as the very first line of your file, using the pathname for where the Python interpreter is installed on your platform.
If you would like the script to be independent of where the Python interpreter lives, you can use the
env
program. Almost all Unix variants support the following, assuming the Python interpreter is in a directory on the user’s
PATH
:
The minor disadvantage is that this defines the script’s __doc__ string. However, you can fix that by adding
Python comes with two testing frameworks. The
doctest
module finds examples in the docstrings for a module and runs them, comparing the output with the expected output given in the docstring.
The
unittest
module is a fancier testing framework modelled on Java and Smalltalk testing frameworks.
To make testing easier, you should use good modular design in your program. Your program should have almost all functionality encapsulated in either functions or class methods – and this sometimes has the surprising and delightful effect of making the program run faster (because local variable accesses are faster than global accesses). Furthermore the program should avoid depending on mutating global variables, since this makes testing much more difficult to do.
The “global main logic” of your program may be as simple as
if __name__ == "__main__":
main_logic()
at the bottom of the main module of your program.
Once your program is organized as a tractable collection of function and class behaviours, you should write test functions that exercise the behaviours. A test suite that automates a sequence of tests can be associated with each module. This sounds like a lot of work, but since Python is so terse and flexible it’s surprisingly easy. You can make coding much more pleasant and fun by writing your test functions in parallel with the “production code”, since this makes it easy to find bugs and even design flaws earlier.
“Support modules” that are not intended to be the main module of a program may include a self-test of the module.
if __name__ == "__main__":
self_test()
Even programs that interact with complex external interfaces may be tested when the external interfaces are unavailable by using “fake” interfaces implemented in Python.
The
pydoc
module can create HTML from the doc strings in your Python source code. An alternative for creating API documentation purely from docstrings is
epydoc
.
Sphinx
can also include docstring content.
For Unix variants there are several solutions. It’s straightforward to do this using curses, but curses is a fairly large module to learn.
As soon as the main thread exits, all threads are killed. Your main thread is running too quickly, giving the threads no time to do any work.
A simple fix is to add a sleep to the end of the program that’s long enough for all the threads to finish:
import threading, time
def thread_task(name, n):
for i in range(n):
print(name, i)
for i in range(10):
T = threading.Thread(target=thread_task, args=(str(i), i))
T.start()
time.sleep(10) # <---------------------------!
But now (on many platforms) the threads don’t run in parallel, but appear to run sequentially, one at a time! The reason is that the OS thread scheduler doesn’t start a new thread until the previous thread is blocked.
A simple fix is to add a tiny sleep to the start of the run function:
def thread_task(name, n):
time.sleep(0.001) # <--------------------!
for i in range(n):
print(name, i)
for i in range(10):
T = threading.Thread(target=thread_task, args=(str(i), i))
T.start()
time.sleep(10)
Instead of trying to guess a good delay value for
time.sleep()
, it’s better to use some kind of semaphore mechanism. One idea is to use the
queue
module to create a queue object, let each thread append a token to the queue when it finishes, and let the main thread read as many tokens from the queue as there are threads.
The easiest way is to use the
concurrent.futures
module, especially the
ThreadPoolExecutor
类。
Or, if you want fine control over the dispatching algorithm, you can write your own logic manually. Use the
queue
module to create a queue containing a list of jobs. The
Queue
class maintains a list of objects and has a
.put(obj)
method that adds items to the queue and a
.get()
method to return them. The class will take care of the locking necessary to ensure that each job is handed out exactly once.
Here’s a trivial example:
import threading, queue, time
# The worker thread gets jobs off the queue. When the queue is empty, it
# assumes there will be no more work and exits.
# (Realistically workers will run until terminated.)
def worker():
print('Running worker')
time.sleep(0.1)
while True:
try:
arg = q.get(block=False)
except queue.Empty:
print('Worker', threading.current_thread(), end=' ')
print('queue empty')
break
else:
print('Worker', threading.current_thread(), end=' ')
print('running with argument', arg)
time.sleep(0.5)
# Create queue
q = queue.Queue()
# Start a pool of 5 workers
for i in range(5):
t = threading.Thread(target=worker, name='worker %i' % (i+1))
t.start()
# Begin adding work to the queue
for i in range(50):
q.put(i)
# Give threads time to run
print('Main thread sleeping')
time.sleep(5)
When run, this will produce the following output:
Running worker
Running worker
Running worker
Running worker
Running worker
Main thread sleeping
Worker <Thread(worker 1, started 130283832797456)> running with argument 0
Worker <Thread(worker 2, started 130283824404752)> running with argument 1
Worker <Thread(worker 3, started 130283816012048)> running with argument 2
Worker <Thread(worker 4, started 130283807619344)> running with argument 3
Worker <Thread(worker 5, started 130283799226640)> running with argument 4
Worker <Thread(worker 1, started 130283832797456)> running with argument 5
...
Consult the module’s documentation for more details; the
Queue
class provides a featureful interface.
A
全局解释器锁
(GIL) is used internally to ensure that only one thread runs in the Python VM at a time. In general, Python offers to switch among threads only between bytecode instructions; how frequently it switches can be set via
sys.setswitchinterval()
. Each bytecode instruction and therefore all the C implementation code reached from each instruction is therefore atomic from the point of view of a Python program.
In theory, this means an exact accounting requires an exact understanding of the PVM bytecode implementation. In practice, it means that operations on shared variables of built-in data types (ints, lists, dicts, etc) that “look atomic” really are.
For example, the following operations are all atomic (L, L1, L2 are lists, D, D1, D2 are dicts, x, y are objects, i, j are ints):
L.append(x)
L1.extend(L2)
x = L[i]
x = L.pop()
L1[i:j] = L2
L.sort()
x = y
x.field = y
D[x] = y
D1.update(D2)
D.keys()
These aren’t:
i = i+1
L.append(L[-1])
L[i] = L[j]
D[x] = D[x] + 1
Operations that replace other objects may invoke those other objects’
__del__()
method when their reference count reaches zero, and that can affect things. This is especially true for the mass updates to dictionaries and lists. When in doubt, use a mutex!
The
全局解释器锁
(GIL) is often seen as a hindrance to Python’s deployment on high-end multiprocessor server machines, because a multi-threaded Python program effectively only uses one CPU, due to the insistence that (almost) all Python code can only run while the GIL is held.
With the approval of
PEP 703
work is now underway to remove the GIL from the CPython implementation of Python. Initially it will be implemented as an optional compiler flag when building the interpreter, and so separate builds will be available with and without the GIL. Long-term, the hope is to settle on a single build, once the performance implications of removing the GIL are fully understood. Python 3.13 is likely to be the first release containing this work, although it may not be completely functional in this release.
The current work to remove the GIL is based on a
fork of Python 3.9 with the GIL removed
by Sam Gross. Prior to that, in the days of Python 1.5, Greg Stein actually implemented a comprehensive patch set (the “free threading” patches) that removed the GIL and replaced it with fine-grained locking. Adam Olsen did a similar experiment in his
python-safethread
project. Unfortunately, both of these earlier experiments exhibited a sharp drop in single-thread performance (at least 30% slower), due to the amount of fine-grained locking necessary to compensate for the removal of the GIL. The Python 3.9 fork is the first attempt at removing the GIL with an acceptable performance impact.
The presence of the GIL in current Python releases doesn’t mean that you can’t make good use of Python on multi-CPU machines! You just have to be creative with dividing the work up between multiple
processes
rather than multiple
threads
。
ProcessPoolExecutor
class in the new
concurrent.futures
module provides an easy way of doing so; the
multiprocessing
module provides a lower-level API in case you want more control over dispatching of tasks.
Judicious use of C extensions will also help; if you use a C extension to perform a time-consuming task, the extension can release the GIL while the thread of execution is in the C code and allow other threads to get some work done. Some standard library modules such as
zlib
and
hashlib
already do this.
An alternative approach to reducing the impact of the GIL is to make the GIL a per-interpreter-state lock rather than truly global. This was
first implemented in Python 3.12
and is available in the C API. A Python interface to it is expected in Python 3.13. The main limitation to it at the moment is likely to be 3rd party extension modules, since these must be written with multiple interpreters in mind in order to be usable, so many older extension modules will not be usable.
使用
os.remove(filename)
or
os.unlink(filename)
; for documentation, see the
os
module. The two functions are identical;
unlink()
is simply the name of the Unix system call for this function.
To remove a directory, use
os.rmdir()
；使用
os.mkdir()
to create one.
os.makedirs(path)
will create any intermediate directories in
path
that don’t exist.
os.removedirs(path)
will remove intermediate directories as long as they’re empty; if you want to delete an entire directory tree and its contents, use
shutil.rmtree()
.
To rename a file, use
os.rename(old_path, new_path)
.
To truncate a file, open it using
f = open(filename, "rb+")
, and use
f.truncate(offset)
; offset defaults to the current seek position. There’s also
os.ftruncate(fd, offset)
for files opened with
os.open()
，其中
fd
is the file descriptor (a small integer).
The
shutil
module also contains a number of functions to work on files including
copyfile()
,
copytree()
，和
rmtree()
.
The
shutil
module contains a
copyfile()
function. Note that on Windows NTFS volumes, it does not copy
alternate data streams
nor
resource forks
on macOS HFS+ volumes, though both are now rarely used. It also doesn’t copy file permissions and metadata, though using
shutil.copy2()
instead will preserve most (though not all) of it.
To read or write complex binary data formats, it’s best to use the
struct
module. It allows you to take a string containing binary data (usually numbers) and convert it to Python objects; and vice versa.
For example, the following code reads two 2-byte integers and one 4-byte integer in big-endian format from a file:
import struct
with open(filename, "rb") as f:
s = f.read(8)
x, y, z = struct.unpack(">hhl", s)
The ‘>’ in the format string forces big-endian data; the letter ‘h’ reads one “short integer” (2 bytes), and ‘l’ reads one “long integer” (4 bytes) from the string.
For data that is more regular (e.g. a homogeneous list of ints or floats), you can also use the
array
模块。
注意
To read and write binary data, it is mandatory to open the file in binary mode (here, passing
"rb"
to
open()
). If you use
"r"
instead (the default), the file will be open in text mode and
f.read()
将返回
str
objects rather than
bytes
对象。
os.read()
is a low-level function which takes a file descriptor, a small integer representing the opened file.
os.popen()
creates a high-level file object, the same type returned by the built-in
open()
function. Thus, to read
n
bytes from a pipe
p
created with
os.popen()
, you need to use
p.read(n)
.
Python
文件对象
are a high-level layer of abstraction on low-level C file descriptors.
For most file objects you create in Python via the built-in
open()
函数，
f.close()
marks the Python file object as being closed from Python’s point of view, and also arranges to close the underlying C file descriptor. This also happens automatically in
f
’s destructor, when
f
becomes garbage.
But stdin, stdout and stderr are treated specially by Python, because of the special status also given to them by C. Running
sys.stdout.close()
marks the Python-level file object as being closed, but does
not
close the associated C file descriptor.
To close the underlying C file descriptor for one of these three, you should first be sure that’s what you really want to do (e.g., you may confuse extension modules trying to do I/O). If it is, use
os.close()
:
os.close(stdin.fileno())
os.close(stdout.fileno())
os.close(stderr.fileno())
Or you can use the numeric constants 0, 1 and 2, respectively.
Use the standard library module
smtplib
.
Here’s a very simple interactive mail sender that uses it. This method will work on any host that supports an SMTP listener.
import sys, smtplib
fromaddr = input("From: ")
toaddrs = input("To: ").split(',')
print("Enter message, end with ^D:")
msg = ''
while True:
line = sys.stdin.readline()
if not line:
break
msg += line
# The actual mail send
server = smtplib.SMTP('localhost')
server.sendmail(fromaddr, toaddrs, msg)
server.quit()
A Unix-only alternative uses sendmail. The location of the sendmail program varies between systems; sometimes it is
/usr/lib/sendmail
, sometimes
/usr/sbin/sendmail
. The sendmail manual page will help you out. Here’s some sample code:
import os
SENDMAIL = "/usr/sbin/sendmail" # sendmail location
p = os.popen("%s -t -i" % SENDMAIL, "w")
p.write("To: receiver@example.com\n")
p.write("Subject: test\n")
p.write("\n") # blank line separating headers from body
p.write("Some text\n")
p.write("some more text\n")
sts = p.close()
if sts != 0:
print("Sendmail exit status", sts)
The
select
module is commonly used to help with asynchronous I/O on sockets.
To prevent the TCP connect from blocking, you can set the socket to non-blocking mode. Then when you do the
connect()
, you will either connect immediately (unlikely) or get an exception that contains the error number as
.errno
.
errno.EINPROGRESS
indicates that the connection is in progress, but hasn’t finished yet. Different OSes will return different values, so you’re going to have to check what’s returned on your system.
可以使用
connect_ex()
method to avoid creating an exception. It will just return the errno value. To poll, you can call
connect_ex()
again later –
0
or
errno.EISCONN
indicate that you’re connected – or you can pass this socket to
select.select()
to check if it’s writable.
注意
The
asyncio
module provides a general purpose single-threaded and concurrent asynchronous library, which can be used for writing non-blocking network code. The third-party
Twisted
library is a popular and feature-rich alternative.
