, release=None, release_level=None, serial=None, post=None, dev=None, local=None)Constructs a Version object, which represents a version number.
You may define the version one of two ways:
s positional parameter.
Example: Version("1.3.24rc37")Version(release=(1, 3, 24), release_level="rc", serial=37)big's Version objects conform to the PEP 440
version scheme, parsing version strings using that PEP's official regular
expression.
Version objects support the following features:
epochreleasemajor (release[0])minor (a safe version of release[1])micro (a safe version of release[2])release_levelserialpostdevlocalVersion objects are hashable.Version objects support ordering and comparison; you can ask if two Version
objects are equal, or if one is less than the other.str() on a Version object returns a normalized version string
for that version. repr() on a Version object returns a string that,
if eval'd, reconstructs that object.Version objects normalize themselves at initialization time:
release (and trailing .0 strings in
the equivalent part of a version string) are stripped.release_level are
normalized.sys.version_info
object or a packaging.Version object into the constructor
instead of a version string.When constructing a Version by passing in a string s, the string must conform to this scheme,
where square brackets denote optional substrings and names in angle brackets represent parameterized
substrings:
[<epoch>!]<major>(.<minor_etc>)*[<release_level>[<serial>]][.post<post>][.dev<dev>][+<local>]
All fields should be non-negative integers except for:
<major>(.<minor_etc>)* is meant to connote a conventional dotted version number, like 1.2 or 1.5.3.8.
This section can contain only numeric digits and periods ('.').
You may have as few or as many periods as you prefer. Trailing .0 entries will be stripped.<release_level> can only be be one of the following strings:
a, meaning an alpha release,b, meaning a beta release, orrc, meaning a release candidate.
For a final release, skip the release_level (and the serial).<local> represents an arbitrary sequence of alphanumeric characters punctuated by periods.Alternatively, you can construct a Version object by passing in these keyword-only arguments:
epoch
A non-negative int or None. Represents an "epoch" of version numbers. A version number
with a higher "epoch" is always a later release, regardless of all other fields.
release
A tuple containing one or more non-negative integers. Represents the conventional part
of the version number; the version string 1.3.8 would translate to Version(release=(1, 3, 8)).
release_level
A str or None. If it's a str, it must be one of the following strings:
a, meaning an alpha release,b, meaning a beta release, orrc, meaning a release candidate.serial
A non-negative int or None. Represents how many releases there have been at this release_level.
(The name is taken from Python's sys.version_info.)
post
A non-negative int or None. Represents "post-releases", extremely minor releases made after a release:
Version(release=(1, 3, 5)) < Version(release=(1, 3, 5), post=1)
dev
A non-negative int or None. Represents an under-development release. Higher dev numbers represent
later releases, but any release where dev is not None comes before any release where dev is None.
In other words:
Version(release=(1, 3, 5), dev=34) < Version(release=(1, 3, 5), dev=35)
Version(release=(1, 3, 5), dev=35) < Version(release=(1, 3, 5))
local
A tuple of one or more str objects containing only one or more
alphanumeric characters
or None. Represents a purely local version number,
allowing for minor build and patch differences
but with no API or ABI changes.
Version.format(s)Returns a formatted version of s, substituting attributes from
self into s using str.format_map.
For example,
Version("1.3.5").format('{major}.{minor}')
returns the string '1.3'.
multi- family of string functionsThis family of string functions was inspired by Python's str.split,
str.rsplit, and str.splitlines methods. These string splitting
methods are well-designed and often do what you want. But they're
surprisingly narrow and opinionated. What if your use case doesn't
map neatly to one of these functions? str.split supports two
very specific modes of operation--unless you want to split your
string in exactly one of those two modes, you probably can't use
str.split to solve your problem.
So what can you use? There's
re.split,
but that can be hard to use.1
Regular expressions can be so hard to get right, and the
semantics of re.split are subtly different from the usual
string splitting functions. Not to mention, it doesn't support
reverse!
Now there's a new answer:
multisplit.
The goal of multisplit is to be the be-all end-all string splitting function.
It's designed to supercede every mode of operation provided by
str.split, str.rsplit, and str.splitlines, and it
can even replace str.partition and str.rpartition too.
multisplit does it all!
The downside of multisplit's
awesome flexibility is that it can be hard to use... after all,
it takes five keyword-only parameters. However, these parameters
and their defaults are designed to be easy to remember.
The best way to cope with
multisplit's
complexity is to use it as a building block for your own
text splitting functions. For example, big uses
multisplit
to implement
multipartition,
normalize_whitespace,
lines,
and several other functions.
multisplitTo use
multisplit,
pass in the string you want to split, the separators you
want to split on, and tweak its behavior with its five
keyword arguments. It returns an iterator that yields
string segments from the original string in your preferred
format. The separator list is optional; if you don't
pass one in, it defaults to an iterable of whitespace separators
(either
big.whitespace
or
big.ascii_whitespace,
as appropriate).
The cornerstone of multisplit
is the separators argument.
This is an iterable of strings, of the same type (str or bytes)
as the string you want to split (s). multisplit will split
the string at each non-overlapping instance of any string
specified in separators.
multisplit
lets you fine-tune its behavior via five keyword-only
parameters:
keep lets you include the separator strings in the output,
in a number of different formats.separate lets you specify whether adjacent separator strings
should be grouped together (like str.split operating on
whitespace) or regarded as separate (like str.split when
you pass in an explicit separator).strip lets you strip separator strings from the beginning,
end, or both ends of the string you're splitting. It also
supports a special progressive mode that duplicates the
behavior of str.split when you use None as the separator.maxsplit lets you specify the maximum number of times to
split the string, exactly like the maxsplit argument to str.split.reverse makes multisplit behave like str.rsplit,
starting at the end of the string and working backwards.
(This only changes the behavior of multisplit if you use
maxsplit, or if your string contains overlapping separators.)To make it slightly easier to remember, all these keyword-only
parameters default to a false value. (Well, technically,
maxsplit defaults to the special value -1, for compatibility
with str.split. But that's its special "don't do anything"
magic value. All the other keyword-only parameters default
to False.)
multisplit
also inspired multistrip
and multipartition,
which also take this same separators arguments. There are also
other big functions that take a separators argument,
for example comment_markers for
lines_filter_line_comment_lines.)
multisplit keyword-only parameterTo give you a sense of how the five keyword-only parameters changes the behavior of
multisplit,
here's a breakdown of each of these parameters with examples.
maxsplitmaxsplit specifies the maximum number of times the string should be split.
It behaves the same as the maxsplit parameter to str.split.
The default value of -1 means "split as many times as you can". In our
example here, the string can be split a maximum of three times. Therefore,
specifying a maxsplit of -1 is equivalent to specifying a maxsplit of
2 or greater:
>>> list(big.multisplit('apple^banana_cookie', ('_', '^'))) # "maxsplit" defaults to -1
['apple', 'banana', 'cookie']
>>> list(big.multisplit('apple^banana_cookie', ('_', '^'), maxsplit=0))
['appleXbananaYcookie']
>>> list(big.multisplit('apple^banana_cookie', ('_', '^'), maxsplit=1))
['apple', 'bananaYcookie']
>>> list(big.multisplit('apple^banana_cookie', ('_', '^'), maxsplit=2))
['apple', 'banana', 'cookie']
>>> list(big.multisplit('apple^banana_cookie', ('_', '^'), maxsplit=3))
['apple', 'banana', 'cookie']
maxsplit has interactions with reverse and strip. For more
information, see the documentation regarding those parameters below.
keepkeep indicates whether or not multisplit should preserve the separator
strings in the strings it yields. It supports four values: false, true,
and the special values ALTERNATING and AS_PAIRS.
When keep is false, multisplit throws away the separator strings;
they won't appear in the output.
>>> list(big.multisplit('apple#banana-cookie', ('#', '-'))) # "keep" defaults to False
['apple', 'banana', 'cookie']
>>> list(big.multisplit('apple-banana#cookie', ('#', '-'), keep=False))
['apple', 'banana', 'cookie']
When keep is true, multisplit keeps the separators, appending them to
the end of the separated string:
>>> list(big.multisplit('apple$banana~cookie', ('$', '~'), keep=True))
['appleX', 'bananaY', 'cookie']
When keep is ALTERNATING, multisplit keeps the separators as separate
strings. The first string yielded is always a non-separator string, and
from then on it always alternates between a separator string and a non-separator
string. Put another way, if you store the output of multisplit in a list,
entries with an even-numbered index (0, 2, 4, ...) are always non-separator strings,
and entries with an odd-numbered index (1, 3, 5, ...) are always separator strings.
>>> list(big.multisplit('appleXbananaYcookie', ('X', 'Y'), keep=big.ALTERNATING))
['apple', 'X', 'banana', 'Y', 'cookie']
Note that ALTERNATING always emits an odd number of strings, and the first and
last strings yielded are always non-separator strings. Like str.split,
if the string you're splitting starts or ends with a separator string,
multisplit will emit an empty string at the beginning or end, to preserve
the "always begin and end with non-separator string" invariant:
>>> list(big.multisplit('1a1z1', ('1',), keep=big.ALTERNATING))
['', '1', 'a', '1', 'z', '1', '']
Finally, when keep is AS_PAIRS, multisplit keeps the separators as separate
strings. But it doesn't yield bare strings; instead, it yields 2-tuples of strings.
Every 2-tuple contains a non-separator string followed by a separator string.
If the original string starts with a separator, the first 2-tuple will contain an empty non-separator string and the separator:
>>> list(big.multisplit('^apple-banana^cookie', ('-', '^'), keep=big.AS_PAIRS))
[('', '^'), ('apple', '-'), ('banana', '^'), ('cookie', '')]
The last 2-tuple will always contain an empty separator string:
>>> list(big.multisplit('apple*banana+cookie', ('*', '+'), keep=big.AS_PAIRS))
[('apple', '*'), ('banana', '+'), ('cookie', '')]
>>> list(big.multisplit('apple*banana+cookie***', ('*', '+'), keep=big.AS_PAIRS, strip=True))
[('apple', '*'), ('banana', '+'), ('cookie', '')]
(This rule means that AS_PAIRS always emits an even number of strings.
Contrast that with ALTERNATING, which always emits an odd number of strings,
and the last string it emits is always a non-separator string. Put another
way: if you ignore the tuples, the list of strings emitted by AS_PAIRS is the
same as those emitted by ALTERNATING, except AS_PAIRS appends an empty
string.)
Because of this rule, if the original string ends with a separator,
and multisplit doesn't strip the right side, the final tuple
emitted by AS_PAIRS will be a 2-tuple containing two empty strings:
>>> list(big.multisplit('appleXbananaYcookieX', ('X', 'Y'), keep=big.AS_PAIRS))
[('apple', 'X'), ('banana', 'Y'), ('cookie', 'X'), ('', '')]
This looks strange and unnecessary. But it is what you want. This odd-looking behavior is discussed at length in the section below, titled Why do you sometimes get empty strings when you split?
The behavior of keep can be affected by the value of separate.
For more information, see the next section, on separate.
separateseparate indicates whether multisplit should consider adjacent
separator strings in s as one separator or as multiple separators
each separated by a zero-length string. It can be either false or
true.
>>> list(big.multisplit('apple=?banana?=?cookie', ('=', '?'))) # separate defaults to False
['apple', 'banana', 'cookie']
>>> list(big.multisplit('apple=?banana?=?cookie', ('=', '?'), separate=False))
['apple', 'banana', 'cookie']
>>> list(big.multisplit('apple=?banana?=?cookie', ('=', '?'), separate=True))
['apple', '', 'banana', '', '', 'cookie']
If separate and keep are both true values, and your string
has multiple adjacent separators, multisplit will view s
as having zero-length non-separator strings between the
adjacent separators:
>>> list(big.multisplit('appleXYbananaYXYcookie', ('X', 'Y'), separate=True, keep=True))
['appleX', 'Y', 'bananaY', 'X', 'Y', 'cookie']
>>> list(big.multisplit('appleXYbananaYXYcookie', ('X', 'Y'), separate=True, keep=big.AS_PAIRS))
[('apple', 'X'), ('', 'Y'), ('banana', 'Y'), ('', 'X'), ('', 'Y'), ('cookie', '')]
stripstrip indicates whether multisplit should strip separators from
the beginning and/or end of s. It supports five values:
false, true, big.LEFT, big.RIGHT, and big.PROGRESSIVE.
By default, strip is false, which means it doesn't strip any
leading or trailing separators:
>>> list(big.multisplit('%|apple%banana|cookie|%|', ('%', '|'))) # strip defaults to False
['', 'apple', 'banana', 'cookie', '']
Setting strip to true strips both leading and trailing separators:
>>> list(big.multisplit('%|apple%banana|cookie|%|', ('%', '|'), strip=True))
['apple', 'banana', 'cookie']
big.LEFT and big.RIGHT tell multistrip to only strip on that
side of the string:
>>> list(big.multisplit('.?apple.banana?cookie.?.', ('.', '?'), strip=big.LEFT))
['apple', 'banana', 'cookie', '']
>>> list(big.multisplit('.?apple.banana?cookie.?.', ('.', '?'), strip=big.RIGHT))
['', 'apple', 'banana', 'cookie']
big.PROGRESSIVE duplicates a specific behavior of str.split when using
maxsplit. It always strips on the left, but it only strips on the right
if the string is completely split. If maxsplit is reached before the entire
string is split, and strip is big.PROGRESSIVE, multisplit won't strip
the right side of the string. Note in this example how the trailing separator
Y isn't stripped from the input string when maxsplit is less than 3.
>>> list(big.multisplit('^apple^banana_cookie_', ('^', '_'), strip=big.PROGRESSIVE))
['apple', 'banana', 'cookie']
>>> list(big.multisplit('^apple^banana_cookie_', ('^', '_'), maxsplit=0, strip=big.PROGRESSIVE))
['apple^banana_cookie_']
>>> list(big.multisplit('^apple^banana_cookie_', ('^', '_'), maxsplit=1, strip=big.PROGRESSIVE))
['apple', 'banana_cookie_']
>>> list(big.multisplit('^apple^banana_cookie_', ('^', '_'), maxsplit=2, strip=big.PROGRESSIVE))
['apple', 'banana', 'cookie_']
>>> list(big.multisplit('^apple^banana_cookie_', ('^', '_'), maxsplit=3, strip=big.PROGRESSIVE))
['apple', 'banana', 'cookie']
>>> list(big.multisplit('^apple^banana_cookie_', ('^', '_'), maxsplit=4, strip=big.PROGRESSIVE))
['apple', 'banana', 'cookie']
reversereverse specifies where multisplit starts parsing the string--from
the beginning, or the end--and in what direction it moves when parsing
the string--towards the end, or towards the beginning_ It only supports
two values: when it's false, multisplit starts at the beginning of the
string, and parses moving to the right (towards the end of the string).
But when reverse is true, multisplit starts at the end of the
string, and parses moving to the left (towards the beginning
of the string).
This has two noticable effects on multisplit's output. First, this
changes which splits are kept when maxsplit is less than the total number
of splits in the string. When reverse is true, the splits are counted
starting on the right and moving towards the left:
>>> list(big.multisplit('apple-banana|cookie', ('-', '|'), reverse=True)) # maxsplit defaults to -1
['apple', 'banana', 'cookie']
>>> list(big.multisplit('apple-banana|cookie', ('-', '|'), maxsplit=0, reverse=True))
['apple-banana|cookie']
>>> list(big.multisplit('apple-banana|cookie', ('-', '|'), maxsplit=1, reverse=True))
['apple-banana', 'cookie']
>>> list(big.multisplit('apple-banana|cookie', ('-', '|'), maxsplit=2, reverse=True))
['apple', 'banana', 'cookie']
>>> list(big.multisplit('apple-banana|cookie', ('-', '|'), maxsplit=3, reverse=True))
['apple', 'banana', 'cookie']
The second effect is far more subtle. It's only relevant when splitting strings
containing multiple overlapping separators. When reverse is false, and there
are two (or more) overlapping separators, the string is split by the leftmost
overlapping separator. When reverse is true, and there are two (or more)
overlapping separators, the string is split by the rightmost overlapping
separator.
Consider these two calls to multisplit. The only difference between them is
the value of reverse. They produce different results, even though neither
one uses maxsplit.
>>> list(big.multisplit('appleXYZbananaXYZcookie', ('XY', 'YZ'))) # reverse defaults to False
['apple', 'Zbanana', 'Zcookie']
>>> list(big.multisplit('appleXYZbananaXYZcookie', ('XY', 'YZ'), reverse=True))
['appleX', 'bananaX', 'cookie']
multisplitHere are some examples of how you could use
multisplit
to replace some common Python string splitting methods. These exactly duplicate the
behavior of the originals.
def _multisplit_to_split(s, sep, maxsplit, reverse):
separate = sep != None
if separate:
strip = False
else:
sep = big.ascii_whitespace if isinstance(s, bytes) else big.whitespace
strip = big.PROGRESSIVE
result = list(big.multisplit(s, sep,
maxsplit=maxsplit, reverse=reverse,
separate=separate, strip=strip))
if not separate:
# ''.split() == ' '.split() == []
if result and (not result[-1]):
result.pop()
return result
def str_split(s, sep=None, maxsplit=-1):
return _multisplit_to_split(s, sep, maxsplit, False)
def str_rsplit(s, sep=None, maxsplit=-1):
return _multisplit_to_split(s, sep, maxsplit, True)
def str_splitlines(s, keepends=False):
linebreaks = big.ascii_linebreaks if isinstance(s, bytes) else big.linebreaks
l = list(big.multisplit(s, linebreaks,
keep=keepends, separate=True, strip=False))
if l and not l[-1]:
# yes, ''.splitlines() returns an empty list
l.pop()
return l
def _partition_to_multisplit(s, sep, reverse):
if not sep:
raise ValueError("empty separator")
l = tuple(big.multisplit(s, (sep,),
keep=big.ALTERNATING, maxsplit=1, reverse=reverse, separate=True))
if len(l) == 1:
empty = b'' if isinstance(s, bytes) else ''
if reverse:
l = (empty, empty) + l
else:
l = l + (empty, empty)
return l
def str_partition(s, sep):
return _partition_to_multisplit(s, sep, False)
def str_rpartition(s, sep):
return _partition_to_multisplit(s, sep, True)
You wouldn't want to use these, of course--Python's built-in functions are so much faster!
Sometimes when you split using
multisplit,
you'll get empty strings in the return value. This might be unexpected,
violating the Principle Of Least Astonishment.
But there are excellent reasons for this behavior.
Let's start by observing what str.split does. str.split really has two
major modes of operation: when you don't pass in a separator (or pass in None for the
separator), and when you pass in an explicit separator string. In this latter mode,
the documentation says it regards every instance of a separator string as an individual
separator splitting the string. What does that mean? Watch what happens when you have
two adjacent separators in the string you're splitting:
>>> '1,2,,3'.split(',')
['1', '2', '', '3']
What's that empty string doing between '2' and '3'? Here's how you should think about it:
when you pass in an explicit separator, str.split splits at every occurance of that
separator in the string. It always splits the string into two places, whenever there's
a separator. And when there are two adjacent separators, conceptually, they have a
zero-length string in between them:
>>> '1,2,,3'[4:4]
''
The empty string in the output of str.split represents the fact that there
were two adjacent separators. If str.split didn't add that empty string,
the output would look like this:
['1', '2', '3']
But then it'd be indistinguishable from splitting the same string without two separators in a row:
>>> '1,2,3'.split(',')
['1', '2', '3']
This difference is crucial when you want to reconstruct the original string from
the split list. str.split with a separator should always be reversable using
str.join, and with that empty string there it works correctly:
>>> ','.join(['1', '2', '3'])
'1,2,3'
>>> ','.join(['1', '2', '', '3'])
'1,2,,3'
Now take a look at what happens when the string you're splitting starts or ends with a separator:
>>> ',1,2,3,'.split(',')
['', '1', '2', '3', '']
This might seem weird. But, just like with two adjacent separators,
this behavior is important for consistency. Conceptually there's
a zero-length string between the beginning of the string and the first
comma. And str.join needs those empty strings in order to correctly
recreate the original string.
>>> ','.join(['', '1', '2', '3', ''])
',1,2,3,'
Naturally,
multisplit
lets you duplicate this behavior. When you want
multisplit
to behave just like str.split does with an explicit separator
string, just pass in keep=False, separate=True, and strip=False.
That is, if a and b are strings,
big.multisplit(a, (b,), keep=False, separate=True, strip=False)
always produces the same output as
a.split(b)
For example, here's
multisplit
splitting the strings we've been playing with, using these parameters:
>>> list(big.multisplit('1,2,,3', (',',), keep=False, separate=True, strip=False))
['1', '2', '', '3']
>>> list(big.multisplit(',1,2,3,', (',',), keep=False, separate=True, strip=False))
['', '1', '2', '3', '']
This "emit an empty string" behavior also has ramifications when keep isn't false.
The behavior of keep=True is easy to predict; multisplit just appends the separators
to the previous string segment:
>>> list(big.multisplit('1,2,,3', (',',), keep=True, separate=True, strip=False))
['1,', '2,', ',', '3']
>>> list(big.multisplit(',1,2,3,', (',',), keep=True, separate=True, strip=False))
[',', '1,', '2,', '3,', '']
The principle here is that, when you use keep=True, you should be able to reconstitute
the original string with ''.join:
>>> ''.join(['1,', '2,', ',', '3'])
'1,2,,3'
>>> ''.join([',', '1,', '2,', '3,', ''])
',1,2,3,'
keep=big.ALTERNATING is much the same, except we insert the separators as their
own segments, rather than appending each one to the previous segment:
>>> list(big.multisplit('1,2,,3', (',',), keep=big.ALTERNATING, separate=True, strip=False))
['1', ',', '2', ',', '', ',', '3']
>>> list(big.multisplit(',1,2,3,', (',',), keep=big.ALTERNATING, separate=True, strip=False))
['', ',', '1', ',', '2', ',', '3', ',', '']
Remember, ALTERNATING output always begins and ends with a non-separator string.
If the string you're splitting begins or ends with a separator, the output
from multisplit specifying keep=ALTERNATING will correspondingly begin or end
with an empty string.
And, as with keep=True, you can also recreate the original string by passing
these arrays in to ''.join:
>>> ''.join(['1', ',', '2', ',', '', ',', '3'])
'1,2,,3'
>>> ''.join(['', ',', '1', ',', '2', ',', '3', ',', ''])
',1,2,3,'
Finally there's keep=big.AS_PAIRS. The behavior here seemed so strange,
initially I thought it was wrong. But I've given it a lot of thought, and
I've convinced myself that this is correct:
>>> list(big.multisplit('1,2,,3', (',',), keep=big.AS_PAIRS, separate=True, strip=False))
[('1', ','), ('2', ','), ('', ','), ('3', '')]
>>> list(big.multisplit(',1,2,3,', (',',), keep=big.AS_PAIRS, separate=True, strip=False))
[('', ','), ('1', ','), ('2', ','), ('3', ','), ('', '')]
That tuple at the end, just containing two empty strings:
('', '')
It's so strange. How can that be right?
In short, it's similar to the str.split situation.
When called with keep=AS_PAIRS,
multisplit
guarantees that the final tuple will contain an empty separator string.
If the string you're splitting ends with a separator, it must emit
the empty non-separator string, followed by the empty separator string.
Think of it this way: with the tuple of empty strings there, you can easily
convert one keep format into any another. (Provided that you know
what the separators were--either the source keep format was not false,
or you only used one separator string when calling multisplit).
Without that tuple of empty strings at the end, you'd also have to have an
if statement to add or remove empty stuff from the end.
I'll demonstrate this with a simple example. Here's the output of
multisplit splitting the string '1a1z1' by the separator '1',
in each of the four keep formats:
>>> list(big.multisplit('1a1z1', '1', keep=False))
['', 'a', 'z', '']
>>> list(big.multisplit('1a1z1', '1', keep=True))
['1', 'a1', 'z1', '']
>>> list(big.multisplit('1a1z1', '1', keep=big.ALTERNATING))
['', '1', 'a', '1', 'z', '1', '']
>>> list(big.multisplit('1a1z1', '1', keep=big.AS_PAIRS))
[('', '1'), ('a', '1'), ('z', '1'), ('', '')]
Because the AS_PAIRS output ends with that tuple of empty
strings, we can mechanically convert it into any of the other
formats, like so:
>>> result = list(big.multisplit('1a1z1', '1', keep=big.AS_PAIRS))
>>> result
[('', '1'), ('a', '1'), ('z', '1'), ('', '')]
>>> [s[0] for s in result] # convert to keep=False
['', 'a', 'z', '']
>>> [s[0]+s[1] for s in result] # convert to keep=True
['1', 'a1', 'z1', '']
>>> [s for t in result for s in t][:-1] # convert to keep=big.ALTERNATING
['', '1', 'a', '1', 'z', '1', '']
If the AS_PAIRS output didn't end with that tuple of empty strings,
you'd need to add an if statement to restore the trailing empty
strings as needed.
str.split returns an empty list when you split an
empty string by whitespace:
>>> ''.split()
[]
But not when you split by an explicit separator:
>>> ''.split('x')
['']
multisplit
is consistent here. If you split an empty string, it always returns an empty string,
as long as the separators are valid:
>>> list(big.multisplit(''))
['']
>>> list(big.multisplit('', ('a', 'b', 'c')))
['']
Similarly, when splitting a string that only contains whitespace, str.split also
returns an empty list:
>>> ' '.split()
[]
This is really the same as "splitting an empty string", because when str.split
splits on whitespace, the first thing it does is strip leading whitespace.
If you multisplit
a string that only contains whitespace, and you split on whitespace characters,
it returns two empty strings:
>>> list(big.multisplit(' '))
['', '']
This is because the string conceptually starts with a zero-length string,
then has a run of whitespace characters, then ends with another zero-length
string. So those two empty strings are the leading and trailing zero-length
strings, separated by whitespace. If you tell
multisplit
to also strip the string, you'll get back a single empty string:
>>> list(big.multisplit(' ', strip=True))
['']
And
multisplit
behaves consistently even when you use different separators:
>>> list(big.multisplit('ababa', 'ab'))
['', '']
>>> list(big.multisplit('ababa', 'ab', strip=True))
['']
And I should know--multisplit is implemented using re.split!
Several functions in big take a separators
argument, an iterable of separator strings.
Examples of these functions include
lines
and
multisplit.
Although you can use any iterable of strings
you like, most often you'll be separating on some
form of whitespace. But what, exactly, is
whitespace? There's more to this topic than you
might suspect.
The good news is, you can almost certainly ignore all the complexity. These days the only whitespace characters you're likely to encounter are spaces, tabs, newlines, and maybe carriage returns. Python and big handle all those easily.
With respect to big and these separators arguments,
big provides four values designed for use as separators.
All four of these are tuples containing whitespace characters:
str objects, you'll want to use either
big.whitespace or big.linebreaks.
big.whitespace contains all the whitespace characters,
big.linebreaks contains just the line-breaking
whitespace characters.bytes
objects: bytes_whitespace
and bytes_linebreaks,
respectively.Apart from exceptionally rare occasions, these are all you'll ever need. And if that's all you need, you can stop reading this section now.
But what about those exceptionally rare occasions? You'll be pleased to know big handles them too. The rest of this section is a deep dive into these rare occasions.
Here's the list of all characters recognized by
Python str objects as whitespace characters:
# char decimal hex name
##########################################
'\t' , # 9 - 0x0009 - tab
'\n' , # 10 - 0x000a - newline
'\v' , # 11 - 0x000b - vertical tab
'\f' , # 12 - 0x000c - form feed
'\r' , # 13 - 0x000d - carriage return
'\x1c' , # 28 - 0x001c - file separator
'\x1d' , # 29 - 0x001d - group separator
'\x1e' , # 30 - 0x001e - record separator
'\x1f' , # 31 - 0x001f - unit separator
' ' , # 32 - 0x0020 - space
'\x85' , # 133 - 0x0085 - next line
'\xa0' , # 160 - 0x00a0 - non-breaking space
'\u1680', # 5760 - 0x1680 - ogham space mark
'\u2000', # 8192 - 0x2000 - en quad
'\u2001', # 8193 - 0x2001 - em quad
'\u2002', # 8194 - 0x2002 - en space
'\u2003', # 8195 - 0x2003 - em space
'\u2004', # 8196 - 0x2004 - three-per-em space
'\u2005', # 8197 - 0x2005 - four-per-em space
'\u2006', # 8198 - 0x2006 - six-per-em space
'\u2007', # 8199 - 0x2007 - figure space
'\u2008', # 8200 - 0x2008 - punctuation space
'\u2009', # 8201 - 0x2009 - thin space
'\u200a', # 8202 - 0x200a - hair space
'\u2028', # 8232 - 0x2028 - line separator
'\u2029', # 8233 - 0x2029 - paragraph separator
'\u202f', # 8239 - 0x202f - narrow no-break space
'\u205f', # 8287 - 0x205f - medium mathematical space
'\u3000', # 12288 - 0x3000 - ideographic space
This list was derived by iterating over every character
defined in Unicode, and testing to see if the split()
method on a Python str object splits at that character.
The first surprise: this isn't the same as the list of
all characters defined by Unicode as whitespace.
It's almost the same list, except Python adds four extra
characters: '\x1c', '\x1d', '\x1e', and '\x1f',
which respectively are called "file separator", "group separator",
"record separator", and "unit separator".
I'll refer to these as "the four ASCII separator characters".
These characters were defined as part of the original ASCII standard, way back in 1963. As their names suggest, they were intended to be used as separator characters for data, the same way Ctrl-Z was used to indicate end-of-file in the CPM and earliest FAT filesystems. But the four ASCII separator characters were rarely used even back in the day. Today they're practically unheard of.
As a rule, printing these characters to the screen generally doesn't do anything--they don't move the cursor, and the screen doesn't change. So their behavior is a bit mysterious. A lot of people (including early Python programmers it seems!) thought that meant they're whitespace. This seems like an odd conclusion to me. After all, all the other whitespace characters move the cursor, either right or down or both; these don't move the cursor at all.
The Unicode standard is unambiguous: these characters are not whitespace. And yet Python's "Unicode object" behaves as if they are. So I'd say this is a bug; Python's Unicode object should implement what the Unicode standard says.
It seems that the C library used by GCC and clang on my workstation agree. I wrote a quick C program to print out what characters are and aren't whitespace, according to the C function isspace(). It seems the C library agrees with Unicode: it doesn't consider the four ASCII separator characters to be whitespace.
Here's the program, in case you want to try it yourself.
#include <stdio.h> #include <ctype.h> int main(int c, char *a[]) { int i; printf("\nisspace table.\nAdd the row and column numbers together (in hex).\n\n"); printf(" | 0 1 2 3 4 5 6 7 8 9 a b c d e f\n"); printf("-----+--------------------------------\n"); for (i = 0 ; i < 256 ; i++) { char *message = isspace(i) ? "Y" : "n"; if ((i % 16) == 0) printf("0x%02x |", i); printf(" %s", message); if ((i % 16) == 15) printf("\n"); } return 0; }Here's its output on my workstation:
isspace table. Add the row and column numbers together (in hex). | 0 1 2 3 4 5 6 7 8 9 a b c d e f -----+-------------------------------- 0x00 | n n n n n n n n n Y Y Y Y Y n n 0x10 | n n n n n n n n n n n n n n n n 0x20 | Y n n n n n n n n n n n n n n n 0x30 | n n n n n n n n n n n n n n n n 0x40 | n n n n n n n n n n n n n n n n 0x50 | n n n n n n n n n n n n n n n n 0x60 | n n n n n n n n n n n n n n n n 0x70 | n n n n n n n n n n n n n n n n 0x80 | n n n n n n n n n n n n n n n n 0x90 | n n n n n n n n n n n n n n n n 0xa0 | n n n n n n n n n n n n n n n n 0xb0 | n n n n n n n n n n n n n n n n 0xc0 | n n n n n n n n n n n n n n n n 0xd0 | n n n n n n n n n n n n n n n n 0xe0 | n n n n n n n n n n n n n n n n 0xf0 | n n n n n n n n n n n n n n n n0x1c through 0x1f are represented by the last four
ncharacters on the second line, the0x10line. The fact that they'rens tells you that this C standard library doesn't consider those characters to be whitespace.
Like many bugs, this one has lingered for a long time. The behavior is present in Python 2, there's a ten-year-old issue on the Python issue tracker about this, and it's not making progress.
The second surprise has to do with bytes objects.
Of course, bytes objects represent binary data, and don't
necessarily represent characters. Even if they do, they don't
have any encoding associated with them. However, for
convenience--and backwards-compatibility with Python 2--Python's
bytes objects support several method calls that treat the data
as if it were "ASCII-compatible".
The surprise: These methods on Python bytes objects recognize
a different set of whitespace characters. Here's the list of
all bytes recognized by Python bytes objects as whitespace:
# char decimal hex name
#######################################
'\t' , # 9 - 0x09 - tab
'\n' , # 10 - 0x0a - newline
'\v' , # 11 - 0x0b - vertical tab
'\f' , # 12 - 0x0c - form feed
'\r' , # 13 - 0x0d - carriage return
' ' , # 32 - 0x20 - space
This list was derived by iterating over every possible
byte value, and testing to see if the split() method
on a Python bytes object splits at that byte.
The good news is, this list is the same as ASCII's list,
and it agrees with Unicode.
In fact this list is quite familiar to C programmers;
it's the same whitespace characters recognized by the
standard C function
isspace() (in ctypes.h).
Python has used this function to decide which characters
are and aren't whitespace in 8-bit strings since its very
beginning.
Notice that this list doesn't contain the four ASCII separator characters. That these two types in Python don't agree only enhances the mystery.
The situation is slightly worse with line-breaking
characters. Line-breaking characters (aka linebreaks)
are a subset of whitespace characters; they're whitespace
characters that always move the cursor down to the next
line. And, as with whitespace generally, Python str
objects don't agree with Unicode about what is and is
not a line-breaking character, and Python bytes objects
don't agree with either of those.
Here's the list of all Unicode characters recognized by
Python str objects as line-breaking characters:
# char decimal hex name
##########################################
'\n' , # 10 0x000a - newline
'\v' , # 11 0x000b - vertical tab
'\f' , # 12 0x000c - form feed
'\r' , # 13 0x000d - carriage return
'\x1c' , # 28 0x001c - file separator
'\x1d' , # 29 0x001d - group separator
'\x1e' , # 30 0x001e - record separator
'\x85' , # 133 0x0085 - next line
'\u2028', # 8232 0x2028 - line separator
'\u2029', # 8233 0x2029 - paragraph separator
This list was derived by iterating over every character
defined in Unicode, and testing to see if the splitlines()
method on a Python str object splits at that character.
Again, this is different from the list of characters defined as line-breaking whitespace in Unicode. And again it's because Python defines some of the four ASCII separator characters as line-breaking characters. In this case it's only the first three; Python doesn't consider the fourth, "unit separator", as a line-breaking character. (I don't know why Python draws this distinction... but then again, I don't know why it considers the first three to be line-breaking. It's all a mystery to me.)
Here's the list of all characters recognized by
Python bytes objects as line-breaking characters:
# char decimal hex name
#######################################
'\n' , # 10 0x000a - newline
'\r' , # 13 0x000d - carriage return
This list was derived by iterating over every possible
byte, and testing to see if the splitlines()
method on a Python bytes object splits at that byte.
It's here we find our final unpleasant surprise:
the methods on Python bytes objects don't consider
'\v' (vertical tab)
and
'\f' (form feed)
to be line-break characters. I assert this is also a bug.
These are well understood to be line-breaking characters;
"vertical tab" is like a "tab", except it moves the cursor
down instead of to the right. And "form feed" moves the
cursor to the top left of the next "page", which requires
advancing at least one line.
To be crystal clear: the odds that any of this will cause a problem for you are extremely low. In order for it to make a difference:
It seems extremely unlikely that all of these will be true for you.
In case this does affect you, big has a complete set of predefined whitespace tuples that will handle any of these situations. big defines a total of ten tuples, sorted into five categories.
In every category there are two values: one that contains
whitespace, the other contains linebreaks. The
whitespace tuple contains all the possible values of
whitespace--characters that move the cursor either
horizontally, or vertically, or both, but don't
print anything visible to the screen. The linebreaks
tuple contains the subset of whitespace characters that
move the cursor vertically.
The most important two values start with str_:
str_whitespace
and
str_linebreaks.
These contain all the whitespace characters
recognized by the Python str object.
Next are two values that start with unicode_:
unicode_whitespace
and
unicode_linebreaks.
These contain all the whitespace characters
defined in the Unicode standard. They're the
same as the str_ tuples except we remove the
four ASCII separator characters.
Third, two values that start with ascii_:
ascii_whitespace
and
ascii_linebreaks.
These contain all the whitespace characters
defined in ASCII. (Note that these contain
str objects, not bytes objects.) They're
the same as the unicode_ tuples, except we
throw away all characters with a code point
higher than 127.
Fourth, two values that start with bytes_:
bytes_whitespace
and
bytes_linebreaks.
These contain all the whitespace characters
recognized by the Python bytes object.
These tuples contain bytes objects, encoded
using the ascii encoding. The list of
characters is distinct from the other sets
of tuples, and was derived as described above.
Finally we have the two tuples that lack a prefix:
whitespace
and
linebreaks.
These are the tuples you should use most of the time,
and several big functions use them as default values.
These are simply copies of str_whitespace and
str_linebreaks respectively.
(big actually defines an additional ten tuples, as discussed in the very next section.)
Historically, different platforms used different ASCII characters--or sequences of ASCII characters--to represent "go to the next line" in text files. Here are the most popular conventions:
\n - UNIX, Amiga, macOS 10+
\r - macOS 9 and earlier, many 8-bit computers
\r\n - Windows, DOS
(There are a couple more conventions, and a lot more history, in the Wikipedia article on newlines.)
Handling these differing conventions was a real mess, for a long time--not just for computer programmers, but in the daily lives of many computer users. It was a continual problem for software developers back in the 90s, particularly those who frequently switched back and forth between the two platforms. And it took a long time before software development tooling figured out how to seamlessly handle all the newline conventions.
Python itself went through several iterations on how to handle this, eventually implementing "universal newlines" support, added way back in Python 2.3.
These days the world seems to have converged on the UNIX
standard, '\n'; Windows supports it, and it's the default
on every other modern platform. So in practice these days
you probably don't have end-of-line conversion problems;
as long as you're decoding files to Unicode, and you don't
disable "universal newlines", it probably all works fine
and you never even noticed.
However! big strives to behave identically to Python in every way. And even today, Python considers the DOS linebreak sequence to be one linebreak, not two.
The Python splitlines method on a string splits the
string at linebreaks. And if the keepends positional
parameter is True, it appends the linebreak character(s)
at the end of each substring. A quick experiment with
splitlines will show us what Python thinks is and
isn't a linebreak. Sure enough, splitlines considers
'\n\r' to be two linebreaks, but it treats \r\n as a
single linebreak:
' a \n b \r c \r\n d \n\r e '.splitlines(True)
produces
[' a \n', ' b \r', ' c \r\n', ' d \n', '\r', ' e ']
Naturally, if you use big to split by lines, you get the same result:
list(big.multisplit(' a \n b \r c \r\n d \n\r e ', big.linebreaks, separate=True, keep=True))
How do we achieve this? big has one more trick. All of
the tuples defined in the previous section--from whitespace
to ascii_linebreaks--also contain the DOS linebreak
convention:
'\r\n'
(The equivalent bytes_ tuples contain the bytes equivalent,
b'\r\n.)
Because of this inclusion, when you use one of these tuples
with one of the big functions that take separators,
it'll recognize \r\n as if it was one whitespace "character".
(Just in case one happens to creep into your data.) And since
functions like multisplit are "greedy", preferring the longest
matching separator, if the string you're splitting contains
'\r\n', it'll prefer matching '\r\n' to just '\r'.
If you don't want this behavior, just add the suffix
_without_crlf to the end of any of the ten tuples,
e.g. whitespace_without_crlf, bytes_linebreaks_without_crlf.
What if you need to split text by whitespace, or by lines,
but that text is in bytes format with an unusual encoding?
big makes that easy too. If one of the builtin tuples
won't work for you, you can can make your own tuple from scratch,
or modify an existing tuple to meet your needs.
For example, let's say you need to split a document by whitespace, and the document is encoded in code page 850 or code page 437. (These two code pages are the most common code pages in English-speaking countries.)
Normally the easiest thing would be to decode it a str object
using the 'cp850' or 'cp437' text codec as appropriate,
then operate on it normally.
But you might have reasons why you don't want to decode it--maybe
the document is damaged and doesn't decode properly, and it's
easier to work with the encoded bytes than to fix it. If you
want to process the text with a big function that accepts a
separator argument, you could make your own custom tuples
of whitespace characters. These two codepages have the same
whitespace characters as ASCII, but they both add one more:
value 255, "non-breaking space", a space character that is
not line-breaking. (The intention is, this character should
behave like a space, except you shouldn't break a line at this
character when word wrapping.)
It's easy to make the appropriate tuples yourself:
cp437_linebreaks = cp850_linebreaks = big.bytes_linebreaks
cp437_whitespace = cp850_whitespace = big.bytes_whitespace + (b'\xff',)
Those tuples would work fine as the separators argument for
any big function that takes one.
What if you want to process a bytes object containing
UTF-8? That's easy too. Just convert one of the existing
tuples containing str objects using
big.encode_strings.
For example, to split a UTF-8 encoded bytes object b using
the Unicode line-breaking characters, you could call:
multisplit(b, encode_strings(unicode_linebreaks, encoding='utf-8'))
Note that this technique probably won't work correctly for most other
multibyte encodings, for example UTF-16.
For these encodings, you should decode to str before processing.
Why? It's because multisplit could find matches in multibyte
sequences straddling characters. Consider this example:
>>> haystack = '\u0101\u0102'
>>> needle = '\u0201'
>>> needle in haystack
False
>>>
>>> encoded_haystack = haystack.encode('utf-16-le')
>>> encoded_needle = needle.encode('utf-16-le')
>>> encoded_needle in encoded_haystack
True
The character '\u0201' doesn't appear in the original string,
but the encoded version appears in the encoded string,
as the second byte of the first character and the first
byte of the second character:
>>> encoded_haystack
b'\x01\x01\x02\x01'
>>> encoded_needle
b'\x01\x02'
But you can avoid this problem if you know you're working in bytes on two-byte sequences. Split the bytes string into two-byte segments and operate on those.
lines and lines modifier functionslines
is a function that makes it easy to write well-behaved, feature-rich text parsers.
lines itself
iterates over a string, returning an iterator that yields individual lines
split from that string. The iterator yields a 2-tuple:
(LinesInfo, line)
The LinesInfo object provides the line number and starting column number
for each line. This makes it easy for your parser to provide
line and column information for error messages.
This iterator is designed to be modified by "lines modifier"
functions. These are functions that consume a lines
iterator and re-yield the values, possibly modifying or
discarding them along the way. For example, passing
a lines iterator into lines_filter_empty_lines results
in an iterator that skips over the empty lines.
All the lines modifier functions that ship with big
start with the string lines_.
Most lines modifier function names belong to a category,
encoded as the second word in the function name
(immediately after lines_). Some examples:
lines_filter_ functions conditionally remove
lines from the output. For example, lines_filter_empty_lines
will only yield a line if it isn't empty.lines_strip_ functions may remove one or
more substrings from the line. For example,
lines_strip_indent
strips the leading whitespace from a line before yielding
it. (Whenever a lines modifier removes leading text from a line,
it will add a leading field to the accompanying
LineInfo
object containing the removed substring, and will also update the
column_number of the line to reflect the new starting column.)lines_convert_ functions means this lines modifier may change one
or more substrings in the line. For example,
lines_convert_tabs_to_spaces changes tab characters
to space characters in any lines it processes.(big isn't strictly consistent about these category names though.
For example,
lines_containing
and
lines_grep
are obviously "filter" modifiers, but their names
don't start with lines_filter_.)
All lines modifier functions are composable with each other; you can "stack" them together simply by passing the output of one into the input of another. For example,
with open("textfile.txt", "rt") as f:
for info, line in big.lines_filter_empty_lines(
big.lines_rstrip(lines(f.read()))):
...
will iterate over the lines of textfile.txt, skipping
over all empty lines and lines that consist only of
whitespace.
When you stack line modifiers in this way, note that the outer modifiers happen later. In the above example, each line is first "r-stripped", and then discarded if it's empty. If you stacked the line modifiers in the opposite order:
with open("textfile.txt", "rt") as f:
for info, line in big.lines_rstrip(
big.lines_filter_empty_lines(lines(f.read()))):
...
then it'd filter out empty lines first, and then "r-strip" the lines. So lines in the input that contained only whitespace would still get yielded as empty lines, which is probably not what you want. Ordering is important!
It's probably clearer to constructed nested lines modifiers this way:
with open("textfile.txt", "rt") as f:
li = lines(f.read())
li = big.lines_filter_empty_lines(li)
li = big.lines_rstrip(li)
for info, line in li:
...
This is much easier to read, particularly when one or more lines modifiers take additional arguments.
Of course, you can write your own lines modifier functions.
Simply accept a lines iterator as an argument, iterate over
it, and yield each line info and line--modifying them
(or not yielding them!) as you see fit. You could
even write your own lines iterator, a replacement for
lines,
if you need functionality
lines doesn't provide.
Note that if you write your own lines modifier function,
and it removes text from the beginning the line, you must
update column_number in the
LineInfo
object manually--it
doesn't happen automatically. The easiest way to handle this
is also the best way: whenever clipping text from the beginning
or end of the line, use the
clip_leading
and
clip_trailing
methods on the LineInfo object.
Speaking of best practices for lines modifier functions,
it's also best practice to modify the existing
LineInfo object that was yielded to you, rather than
throwing it away, creating a new one, and yielding that
instead. Previous lines modifier iterators may have added
fields to the
LineInfo
that you'd to preserve.
Generally speaking,
LineInfo
objects obey an invariant.
For any (info, line) pair yielded by lines or a lines
modifier:
info.leading + line + info.trailing + info.end == info.line
That is, you can recreate the original line by concatenating the "leading" string, the modified line, the "trailing" string, and the "end" string.
Of course, this won't be true if you use lines modifiers that
replace characters in the line. For example, lines_convert_tabs_to_spaces
replaces tab characters with one or more space characters.
If the original line contains tabs, obviously the above invariant
will no longer hold true.
big contains three functions used to reflow and format text
in a pleasing manner. In the order you should use them, they are
split_text_with_code,
wrap_words(),,
and optionally
merge_columns.
This trio of functions gives you the following word-wrap superpowers:
The big word wrapping functions also distinguish between "text" and "code". The main distinction is, "text" lines can get word-wrapped, but "code" lines shouldn't. big considers any line starting with enough whitespace to be a "code" line; by default, this is four spaces. Any non-blank line that starting with four spaces is a "code" line, and any non-blank line that starts with less than four spaces is a "text" line.
In "text" mode:
In "code" mode:
Also, whenever
split_text_with_code
switches between
"text" and "code" mode, it emits a paragraph break.
A split text array is an intermediary data structure used by big.text functions to represent text. It's literally just an array of strings, where the strings represent individual word-wrappable substrings.
split_text_with_code
returns a split text array, and
wrap_words()
consumes a split text array.
You'll see four kinds of strings in a split text array:
'\n'.'\n\n'.This might be clearer with an example or two. The following text:
hello there!
this is text.
this is a second paragraph!
would be represented in a Python string as:
"hello there!\nthis is text.\n\n\nthis is a second paragraph!"
Note the three newlines between the second and third lines.
If you then passed this string in to
split_text_with_code,
it'd return this split text array:
[ 'hello', 'there!', 'this', 'is', 'text.', '\n\n',
'this', 'is', 'a', 'second', 'paragraph!']
split_text_with_code
merged the first two lines together into
a single paragraph, and collapsed the three newlines separating
the two paragraphs into a "paragraph break" marker
(two newlines in one string).
Now let's add an example of text with some "code". This text:
What are the first four squared numbers?
for i in range(1, 5):
print(i**2)
Python is just that easy!
would be represented in a Python string as (broken up into multiple strings for clarity):
"What are the first four squared numbers?\n\n"
+
" for i in range(1, 5):\n\n\n"
+
" print(i**2)\n\nPython is just that easy!"
split_text_with_code
considers the two lines with initial whitespace as "code" lines,
and so the text is split into the following split text array:
['What', 'are', 'the', 'first', 'four', 'squared', 'numbers?', '\n\n',
' for i in range(1, 5):', '\n', '\n', '\n', ' print(i**2)', '\n\n',
'Python', 'is', 'just', 'that', 'easy!']
Here we have a "text" paragraph, followed by a "code" paragraph, followed by a second "text" paragraph. The "code" paragraph preserves the internal newlines, though they are represented as individual "line break" markers (strings containing a single newline). Every paragraph is separated by a "paragraph marker".
Here's a simple algorithm for joining a split text array back into a single string:
prev = None
a = []
for word in split_text_array:
if not (prev and prev.isspace() and word.isspace()):
a.append(' ')
a.append(word)
text = "".join(a)
Of course, this algorithm is too simple to do word wrapping.
Nor does it handle adding two spaces after sentence-ending
punctuation. In practice, you shouldn't do this by hand;
you should use
wrap_words.
merge_columns
merges multiple strings into columns on the same line.
For example, it could merge these three Python strings:
[
"Here's the first\ncolumn of text.",
"More text over here!\nIt's the second\ncolumn! How\nexciting!",
"And here's a\nthird column.",
]
into the following text:
Here's the first More text over here! And here's a
column of text. It's the second third column.
column! How
exciting!
(Note that
merge_columns
doesn't do its own word-wrapping;
instead, it's designed to consume the output of
wrap_words.)
Each column is passed in to
merge_columns
as a "column tuple":
(s, min_width, max_width)
s is the string,
min_width is the minimum width of the column, and
max_width is the minimum width of the column.
As you saw above, s can contain newline characters,
and
merge_columns
obeys those when formatting each column.
For each column,
merge_columns
measures the longest
line of each column. The width of the column is determined
as follows:
min_width characters long,
the column will be min_width characters wide.min_width
characters long, and less than or equal to max_width
characters long, the column will be as wide as the longest line.max_width characters long,
the column will be max_width characters wide, and lines that
are longer than max_width characters will "overflow".What is "overflow"? It's a condition
merge_columns
may encounter when the text in a column is wider than that
column's max_width.
merge_columns
needs to consider both "overflow lines",
lines that are longer than max_width, and "overflow columns",
columns that contain one or more overflow lines.
What does
merge_columns
do when it encounters overflow?
merge_columns
supports three "strategies" to deal with this condition, and you can specify
which one you want using its overflow_strategy parameter. The three
strategies are:
OverflowStrategy.RAISE: Raise an OverflowError exception. The default.
OverflowStrategy.INTRUDE_ALL: Intrude into all subsequent columns on
all lines where the overflowed column is wider than its max_width.
The subsequent columns "make space" for the overflow text by not adding
text on those overflowed lines; this is called "pausing" their output.
OverflowStrategy.DELAY_ALL: Delay all columns after the overflowed
column, not beginning any until after the last overflowed line
in the overflowed column. This is like the INTRUDE_ALL strategy,
except that the columns "make space" by pausing their output until
the last overflowed line.
When overflow_strategy is INTRUDE_ALL or DELAY_ALL, and
either overflow_before or overflow_after is nonzero, these
specify the number of extra lines before or after
the overflowed lines in a column where the subsequent columns
"pause".
TopologicalSorterbig's TopologicalSorter
is a drop-in replacement for
graphlib.TopologicalSorter
in the Python standard library (new in 3.9).
However, the version in big has been greatly upgraded:
prepare is now optional, though it still performs a cycle check.g with the new method g.remove(node).
Again, you can do this at any time, even while iterating over the graph.
Removing a node from the graph always succeeds, assuming the node is in the graph.get_ready, done, and __bool__
methods. There's a default view built in to the graph object;
the get_ready, done, and __bool__ methods on a graph just call
into the graph's default view. You can create a new view at any time
by calling the new view method.Note that if you're using a view to iterate over the graph, and you modify the graph,
and the view now represents a state that isn't coherent with the graph,
attempting to use that view raises a RuntimeError. (I'll define what I mean
by view "coherence" in the next subsection.)
This implementation also fixes some minor warts with the existing API:
static_order and get_ready/done are mutually exclusive. If you ever call
get_ready on a graph, you can never call static_order, and vice-versa. The implementaiton in big
doesn't have this restriction, because its implementation of static_order creates and uses a new view object
every time it's called.static_order once.
The implementation in big solves this in several ways: it allows you to create as many views as you
want, and you can call the new reset method on a view to reset it to its initial state.So what does it mean for a view to no longer be coherent with the graph? Consider the following code:
g = big.TopologicalSorter()
g.add('B', 'A')
g.add('C', 'A')
g.add('D', 'B', 'C')
g.add('B', 'A')
v = g.view()
g.ready() # returns ('A',)
g.add('A', 'Q')
First this creates a graph g with a classic "diamond"
dependency pattern. Then it creates a new view v, and gets
the currently "ready" nodes, which consists just of the node
'A'. Finally it adds a new dependency: 'A' depends on 'Q'.
At this moment, view v is no longer coherent. 'A' has been
marked as "ready", but 'Q' has not. And yet 'A' depends on 'Q'.
All those statements can't be true at the same time!
So view v is no longer coherent, and any attempt to interact
with v raises an exception.
To state it more precisely: if view v is a view on graph g,
and you call g.add('Z', 'Y'),
and neither of these statements is true in view v:
'Y' has been marked as done.'Z' has not yet been yielded by get_ready.then v is no longer "coherent".
(If 'Y' has been marked as done, then it's okay to make 'Z' dependent on
'Y' regardless of what state 'Z' is in. Likewise, if 'Z' hasn't been yielded
by get_ready yet, then it's okay to make 'Z' dependent on 'Y' regardless
of what state 'Y' is in.)
Note that you can restore a view to coherence. In this case,
removing either Y or Z from g would resolve the incoherence
between v and g, and v would start working again.
Also note that you can have multiple views, in various states of iteration, and by modifying the graph you may cause some to become incoherent but not others. Views are completely independent from each other.
One minor complaint I have about Python regards inner classes. An "inner class" is a class defined inside another class. And, well, inner classes seem kind of half-baked. Unlike functions, inner classes don't get bound to the object.
Consider this Python code:
class Outer(object):
def method(self):
pass
class Inner(object):
def __init__(self):
pass
o = Outer()
o.method()
i = o.Inner()
When o.method is called, Python automatically passes in the o object as the first parameter
(generally called self). In object-oriented lingo, o is bound to method, and indeed
Python calls this object a bound method:
>>> o.method
<bound method Outer.method of <__main__.Outer object at 0x########>>
But that doesn't happen when o.Inner is called. (It does pass in
a self, but in this case it's the newly-created Inner object.)
There's just no built-in way for the o.Inner object being constructed
to automatically get a reference to o. If you need one, you must
explicitly pass one in, like so:
class Outer(object):
def method(self):
pass
class Inner(object):
def __init__(self, outer):
self.outer = outer
o = Outer()
o.method()
i = o.Inner(o)
This seems redundant. You don't have to pass in o explicitly to method calls,
why should you have to pass it in explicitly to inner classes?
Well--now you don't have to!
You just decorate the inner class with @big.BoundInnerClass,
and BoundInnerClass takes care of the rest!
Let's modify the above example to use our BoundInnerClass
decorator:
from big import BoundInnerClass
class Outer(object):
def method(self):
pass
@BoundInnerClass
class Inner(object):
def __init__(self, outer):
self.outer = outer
o = Outer()
o.method()
i = o.Inner()
Notice that Inner.__init__ now requires an outer parameter,
even though you didn't pass in any arguments to o.Inner.
When it's called, o is magically passed in to outer!
Thanks, BoundInnerClass! You've saved the day!
Decorating an inner class like this always adds a second positional
parameter, after self. And, like self, you don't have
to use the name outer, you can use any name you like.
(Although it's probably a good idea, for consistency's sakes.)
Bound inner classes get slightly complicated when mixed with inheritance. It's not all that difficult, you merely need to obey the following rules:
A bound inner class can inherit normally from any unbound class.
To subclass from a bound inner class while still inside the outer
class scope, or when referencing the inner class from the outer class
(as opposed to an instance of the outer class), you must actually
subclass or reference classname.cls. This is because inside the
outer class, the "class" you see is actually an instance of a
BoundInnerClass object.
All classes that inherit from a bound inner class must always call the
superclass's __init__. You don't need to pass in the outer parameter;
it'll be automatically passed in to the superclass's __init__ as before.
An inner class that inherits from a bound inner class, and which also
wants to be bound to the outer object, should be decorated with
BoundInnerClass.
An inner class that inherits from a bound inner class, but doesn't
want to be bound to the outer object, should be decorated with
UnboundInnerClass.
Restating the last two rules: every class that descends from any
BoundInnerClass
should be decorated with either
BoundInnerClass
or
UnboundInnerClass.
Which one you use depends on what behavior you want--whether or
not you want your inner subclass to automatically get the outer
instance passed in to its __init__.
Here's a simple example using inheritance with bound inner classes:
from big import BoundInnerClass, UnboundInnerClass
class Outer(object):
@BoundInnerClass
class Inner(object):
def __init__(self, outer):
self.outer = outer
@UnboundInnerClass
class ChildOfInner(Inner.cls):
def __init__(self):
super().__init__()
o = Outer()
i = o.ChildOfInner()
We followed the rules:
Inner inherits from object; since object isn't a bound inner class,
there are no special rules about inheritance Inner needs to obey.ChildOfInner inherits from Inner.cls, not Inner.ChildOfInner inherits from a
BoundInnerClass,
it must be
decorated with either BoundInnerClass
or UnboundInnerClass.
It doesn't want the outer object passed in, so it's decorated
with UnboundInnerClass.ChildOfInner.__init__ calls super().__init__.Note that, because ChildOfInner is decorated with
UnboundInnerClass,
it doesn't take an outer parameter. Nor does it pass in an outer
argument when it calls super().__init__. But when the constructor for
Inner is called, the correct outer parameter is passed in--like magic!
Thanks again, BoundInnerClass!
If you wanted ChildOfInner to also get the outer argument passed in to
its __init__, just decorate it with BoundInnerClass
instead of
UnboundInnerClass,
like so:
from big import BoundInnerClass
class Outer(object):
@BoundInnerClass
class Inner(object):
def __init__(self, outer):
self.outer = outer
@BoundInnerClass
class ChildOfInner(Inner.cls):
def __init__(self, outer):
super().__init__()
assert self.outer == outer
o = Outer()
i = o.ChildOfInner()
Again, ChildOfInner.__init__ doesn't need to explicitly
pass in outer when calling super.__init__.
You can see more complex examples of using inheritance with
BoundInnerClass
(and UnboundInnerClass)
in the big test suite.
If you refer to a bound inner class directly from the outer class,
rather than using the outer instance, you get the original class.
This ensures that references to Outer.Inner are consistent; this
class is also a base class of all the bound inner classes. Additionally,
if you attempt to construct an instance of an unbound Outer.Inner
class without referencing it via an instance, you must pass in the
outer parameter by hand--just like you'd have to pass in the self
parameter by hand when calling a method on the class itself rather
than on an instance of the class.
If you refer to a bound inner class from an outer instance, you get a subclass of the original class.
Bound classes are cached in the outer object, which both provides
a small speedup and ensures that isinstance relationships are
consistent.
You must not rename inner classes decorated with either
BoundInnerClass
or UnboundInnerClass!
The implementation of
BoundInnerClass
looks up
the bound inner class in the outer object by name in several places.
Adding aliases to bound inner classes is harmless, but the original
attribute name must always work.
Bound inner classes from different objects are different classes.
This is symmetric with bound methods; if you have two objects
a and b that are instances of the same class,
a.BoundInnerClass != b.BoundInnerClass, just as a.method != b.method.
The binding only goes one level deep; if you had an inner class C
inside another inner class B inside a class A, the constructor
for C would be called with the B object, not the A object.
Similarly, if you have a bound inner class B inside a class A,
and another bound inner class D inside a class C, and D
inherits from B, the constructor for D will be called with
the B object but not the A object. When D calls super().__init__
it'll have to fill in the outer parameter by hand.
There's a race condition in the implementation: if you access a
bound inner class through an outer instance from two separate threads,
and the bound inner class was not previously cached, the two threads
may get different (but equivalent) bound inner class objects, and only
one of those instances will get cached on the outer object. This could
lead to confusion and possibly cause bugs. For example, you could have
two objects that would be considered equal if they were instances of
the same bound inner class, but would not be considered equal if
instantiated by different instances of that same bound inner class.
There's an easy workaround for this problem: access the bound inner
class from the __init__ of the outer class, which should allow
the code to cache the bound inner class instance before a second
thread could ever get a reference to the outer object.
2025/01/06
search_path to the big.file module. search_path
implements "search path" functionality; given a list of
directories, a filename, and optionally a list of file
extensions to try, returns the first existing file that matches.multisplit and split_delimiters now properly support
subclasses of str. All strings yielded by these functions
are now guaranteed to be slices of the original s parameter
passed in, or otherwise produced by making method calls on the
original s parameter that return strings.2024/12/15
A teeny tiny new feature.
LineInfo now supports a copy method, which returns a copy of the LineInfo
object in its current state.2024/12/13
It's a big release tradition! Here's another small big release, less than a day after the last big big release.
decode_python_script
now supports "universal newlines". It accepts a new newline parameter
which behaves identically to the newline parameter for Python's built-in
open function.read_python_file
was broken in 0.12.5; the newline parameter was simply ignored.
It now works great--it passes newline to decode_python_script.
(Sorry I missed this; I use Linux and don't need to convert newlines.)Note: Whoops! Forgot to ever release 0.12.6 as a package. Oh well.
2024/12/13
Added decode_python_script
to the big.text module.
decode_python_script scans a binary Python script and
decodes it to Unicode--correctly. Python scripts can
specify an explicit encoding in two diferent ways:
a Unicode "byte order mark",
or a PEP 263 "source file encoding" line.
decode_python_script handles either, both, or neither.
Added read_python_file
to the big.file module.
read_python_file reads a binary Python file from the
filesystem and decodes it using decode_python_script.
Added python_delimiters
to the big.text module. This is
a new predefined set of delimiters
for use with split_delimeters, enabling it to correctly
process Python scripts. python_delimiters defines all
delimiters defined by Python, including all 100 possible
string delimiters (no kidding!). If you want to parse the
delimiters of Python code, and you don't want to use the
Python tokenizer, you should use python_delimiters
with split_delimiters.
Note that defining python_delimiters correctly was difficult,
and big's Delimiters API isn't expressive enough to
express all of Python's semantics. At this point the
python_delimiters object doesn't itself actually define all its
semantics; rather, at module load time it's compiled into a special
internal runtime format which is cached, and then there's
manually-written code that tweaks this compiled form so python_delimiters
can correctly handle Python's special cases. So, you're encouraged
to use python_delimiters, but if you modify it and use the
modified version, the modified version won't inherit all
those tweaks, and will lose the ability to handle many of
Python's weirder semantics.
Important note: When you use python_delimiters, you must
include the linebreak characters in the lines you split using
split_delimiters. This is necessary to support the comment
delimiter correctly, and to enforce the
no-linebreaks-inside-single-quoted-strings rule.
There can be small differences in Python's syntax from one
version to another. python_delimiters is therefore
version-sensitive, using the semantics appropriate for the
version of Python it's being run under. If you want to
parse Python delimiters using the semantics of another version
of the language, use instead python_delimiters_version[s]
where s is a string containing the dotted Python major and minor
version you want to use, for example python_delimiters_version["3.10"]
to use Python 3.10 semantics. (At the moment there are
no differences between versions; this is planned for future
versions of big.)
Added python_delimiters_version
to the big.text module.
This maps simple Python version strings ("3.6", "3.13")
to python_delimiters values implementing the semantics
for that version. Currently all the values of this dict
are identical, but that should change in the future.
A breaking API change to split_delimiters is coming.
split_delimiters now yields an object that
can yield either three or four values. Previous to 0.12.5, the
split_delimiters iterator always yielded a tuple of three values,
called text, open, and close. But python_delimiters
required adding a fourth value, change.
When change is true, we are changing from one delimiter to
another, without entering a new nested delimiter. The canonical
example of this is inside a Python f-string:
`f"{abc:35}"`
Here the colon (:) is a "change" delimiter. Inside the curly
braces inside the f-string, before the colon, the hash character
(#) acts as a line comment character. But after the colon
it's just another character. We've changed semantics, but we
haven't pushed a new delimiter pair. The only way to accurately
convey this behavior was to add this new change field to the values
yielded by split_delimiters.
The goal is to eventually transition to split_delimiters yielding
all four of these values (text, open, close, and change).
But this will be a gradual process; as of 0.12.5, existing
split_delimiters calls will continue to work unchanged.
split_delimiters now yields a custom object, called
SplitDelimitersValue. This object is configurable to yield
either three or four values. The rules are:
yields=4 to split_delimiters,
the object it yields will yield four values.delimiters=python_delimiters to split_delimiters,
the object it yields will yield four values. (python_delimiters
is new, so any calls using it must be new code, therefore this
change won't break existing calls.)split_delimiters will yield
three values, as it did in versions prior to 0.12.5.split_delimiters will eventually change to always yielding
four values, but big won't publish this change until at least June 2025.
Six months after that change--at least December 2025--big will remove
the yields parameter to split_delimiters.
Minor semantic improvement:
PushbackIterator
no longer
evaluates the iterator you pass in in a boolean context.
(All we really needed to do was compare it to None,
so now that's all we do.)
A minor change to the
Delimiter
object used with
split_delimiters: previously, the quoting and
escape values had to agree, either both being true
or both being false. However, python_delimiters
necessitated relaxing this restriction, as there are
some delimiters (! inside curly braces in an f-string,
: inside curly braces in an f-string) that are "quoting"
but don't have an escape string. So now, the restriction
is simply that if escape is true, quoting must also
be true.
2024/11/15
text module: format_map.
This works like Python's str.format_map method,
except it allows nested curly-braces. Example:
big.format_map("The {extension} file is {{extension} size} bytes.", {'extension': 'mp3', 'mp3 size': 8555})Version.format is like strftime but for Version objects.
You pass in a format string with Version attributes in curly braces
and it formats the string with values from that Version object.Version constructor now accepts a packaging.Version object
as an initializer. Embrace and extend!lines now takes two new arguments:
clip_linebreaks, default is true.
If true, it clips the linebreaks off the lines before yielding them,
otherwise it doesn't. (Either way, the linebreaks are still stored
in info.end.)source, default is an empty string.
source should represent the source of the line in a
meaninful way to the user. It's stored in the LinesInfo
objects yielded by lines, and should be incorporated into
error messages.LineInfo.clip_leading and LineInfo.clip_trailing now automatically
detect if you've clipped the entire line, and if so move all clipped
text to info.trailing (and adjust the column_number accordingly).LineInfo.clip_leading and LineInfo.clip_trailing: Minor performance
upgrade. Previously, if the user passed in the string to clip, the
two functions would throw it away then recreate it. Now they just use
the passed-in string.p.s. 56
2024/09/17
Optimized
split_delimiters.
The new version uses a much more efficient internal representation
of how to react to the various delimiters when processing the text.
Perfunctory timeit experiments suggest this new split_delimiters
is maybe 5-6% faster than it was in 12.2.
Minor breaking change: split_delimiters now consistently
raises SyntaxError for mismatched delimiters. (Previously it
would sometimes raise ValueError.)
2024/09/11
A minor semantic change to lines_strip_indent:
when it encounters a whitespace-only line, it clips the line to trailing
in the LineInfo object. It used to clip such lines to leading. But this
changed LineInfo.column_number in a nonsensical way.
This behavior is policy going forward: if a lines modifer function ever clips
the entire line, it must clip it to trailing rather than leading. It
shouldn't matter one way or another, as whitespace-only lines arguably
shouldn't have any explicit semantics. But it makes intuitive sense to me
that their empty line should be at column number 1, rather than 9 or 13
or whatnot. (Especially considering that with lines_strip_indent their
indent value is synthetic anyway, inferred by looking ahead.)
Major cleanup to the lines modifier test suites.
2024/09/07
In fine big tradition, here's an update published immediately after a big release.
Surprisingly, even though this is only a small update, it still adds two new packages to big: metadata and version.
There's sadly one breaking change.
big.metadataNew package. A package containing metadata about big itself. Currently only contains one thing: version.
big.versionNew package. A package for working with version information.
lines_strip_line_commentsThis API has breaking changes.
The default value for quotes has changed. Now it's
what it should always have been: empty. No quote marks
are defined by default, which means the default behavior of
lines_strip_line_comments
is now to simply truncate the line
at the leftmost comment marker.
Processing quote marks by default was always too opinionated
for this function. Consider: having ' active as a quote
marker meant that single-quotes need to be balanced,
which means you can't process a line like this that only has one.
Wish I'd figured this out before the release yesterday! Hopefully this will only cause smiles, and no teeth-gnashing.
metadata.versionNew value. A
Version
object representing the current version of big.
VersionNew class.
Version
represents a version number. You can
construct them from PEP 440-compliant
version strings, or specify them using keyword-only parameters.
Version
objects are immutable, ordered, and hashable.
2024/09/06
Lots of changes this time! Most of 'em are in the big.text
module, particularly the lines and lines modifier
functions. But plenty of other modules got in on the fun too.
big even has a new module: deprecated. Deprecated
functions and classes get moved into this module. Note that
the contents of deprecated are not automatically imported
into big.all.
The following functions and classes have breaking changes:
These functions have been renamed:
lines_filter_comment_lines is now lines_filter_line_comment_lines
lines_strip_comments is now lines_strip_line_comments
parse_delimiters is now split_delimiters
big has five new functions:
Finally, here's an in-depth description of all changes in big 0.12, sorted by API name.
bytes_linebreaks and bytes_linebreaks_without_crlfExtremely minor change! Python's bytes and str objects
don't agree on which ASCII characters represent line breaks.
The str object obeys the Unicode standard, which means
there are four:
\n \v \f \r
For some reason, Python's bytes object only supports two:
\n \r
I have no idea why this is. We might fix it. And if we do,
big is ready. It now calculates
bytes_linebreaks
and
bytes_linebreaks_without_crlf
on the fly to agree with Python.
If either (or both) work as newline characters for the splitlines
method on a bytes object, they'll automatically be inserted
into these iterables of bytes linebreaks.
combine_splitsNew function. If you split a string two different ways,
producing two arrays that sum to the original string,
combine_splits will merge those splits together, producing
a new array that splits in every place any of the two split
arrays had a split.
Example:
>>> big.combine_splits("abcdefg", ['a', 'bcdef', 'g'], ['abc', 'd', 'efg'])
['a, 'bc', 'd', 'ef', 'g']
DelimiterThis API has breaking changes.
Delimiter is a simple data class, representing information about
delimiters to
split_delimiters
(previously parse_delimiters).
split_delimiters has changed, and some of those changes are
reflected in the Delimiter object; also, some changes to Delimiter
are simply better API choices.
The old Delimiter object is deprecated but still available,
as big.deprecated.Delimiter. It should only be used with
big.deprecated.parse_delimiters, which is also deprecated.
big.deprecated.Delimiter will be removed when
big.deprecated.parse_delimiters is removed, which will be
no sooner than September 2025.
Changes:
Delimiter object was open,
and was stored as the open attribute. These have both been
completely removed. Now, the "open delimiter" is specified
as a key in a dictionary of delimiters, mapping open delimiters
to Delimiter objects.Delimiter object had a boolean backslash attribute;
if it was True, that delimiter allows escaping using a backslash.
Now Delimiter has an escape parameter and attribute,
specifying the escape string you want to use inside that
set of delimiters.Delimiter also now has two new attributes, quoting and
multiline. These default to False and True respectively;
you can specify values for these with keyword-only arguments
to the constructor.Delimiter object is read-only after construction,
and is hashable.encode_stringsSlightly liberalized the types it accepts. It previously
required o to be a collection; now o can be a bytes
or str object. Also, it now explicitly supports set.
get_int_or_floatMinor behavior change. If the o you pass in is a float,
or can be converted to float (but couldn't be converted directly
to an int), get_int_or_float will experimentally convert that
float to an int. If the resulting int compares equal to that
float, it'll return the int, otherwise it'll return the float.
For example, get_int_or_float("13.5") still returns 13.5
(a float), but get_int_or_float("13.0") now returns 13
(an int). (Previously, get_int_or_float("13.0") would
have returned 13.0.)
This better represents the stated aesthetic of the function--it prefers ints to floats. And since the int is exactly equal to the float, I assert this is completely backwards compatible.
HeapMinor updates to the documentation and to the text of some exceptions.
LineInfoThis API has breaking changes.
Breaking change: the
LineInfo
constructor has a
new lines positional parameter, added in front of
the existing positional parameters. This new first argument
should be the lines iterator that yielded this
LineInfo object. It's stored in the lines attribute.
(Why this change? The lines object contains information
needed by the lines modifiers, for example tab_width.)
Minor optimization:
LineInfo
objects previously had many
optional fields, which might or might not be added
dynamically. Now all fields are pre-added. (This makes
the CPython 3.13 runtime happier; it really wants you to
set all your class's attributes in its __init__.)
Minor breaking change: the original string stored in the
line attribute now includes the linebreak character, if any.
This means concatenating all the info.line strings
will reconstruct the original s passed in to lines.
New feature: while some methods used to update the leading
attribute when they clipped leading text from the line,
the "lines modifiers" are now very consistent about updating
leading, and the new symmetrical attribute trailing.
New feature:
LineInfo
now has an end attribute,
which contains the end-of-line character that ended this line.
These three attributes allow us to assert a new invariant:
as long as you modify the contents of line (e.g.
turning tabs into spaces),
info.leading + line + info.trailing + info.end == info.line
LineInfo
objects now always have these attributes:
lines, which contains the base lines iterator.line, which contains the original unmodified line.line_number, which contains the line number of
this line.column_number, which contains the starting column
number of the first character of this line.indent, which contains the indent level of the
line if computed, and None otherwise.leading, which contains the string stripped from
the beginning of the line. Initially this is the
empty string.trailing, which contains the string stripped from
the end of the line. Initially this is the
empty string.end, which is the end-of-line character
that ended the current line. For the last line yielded,
info.end will always be the empty string. If the last
character of the text split by lines was an end-of-line
character, the last line yielded will be the empty string,
and info.end will also be the empty string.match, which contains a Match object if this line
was matched with a regular expression, and None otherwise.LineInfo.clip_leading and LineInfo.clip_trailingLineInfo
also has two new methods:
LineInfo.clip_leading
and
LineInfo.clip_trailing(line, s).
These methods clip a leading or
trailing substring from the current line, and transfer
it to the relevant field in
LineInfo
(either leading or
trailing). clip_leading also updates the column_number
attribute.
The name "clip" was chosen deliberately to be distinct from "strip". "strip" functions on strings remove substrings and throws them away; my "clip" functions on strings removes substrings and puts them somewhere else.
lines_filter_comment_lineslines_filter_comment_lines has been renamed to
lines_filter_line_comment_lines.
For backwards compatibility, the function
is also available under the old name; this old name will
eventually be removed, but not before September 2025.
lines_filter_line_comment_linesThis API has breaking changes.
New name for lines_filter_comment_lines.
Correctness improvements:
lines_filter_line_comment_lines
now enforces that single-quoted strings can't span lines,
and multi-quoted strings must be closed before the end of
the last line.
Minor optimization: for every line, it used to lstrip a copy of
the line, then use a regular expression to see if the line started
with one of the comment characters. Now the regular expression
itself skips past any leading whitespace.
lines_grepNew feature: lines_grep
has always used re.search to examine
the lines yielded. It now writes the result to info.match.
(If you pass in invert=True to lines_grep, lines_grep
still writes to the match attribute--but it always writes None.)
If you want to write the re.Match object to another attribute,
pass in the name of that attribute to the keyword-only
parameter match.
lines_rstrip and lines_stripNew feature:
lines_rstrip
and
lines_strip
now both accept a
separators argument; this is an iterable of separators,
like the argument to
multisplit.
The default value of None preserves the previous behavior,
stripping whitespace.
lines_sortNew feature:
lines_sort
now accepts a key parameter,
which is used as the key argument for list.sort.
The value passed in to key is the (info, line) tuple
yielded by the upstream iterator. The default value preserves
the previous behavior, sorting by the line (ignoring the
info).
lines_strip_commentsThis function has been renamed
lines_strip_line_comments
and
rewritten, see below. The old deprecated version will be
available at big.deprecated.lines_strip_comments until at
least September 2025.
Note that the old version of line_strip_comments still uses
the current version of
LineInfo,
so use of this deprecated
function is still exposed to those breaking changes.
(For example, LineInfo.line now includes the linebreak character
that terminated the current line, if any.)
lines_strip_indentBugfix:
lines_strip_indent
previously required
whitespace-only lines to obey the indenting rules, which was
a mistake. My intention was always for lines_strip_indent
to behave like Python, and that includes not really caring
about the intra-line-whitespace for whitespace-only
lines. Now lines_strip_indent behaves more like Python:
a whitespace-only line behaves as if it has
the same indent as the previous line. (Not that the
indent value of an empty line should matter--but this
behavior is how you'd intuitively expect it to work.)
lines_strip_line_commentsThis API has breaking changes.
lines_strip_line_comments
is the new name for the old
lines_strip_comments lines modifier function. It's also
been completely rewritten.
Changes:
triple_quotes parameter
to support multiline strings. The new version supports separate
parameters for single-line quote marks (quotes) and multiline
quote marks (multiline_quotes).backslash parameter has been renamed to escape.rstrip parameter has been removed. If you need to
rstrip the line after stripping the comment, wrap your
lines_strip_line_comments call with a
lines_rstrip
call.SyntaxError if quoted
strings using non-multiline quote marks contain newlines.(lines_strip_line_comments has always been implemented using
split_quoted_strings;
this is why it now supports multicharacter
quote marks and escape strings. It also benefits from the
new optimizations in split_quoted_strings.)
multisplitMinor optimizations.
multisplit
used to locally define a
new generator function, then call it and return the generator.
I promoted the generator function to module level, which means
we no longer rebind it each time multisplit is called. As
a very rough guess, this can be as much as a 10% speedup for
multisplit run on very short workloads. (It's also never
slower.)
I also applied this same small optimization to several other
functions in the text module. In particular,
merge_columns
was binding functions inside a loop (!!). (Dumb, huh!)
These local functions are still bound inside merge_columns,
but now at least they're outside the loop.
Another minor speedup for multisplit: when reverse=True,
it used to reverse the results three times! multisplit
now explicitly observes and manages the reversed state of the
result to avoid needless reversing.
parse_delimitersThis function has been renamed
split_delimiters
and rewritten,
see below. The old version is still available, using the name
big.deprecated.parse_delimiters module, and will be available
until at least September 2025.
SchedulerCode cleanups both in the implementation and the test suite, including one minor semantic change.
Cleaned up Scheduler._next, the internal method call
that implements the heart of the scheduler. The only externally
visible change: the previous version would call sleep(0) every
time it yielded an event. On modern operating systems this should
yields the rest of the current thread's current time slice back
to the OS's scheduler. This can make multitasking smoother,
particularly in Python programs. But this is too opinionated for
library code--if you want a sleep(0) there, by golly, you can
call that yourself when the Scheduler object yields to you.
I've restructured the code and eliminated this extraneous sleep(0).
Also, rewrote big chunks of the test suite (tests/test_scheduler.py).
The multithreaded tests are now much better synchronized, while
also becoming easier to read. Although it seems intractable to
purge all race conditions from the test suite, this change has
removed most of them.
split_delimitersThis API has breaking changes.
split_delimiters
is the new name for the old parse_delimiters
function. The function has also been completely re-tooled and
re-written.
Changes:
parse_delimiters took an iterable of Delimiters
objects, or strings of length 2. split_delimiters
takes a dictionary mapping open delimiter strings to
Delimiter objects, and Delimiter objects no
longer have an "open" attribute.split_delimiters now accepts an state parameter,
which specifies the initial state of nested delimiters.split_delimiters no longer cares if there were unclosed
open delimiters at the end of the string. (It used to
raise ValueError.) This includes quote marks; if you
don't want quoted strings to span multiple lines, it's up
to you to detect it and react (e.g. raise an exception).parse_delimiters manually parsed the input string
character by character. split_delimiters uses
multisplit,
so it zips past the uninteresting characters and only examines
the delimiters and escape characters. It's always faster,
except for some trivial calls (which are fast enough anyway).multisplit: open delimiters,
close delimiters, and the escape string may now all be
any nonzero length. (In the face of ambiguity,
split_delimiters will always choose the longer delimiter.)See also changes to Delimiter.
split_quoted_stringsThis API has breaking changes.
split_quoted_strings
has been completely re-tooled and
re-written. The new API is simpler, easier to understand,
and conceptually clarified. It's a major upgrade!
Changes:
(is_quote, segment), where
is_quote was a boolean value indicating whether or not
segment was quoted. If segment was quoted, it began
and ended with (single character) quote marks. To reassemble
the original string, join together all the segment strings
in order.(leading_quote, segment, trailing_quote),
where leading_quote and trailing_quote are either matching
quote marks or empty. If they're true values, the segment
string is inside the quotes. To reassemble the original string,
join together all the yielded strings in order.backslash parameter has been replaced by a new parameter,
escape. escape allows specifying the escape string, which
defaults to '\' (backslash). If you specify a false value,
there will be no escape character in strings.quotes only contains ' (single-quote)
and " (double-quote). The previous version also
recognized """ and ''' as multiline quote marks
by default; this is no longer true, as it's too
opinionated and Python-specific.SyntaxError if there's a newline character inside
a string delimited with a quote marker from quotes.triple_quotes parameter.
That's been removed in favor of a new parameter,
multiline_quotes. multiline_quotes is like quotes,
except that newline characters are allowed inside their
quoted strings.split_quoted_string accepts another new parameter,
state, which sets the initial state of quoting.split_quoted_string used a
hand-coded parser, manually analyzing each character in
the input text. Now it uses
multisplit,
so it only bothers to examine
the interesting substrings. multisplit has a large
startup cost the first time you use a particular set of
iterators, but this information is cached for subsequent calls.
Bottom line, the new version is much faster
for larger workloads. (It can be slower for trivial
examples... where speed doesn't matter anyway.)multisplit: quotes
now supports quote delimiters and an escape string
of any nonzero length. In the case of ambiguity--if
more than one quote delimiter matches at a
time--split_quoted_string will always choose the
longer delimiter.split_title_caseNew function.
split_title_case
splits a string at word boundaries,
assuming the string is in "TitleCase".
StateManagerSmall performance upgrade for
StateManager.
observers.
StateManager always uses a copy of the observer
list (specifically, a tuple) when calling the observers; this
means it's safe to modify the observer list at any time.
StateManager used to always make a fresh copy every time you
called an event; now it uses a cached copy, and only recomputes
the tuple when the observer list changes.
(Note that it's not thread-safe to modify the observer list
from one thread while also dispatching events in another.
Your program won't crash, but the list of observers called
may be unpredictable based on which thread wins or loses the
race. But this has always been true. As with many libraries,
the StateManager API leaves locking up to you.)
p.s. I'm getting close to declaring big as being version 1.0. I don't want to do it until I'm done revising the APIs.
p.p.s. Updated copyright notices to 2024.
p.p.p.s. Yet again I thank Eric V. Smith for his willingness to humor me in my how-many-parameters-could-dance-on-the-head-of-a-pin API theological discussions.
released 2023/09/19
Breaking change: renamed almost all the old whitespace and newlines tuples.
Worse yet, one symbol has the same name but a different value: ascii_whitespace!
I've also changed the suffix _without_dos to the more accurate and intuitive
_without_crlf, and similarly changed newlines to linebreaks.
Sorry for all the confusion. This resulted from a lot of research into whitespace
and newline characters, in Python, Unicode, and ASCII; please see the new deep-dive
Whitespace and line-breaking characters in Python and big
to see what all the fuss is about. Here's a summary of all the
changes to the whitespace tuples:
RENAMED TUPLES (old name -> new name)
ascii_newlines -> bytes_linebreaks
ascii_whitespace -> bytes_whitespace
newlines -> linebreaks
ascii_newlines_without_dos -> bytes_linebreaks_without_crlf
ascii_whitespace_without_dos -> bytes_whitespace_without_crlf
newlines_without_dos -> linebreaks_without_crlf
whitespace_without_dos -> whitespace_without_crlf
REMOVED TUPLES
utf8_newlines
utf8_whitespace
utf8_newlines_without_dos
utf8_whitespace_without_dos
UNCHANGED TUPLES (same name, same meaning)
whitespace
NEW TUPLES
ascii_linebreaks
ascii_whitespace
str_linebreaks
str_whitespace
unicode_linebreaks
unicode_whitespace
ascii_linebreaks_without_crlf
ascii_whitespace_without_crlf
str_linebreaks_without_crlf
str_whitespace_without_crlf
unicode_linebreaks_without_crlf
unicode_whitespace_without_crlf
Changed
split_text_with_code
implementation to use StateManager.
(No API or semantic changes, just an change to the internal implementation.)
New function in the big.text module: encode_strings,
which takes a container object containing str objects and returns an equivalent object
containing encoded versions of those strings as bytes.
When you call
multisplit
with a type mismatch
between 's' and 'separators', the exception it raises
now includes the values of 's' and 'separators'.
Added more tests for big.state to exercise all the string arguments
of accessor and dispatch.
The exhaustive
multisplit
tester now lets you
specify test cases as cohesive strings, rather
than forcing you to split the string manually.
The exhaustive
multisplit
tester is better at
internally verifying that it's doing the right
thing. (There are some internal sanity checks,
and those are more accurate now.)
Whoops! The name of the main class in big.state is
StateManager.
I accidentally wrote StateMachine instead in the docs... several times.
Originally the
multisplit
parameter 'separators'
was required. I changed it to optional a while ago,
with a default of None. (If you pass in None
it uses big.str_whitespace or big.bytes_whitespace,
depending on the type of s.) But the documentation
didn't reflect this change until... now.
Improved the prose in
The multi- family of string functions deep-dive.
Hopefully now it does a better job of selling multisplit to the reader.
The usual smattering of small doc fixes and improvements.
My thanks again to Eric V. Smith for his willingness to consider and discuss these issues. Eric is now officially a contributor to big, increasing the project's bus factor to two. Thanks, Eric!
released 2023/09/04
big.state module, with its exciting
StateManager
class!int_to_words
now supports the new ordinal keyword-only parameter, to produce
ordinal strings instead of cardinal strings. (The number 1
as a cardinal string is 'one', but as an ordinal string is 'first').pure_virtual decorator to big.builtin.<dl><dd>). Note that you need a blank
line after the <dl><dd> line, or else Markdown will ignore the
markup in the following paragraph. Thanks to Hugo van Kemenade
for his help confirming this! Oh, and, Hugo also fixed the image markup
so the big banner displays properly on PyPI. Thanks, Hugo!released 2023/07/22
Extremely minor release. No new features or bug fixes.
Log
deterministic. They now use a fake clock
that always returns the same values.pyproject.toml file.released 2023/06/28
released 2023/06/15
Bugfix! If an outer class Outer had an inner class Inner
decorated with @BoundInnerClass, and o is an instance of
Outer, and o evaluated to false in a boolean context,
o.Inner would be the unbound version of Inner. Now
it's the bound version, as is proper.
Modified tests/test_boundinnerclasses.py:
super".released 2023/06/11
int_to_words.sys.path, so you can run the tests on your
local copy without having to install. Especially
convenient for testing with old versions of Python!Note: tomorrow, big will be one year old!
released 2023/05/19
released 2023/05/19
parse_delimiters (ed: now split_delimiters)
and
Delimiter.released 2023/05/18
str and bytes support in big.text.
big.text now uniformly accept str or bytes
or a subclass of either. See the
Support for bytes and str section
for how it works.big.text are now more consistent about raising
TypeError vs ValueError. If you mix bytes and str
objects together in one call, you'll get a TypeError, but
if you pass in an empty iterable (of a correct type) where
a non-empty iterable is required you'll get a ValueError.
big.text generally tries to give the TypeError higher
priority; if you pass in a value that fails both the type
check and the value check, the big.text function will raise
TypeError first.re_rpartition.
I realized it had the same "reverse mode" problem that
I fixed in
multisplit
back in version 0.6.10: the regular expression should really
search the string in "reverse mode", from right to left.
The difference is whether the regular
expression potentially matches against overlapping strings.
When in forwards mode, the regular expression should prefer
the leftmost overlapping match, but in reverse mode it
should prefer the rightmost overlapping match. Most of the
time this produces the same list of matches as you'd
find searching the string forwards--but sometimes the matches come
out very different.
This was way harder to fix with re_rpartition than with multisplit,
because Python's re module only supports searching forwards.
I have to emulate reverse-mode searching by manually checking for
overlapping matches and figuring out which one(s) to keep--a lot of
work! Fortunately it's only a minor speed hit if you don't have
overlapping matches. (And if you do have overlapping matches,
you're probably just happy re_rpartition now produces correct
results--though I did my best to make it performant anyway.)
In the future, big will probably add support for the
PyPI package regex, which reimplements Python's re module
but adds many features... including reverse mode!reversed_re_finditer.
Behaves almost identically to the Python
standard library function re.finditer, yielding
non-overlapping matches of pattern in string. The difference
is, reversed_re_finditer searches string from right to left.
(Written as part of the
re_rpartition
rewrite mentioned above.)apostrophes, double_quotes,
ascii_apostrophes, ascii_double_quotes,
utf8_apostrophes, and utf8_double_quotes
to the big.text module. Previously the first
four of these were hard-coded strings inside
gently_title.
(And the last two didn't exist!)split_text_with_code, removed redundant code.
I think it has about the same number of if statements; if anything
it might be slightly faster.re_partition
and
re_rpartition
slightly, should now be very-slightly faster. (Well, re_rpartition
will be slower if your pattern finds overlapping matches. But at
least now it's correct!)released 2023/03/13
multisplit.
Internally, it does the
string splitting using re.split, which returns a list. It used
to iterate over the list and yield each element. But that meant keeping
the entire list around in memory until multisplit exited. Now,
multisplit
reverses the list, pops off the final element, and yields
that. This means
multisplit
drops all references to the split strings
as it iterates over the string, which may help in low-memory situations.released 2023/03/11
Scheduler:
Scheduler
object and into the
Regulator. Among other things, this
means that the
Scheduler
constructor no longer takes a lock argument.Regulator is now an abstract base class.
big.scheduler also provides two concrete implementations:
SingleThreadedRegulator
and
ThreadSafeRegulator.Regulator and
Event
are now defined in the big.scheduler namespace. They were
previously defined inside the Scheduler class.Event
constructor were rearranged. (You shouldn't care, as you
shouldn't be manually constructing
Event
objects anyway.)Scheduler now guarantees that it will only call now and wake
on a Regulator object while holding that Regulator's lock.released 2023/03/09
multisplit
and
multistrip
argument verification code. Both functions now consistently check all
their inputs, and use consistent error messages when raising an exception.released 2023/03/09
multisplit:
if you passed in a list of separators (or separators
was of any non-hashable type), and reverse was true,
multisplit would crash. It used separators as a key
into a dict, which meant separators had to be hashable.multisplit
now verifies that the s passed in is either str or bytes.released 2023/02/26
dateutils package an optional dependency. Only one function
needs it, parse_timestamp_3339Z().PushbackIterator().
It also uses slots now, which should make it a bit faster.released 2023/01/07
datetime_ensure_timezone(d, timezone) and
datetime_set_timezone(d, timezone).
These allow you to ensure or explicitly set a timezone on a datetime.datetime
object.timezone argument to
parse_timestamp_3339Z().gently_title()
now capitalizes the first letter after a left parenthesis.multirpartition function slightly. Its reverse
parameter now means to un-reverse its reversing behavior. Stated
another way, multipartition(reverse=X) and multirpartition(reverse=not X)
now do the same thing.released 2022/12/11
RuntimeError raised by TopologicalSorter.View
when the view is incoherent. Now it tells you exactly what nodes are
conflicting.multisplit.released 2022/12/11
translate_filename_to_exfat(s)
behavior: when modifying a string with a colon (':') not followed by
a space, it used to convert it to a dash ('-'). Now it converts the
colon to a period ('.'), which looks a little more natural. A colon
followed by a space is still converted to a dash followed by a space.tagged 2022/12/04
TopologicalSorter.print(),
it sorts the list of nodes, for consistency's sakes and for ease of reading.
But if the node objects don't support < or > comparison,
that throws an exception. TopologicalSorter.print() now catches
that exception and simply skips sorting. (It's only a presentation thing anyway.)multirpartition,
which is like
multipartition
but with reverse=True.Note: although version 0.6.12 was tagged, it was never packaged for release.
tagged 2022/11/13
import * all the symbols
from all those modules. But a friend (hi Mark Shannon!) talked
me out of this. It's convenient, but if a user doesn't care about
a particular module, why make them import it. So now the top-level
big module contains nothing but a version number, and you
can either import just the submodules you need, or you can import
big.all to get all the symbols (like big itself used to do).Note: although version 0.6.11 was tagged, it was never packaged for release.
released 2022/10/26
multisplit:
' x ', multisplit will prefer the leftmost instance.
But when reverse=True, it must prefer the rightmost instance.
Thanks to Eric V. Smith for suggesting the clever "reverse everything,
call re.split, and un-reverse everything" approach. That let me
fix this bug while still implementing on top of re.split!PROGRESSIVE mode for the strip keyword. This behaves
like str.strip: when splitting, strip on the left, then start splitting.
If we don't exhaust maxsplit, strip on the right; if we do exhaust
maxsplit, don't strip on the right. (Similarly for str.rstrip
when reverse=True.)strip to False. It used to be
NOT_SEPARATE. But this was too surprising--I'd forget that it
was the default, and turning on keep wouldn't return everything I
thought I should get, and I'd head off to debug multisplit, when in
fact it was behaving as specified. The Principle Of Least Surprise
tells me that strip defaulting to False is less surprising.
Also, maintaining the invariant that all the keyword-only parameters
to multisplit default to False is a helpful mnemonic device in
several ways.NOT_SEPARATE (and the not-yet-implemented STR_STRIP)
modes for strip. They're easy to implement yourself, and this
removes some surface area from the already-too-big
multisplit API.pyproject.toml metadata to make flit happier. This was
necessary to ensure that pip install big also installs its dependencies.released 2022/10/16
lines_filter_contains is now
lines_containing,
and lines_filter_grep is now
lines_grep.released 2022/10/16
text module:
lines_filter_contains,
lines_filter_grep,
and
lines_sort.gently_title
now accepts str or bytes. Also added the apostrophes and
double_quotes arguments.released 2022/10/14
multisplit.
I thought when using keep=AS_PAIRS that it shouldn't ever emit a 2-tuple
containing just empty strings--but on further reflection I've realized that
that's correct. This behavior is now tested and documented, along with
the reasoning behind it.reverse flag to
re_partition.whitespace_without_dos and newlines_without_dos still had the DOS
end-of-line sequence in them! Oops!
whitespace, newlines, and all the variants (utf8_, ascii_,
and _with_dos) exactly match the set of characters Python considers
whitespace and newline characters.released 2022/10/13
itertools module, which so far only contains
PushbackIterator.lines_strip_comments [ed: now lines_strip_line_comments
and
split_quoted_strings
to the
text
module.released 2022/10/13
whitespace should contain the DOS end-of-line
sequence ('\r\n'), as it should be considered a single separator
when splitting etc. I added that, along with whitespace_no_dos,
and naturally utf8_whitespace_no_dos and
ascii_whitespace_no_dos too.released 2022/10/13
A big upgrade!
multisplit,
and added
multistrip
and
multipartition,
collectively called
The multi- family of string functions.
(Thanks to Eric Smith for suggesting
multipartition!
Well, sort of.)
[multisplit](#multisplits-separatorsnone--keepfalse-maxsplit-1-reversefalse-separatefalse-stripfalse)
now supports five (!) keyword-only parameters, allowing the caller
to tune its behavior to an amazing degree.[multisplit](#multisplits-separatorsnone--keepfalse-maxsplit-1-reversefalse-separatefalse-stripfalse)
got its semantics a bit wrong; it was inconsistent and maybe a little buggy.multistrip
is like str.strip but accepts an iterable of
separator strings. It can strip from the left, right, both, or
neither (in which case it does nothing).multipartition
is like str.partition, but accepts an iterable
of separator strings. It can also partition more than once,
and supports reverse=True which causes it to partition from the right
(like str.rpartition).multi functions: whitespace and newlines,
with ascii_ and utf8_ versions of each, and without_dos variants of
all three newlines variants.Scheduler
and
Heap
classes. Scheduler
is a replacement for Python's sched.scheduler class, with a modernized
interface and a major upgrade in functionality. Heap
is an object-oriented interface to Python's heapq module, used by
Scheduler.
These are in their own modules, big.heap and big.scheduler.lines
and all the lines_ modifiers. These are great for writing little text parsers.
For more information, please see the deep-dive on
lines and lines modifier functions.stripped_lines and rstripped_lines from the text module,
as they're superceded by the far superior
lines
family.normalize_whitespace.
Added the separators and replacement parameters,
and added support for bytes objects.count parameter to
re_partition
and
re_rpartition.released 2022/09/12
stripped_lines and rstripped_lines to the text module.len to the TopologicalSorter object.released 2022/09/04
gently_title
and
normalize_whitespace
to the text module.translate_filename_to_exfat
to handle translating ':' in a special way.
If the colon is followed by a space, then the colon is turned into ' -'.
This yields a more natural translation when colons are used in text, e.g.
'xXx: The Return Of Xander Cage' is translated to 'xXx - The Return Of Xander Cage'.
If the colon is not followed by a space, turns the colon into '-'.
This is good for tiresome modern gobbledygook like 'Re:code', which
will now be translated to 'Re-code'.released 2022/06/12
FAQs
The big package is a grab-bag of cool code for use in your programs.
We found that big demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Security News
OpenSSF has published OSPS Baseline, an initiative designed to establish a minimum set of security-related best practices for open source software projects.
Security News
Michigan TypeScript founder Dimitri Mitropoulos implements WebAssembly runtime in TypeScript types, enabling Doom to run after processing 177 terabytes of type definitions.
Security News
vlt's new "reproduce" tool verifies npm packages against their source code, outperforming traditional provenance adoption in the JavaScript ecosystem.