* Make ``gevent.select.select`` accept arbitrary iterables, not just sequences. That is, you can now pass in a generator of file descriptors instead of a realized list. Internally, arbitrary iterables are copied into lists. This better matches what the standard library does. * On Python 3.11 and newer, opt out of Cython's fast exception manipulation, which *may* be causing problems in certain circumstances when combined with greenlets. * On all versions of Python, adjust some error handling in the default * -based loop. This fixes several assertion failures on debug versions of CPython. Hopefully it has a positive impact under real conditions. * Make ``gevent.pywsgi`` comply more closely with the HTTP specification for chunked transfer encoding. In particular, we are much stricter about trailers, and trailers that are invalid (too long or featuring disallowed characters) forcibly close the connection to the client *after* the results have been sent. * Trailers otherwise continue to be ignored and are not available to the WSGI application. Previously, carefully crafted invalid trailers in chunked requests on keep-alive connections might appear as two requests to ``gevent.pywsgi``. Because this was handled exactly as a normal keep-alive connection with two requests, the WSGI application should handle it normally. However, if you were counting on some upstream server to filter incoming requests based on paths or header fields, and the upstream server simply passed trailers through without validating them, then this embedded second request would OBS-URL: https://build.opensuse.org/package/show/devel:languages:python/python-gevent?expand=0&rev=99
1656 lines
86 KiB
Plaintext
1656 lines
86 KiB
Plaintext
-------------------------------------------------------------------
|
||
Mon Sep 18 19:07:56 UTC 2023 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- update to 23.9.0 (CVE-2023-41419):
|
||
* Make ``gevent.select.select`` accept arbitrary iterables, not
|
||
just sequences. That is, you can now pass in a generator of file
|
||
descriptors instead of a realized list. Internally, arbitrary
|
||
iterables are copied into lists. This better matches what the
|
||
standard library does.
|
||
* On Python 3.11 and newer, opt out of Cython's fast exception
|
||
manipulation, which *may* be causing problems in certain
|
||
circumstances when combined with greenlets.
|
||
* On all versions of Python, adjust some error handling in the
|
||
default * -based loop. This fixes several assertion failures
|
||
on debug versions of CPython. Hopefully it has a positive
|
||
impact under real conditions.
|
||
* Make ``gevent.pywsgi`` comply more closely with the HTTP
|
||
specification for chunked transfer encoding. In particular,
|
||
we are much stricter about trailers, and trailers that are
|
||
invalid (too long or featuring disallowed characters) forcibly
|
||
close the connection to the client *after* the results have
|
||
been sent.
|
||
* Trailers otherwise continue to be ignored and are not
|
||
available to the WSGI application.
|
||
Previously, carefully crafted invalid trailers in chunked
|
||
requests on keep-alive connections might appear as two
|
||
requests to ``gevent.pywsgi``. Because this was handled
|
||
exactly as a normal keep-alive connection with two requests,
|
||
the WSGI application should handle it normally. However, if
|
||
you were counting on some upstream server to filter incoming
|
||
requests based on paths or header fields, and the upstream
|
||
server simply passed trailers through without
|
||
validating them, then this embedded second request would
|
||
bypass those checks.
|
||
(If the upstream server validated that the trailers
|
||
meet the* HTTP specification, this could not occur,
|
||
because characters that are required in an HTTP request,
|
||
like a space, are not allowed in trailers.) CVE-2023-41419
|
||
was reserved for this.
|
||
|
||
-------------------------------------------------------------------
|
||
Mon Aug 14 09:20:19 UTC 2023 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- update to 23.7.0:
|
||
* Add preliminary support for Python 3.12, using greenlet
|
||
3.0a1.
|
||
* Update the bundled c-ares version to 1.19.1.
|
||
* Fix an edge case connecting a non-blocking ``SSLSocket`` that
|
||
could result in an AttributeError. In a change to match
|
||
the standard library, calling ``sock.connect_ex()`` on a
|
||
subclass of ``socket`` no longer calls the subclass's
|
||
``connect`` method.
|
||
* Make gevent's ``FileObjectThread`` (mostly used on Windows)
|
||
implement ``readinto`` cooperatively.
|
||
* Work around an ``AttributeError`` during cyclic garbage
|
||
collection when Python finalizers (``__del__`` and the like)
|
||
attempt to use gevent APIs. This is not a recommended practice,
|
||
and it is unclear if catching this ``AttributeError`` will fix
|
||
any problems or just shift them.
|
||
* Remove support for obsolete Python versions. This is
|
||
everything prior to 3.8.
|
||
* Stop using ``pkg_resources`` to find entry points (plugins).
|
||
Instead, use ``importlib.metadata``.
|
||
* Honor ``sys.unraisablehook`` when a callback function
|
||
produces an exception, and handling the exception in the hub
|
||
*also* produces an exception.
|
||
- drop skip-tests-in-leap.patch handle-python-ssl-changes.patch (obsolete)
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Jun 9 08:08:57 UTC 2023 - Daniel Garcia <daniel.garcia@suse.com>
|
||
|
||
- skip test__util.py in s390x arch
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Jun 1 07:05:01 UTC 2023 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- handle-python-ssl-changes.patch: refresh to handle ssl.shared_ciphers()
|
||
behavior change in python 3.11 as well
|
||
|
||
-------------------------------------------------------------------
|
||
Mon May 15 14:18:03 UTC 2023 - Steve Kowalik <steven.kowalik@suse.com>
|
||
|
||
- Add patch handle-python-ssl-changes.patch:
|
||
* Handle Python 3.10 changes where ssl.shared_ciphers() changes
|
||
behaviour.
|
||
|
||
-------------------------------------------------------------------
|
||
Mon May 15 13:44:48 UTC 2023 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- skip one more test from testsuite
|
||
|
||
-------------------------------------------------------------------
|
||
Thu May 4 20:28:36 UTC 2023 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- update to 22.10.2:
|
||
* Update to greenlet 2.0. This fixes a deallocation issue that
|
||
required a change in greenlet's ABI. The design of greenlet 2.0 is
|
||
intended to prevent future fixes and enhancements from
|
||
requiring an ABI change, making it easier to update gevent
|
||
and greenlet independently.
|
||
|
||
-------------------------------------------------------------------
|
||
Sun Apr 23 23:15:24 UTC 2023 - Matej Cepl <mcepl@suse.com>
|
||
|
||
- Switch documentation to be within the main package.
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Apr 21 12:25:42 UTC 2023 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- add sle15_python_module_pythons (jsc#PED-68)
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Apr 13 22:41:35 UTC 2023 - Matej Cepl <mcepl@suse.com>
|
||
|
||
- Make calling of %{sle15modernpython} optional.
|
||
|
||
-------------------------------------------------------------------
|
||
Wed Mar 8 15:30:19 UTC 2023 - Matej Cepl <mcepl@suse.com>
|
||
|
||
- Clean up the SPEC file.
|
||
|
||
-------------------------------------------------------------------
|
||
Sun Oct 16 17:41:54 UTC 2022 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- update to 22.10.0:
|
||
* Update bundled libuv to 1.44.2.
|
||
See :issue:`1913`.
|
||
* Upgrade embedded c-ares to 1.18.1.
|
||
* Upgrade bundled libuv to 1.42.0 from 1.40.0.
|
||
* Added preliminary support for Python 3.11 (rc2 and later).
|
||
Some platforms may or may not have binary wheels at this time.
|
||
.. important:: Support for legacy versions of Python, including 2.7
|
||
and 3.6, will be ending soon. The
|
||
maintenance burden has become too great and the
|
||
maintainer's time is too limited.
|
||
|
||
Ideally, there will be a release of gevent compatible
|
||
with a final release of greenlet 2.0 that still
|
||
supports those legacy versions, but that may not be
|
||
possible; this may be the final release to support them.
|
||
:class:`gevent.threadpool.ThreadPool` can now optionally expire idle
|
||
threads. This is used by default in the implicit thread pool used for
|
||
DNS requests and other user-submitted tasks; other uses of a
|
||
thread-pool need to opt-in to this.
|
||
See :issue:`1867`.
|
||
* Truly disable the effects of compiling with ``-ffast-math``.
|
||
|
||
-------------------------------------------------------------------
|
||
Mon Dec 13 20:30:19 UTC 2021 - Ben Greiner <code@bnavigator.de>
|
||
|
||
- Update to 21.12.0
|
||
* Fix hanging the interpreter on shutdown if gevent monkey
|
||
patching occurred on a non-main thread in Python 3.9.8 and
|
||
above. (Note that this is not a recommended practice.) See
|
||
:issue:`1839`.
|
||
* Update the embedded c-ares from 1.16.1 to 1.17.1. See
|
||
:issue:`1758`.
|
||
* Add support for Python 3.10rc1 and newer. As part of this, the
|
||
minimum required greenlet version was increased to 1.1.0 (on
|
||
CPython), and the minimum version of Cython needed to build
|
||
gevent from a source checkout is 3.0a9. Note that the dnspython
|
||
resolver is not available on Python 3.10. See :issue:`1790`.
|
||
- Meanwhile Cython 0.29.24 and dnspython are compatible
|
||
with python310
|
||
- Revert threading test skip, fixed in 21.12
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Nov 19 07:48:57 UTC 2021 - Steve Kowalik <steven.kowalik@suse.com>
|
||
|
||
- Skip test__threading_monkey_in_thread as it breaks with Python 3.9.9.
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Jul 8 05:50:58 UTC 2021 - Antonio Larrosa <alarrosa@suse.com>
|
||
|
||
- Skip two tests that fail in SLE/Leap:
|
||
* skip-tests-in-leap.patch
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Feb 12 10:18:37 UTC 2021 - Pedro Monreal <pmonreal@suse.com>
|
||
|
||
- Relax the crypto policies for the test-suite
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Feb 12 10:10:53 UTC 2021 - Pedro Monreal <pmonreal@suse.com>
|
||
|
||
- Update to 21.1.2:
|
||
* Features:
|
||
- Update the embedded libev from 4.31 to 4.33.
|
||
- Update the embedded libuv from 1.38.0 to 1.40.0.
|
||
- Update to 21.1.1:
|
||
* Bugfixes:
|
||
- Fix a TypeError on startup on Python 2 with zope.schema
|
||
installed.
|
||
- Update to 21.1.0:
|
||
* Bugfixes:
|
||
- Make gevent FileObjects more closely match the semantics of
|
||
native file objects for the name attribute.: Objects opened
|
||
from a file descriptor integer have that integer as their
|
||
name. (Note that this is the Python 3 semantics; Python 2
|
||
native file objects returned from os.fdopen() have the string
|
||
"<fdopen>" as their name , but here gevent always follows
|
||
Python 3.) The name remains accessible after the file object
|
||
is closed.
|
||
* Misc:
|
||
- Make gevent.event.AsyncResult print a warning when it detects
|
||
improper cross-thread usage instead of hanging.
|
||
- AsyncResult has never been safe to use from multiple threads.
|
||
It, like most gevent objects, is intended to work with greenlets
|
||
from a single thread. Using AsyncResult from multiple threads
|
||
has undefined semantics. The safest way to communicate between
|
||
threads is using an event loop async watcher.
|
||
- Those undefined semantics changed in recent gevent versions,
|
||
making it more likely that an abused AsyncResult would
|
||
misbehave in ways that could cause the program to hang.
|
||
- Now, when AsyncResult detects a situation that would hang, it
|
||
prints a warning to stderr. Note that this is best-effort, and
|
||
hangs are still possible, especially under PyPy 7.3.3.
|
||
- At the same time, AsyncResult is tuned to behave more like it
|
||
did in older versions, meaning that the hang is once again much
|
||
less likely. If you were getting lucky and using AsyncResult
|
||
successfully across threads, this may restore your luck. In
|
||
addition, cross-thread wakeups are faster. Note that the gevent
|
||
hub now uses an extra file descriptor to implement this.
|
||
- Similar changes apply to gevent.event.Event
|
||
- Update to 20.12.1:
|
||
* Features:
|
||
- Make :class:`gevent.Greenlet` objects function as context
|
||
managers. When the with suite finishes, execution doesn't
|
||
continue until the greenlet is finished. This can be a simpler
|
||
alternative to a :class:`gevent.pool.Group` when the lifetime
|
||
of greenlets can be lexically scoped.
|
||
* Bugfixes:
|
||
- Make gevent's Semaphore objects properly handle native thread
|
||
identifiers larger than can be stored in a C long on Python 3,
|
||
instead of raising an OverflowError.
|
||
- Rebase fix-no-return-in-nonvoid-function.patch
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Feb 9 21:32:04 UTC 2021 - Dirk Müller <dmueller@suse.com>
|
||
|
||
- update to 20.12.0:
|
||
* Make worker threads created by :class:`gevent.threadpool.ThreadPool` install
|
||
the :func:`threading.setprofile` and :func:`threading.settrace` hooks
|
||
while tasks are running. This provides visibility to profiling and
|
||
tracing tools like yappi.
|
||
* Incorrectly passing an exception *instance* instead of an exception
|
||
*type* to `gevent.Greenlet.kill` or `gevent.killall` no longer prints
|
||
an exception to stderr.
|
||
* Make destroying a hub try harder to more forcibly stop loop processing
|
||
when there are outstanding callbacks or IO operations scheduled.
|
||
* Improve the ability to use monkey-patched locks, and
|
||
`gevent.lock.BoundedSemaphore`, across threads, especially when the
|
||
various threads might not have a gevent hub or any other active
|
||
greenlets. In particular, this handles some cases that previously
|
||
raised ``LoopExit`` or would hang. Note that this may not be reliable
|
||
on PyPy on Windows; such an environment is not currently recommended.
|
||
* Make error reporting when a greenlet suffers a `RecursionError` more
|
||
reliable.
|
||
* gevent.pywsgi: Avoid printing an extra traceback ("TypeError: not
|
||
enough arguments for format string") to standard error on certain
|
||
invalid client requests.
|
||
* Add support for PyPy2 7.3.3.
|
||
* Python 2: Make ``gevent.subprocess.Popen.stdin`` objects have a
|
||
``write`` method that guarantees to write the entire argument in
|
||
binary, unbuffered mode. This may require multiple trips around the
|
||
event loop, but more closely matches the behaviour of the Python 2
|
||
standard library (and gevent prior to 1.5). The number of bytes
|
||
written is still returned (instead of ``None``).
|
||
See :issue:`1711`.
|
||
* Make `gevent.pywsgi` stop trying to enforce the rules for reading chunked input or
|
||
``Content-Length`` terminated input when the connection is being
|
||
upgraded, for example to a websocket connection. Likewise, if the
|
||
protocol was switched by returning a ``101`` status, stop trying to
|
||
automatically chunk the responses.
|
||
* Remove the ``__dict__`` attribute from `gevent.socket.socket` objects. The
|
||
standard library socket do not have a ``__dict__``.
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Dec 11 23:52:16 UTC 2020 - Matej Cepl <mcepl@suse.com>
|
||
|
||
- mock dependency was actually not needed at all
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Oct 8 10:03:45 UTC 2020 - Hans-Peter Jansen <hpj@urpla.net>
|
||
|
||
- Disable more tests failing for Python 3.6
|
||
- Don't bother with python2 tests
|
||
|
||
-------------------------------------------------------------------
|
||
Sat Oct 3 16:09:59 UTC 2020 - Hans-Peter Jansen <hpj@urpla.net>
|
||
|
||
- Update to version 20.9.0 (2020-09-22)
|
||
+ Features
|
||
* The embedded libev is now asked to detect the availability of
|
||
clock_gettime and use the realtime and/or monotonic clocks,
|
||
if they are available.
|
||
* On Linux, this can reduce the number of system calls libev
|
||
makes. Originally provided by Josh Snyder. See
|
||
:issue:`issue1648`.
|
||
+ Bugfixes
|
||
* On CPython, depend on greenlet >= 0.4.17. This version is
|
||
binary incompatible with earlier releases on CPython 3.7 and
|
||
later.
|
||
* On Python 3.7 and above, the module gevent.contextvars is no
|
||
longer monkey-patched into the standard library. contextvars
|
||
are now both greenlet and asyncio task local. See
|
||
:issue:`1656`. See :issue:`issue1674`.
|
||
* The DummyThread objects created automatically by certain
|
||
operations when the standard library threading module is
|
||
monkey-patched now match the naming convention the standard
|
||
library uses ("Dummy-12345"). Previously (since gevent 1.2a2)
|
||
they used "DummyThread-12345". See :issue:`1659`.
|
||
* Fix compatibility with dnspython 2.
|
||
* Caution!
|
||
* This currently means that it can be imported. But it cannot
|
||
yet be used. gevent has a pinned dependency on dnspython < 2
|
||
for now.
|
||
* See :issue:`1661`.
|
||
|
||
- Update to version 20.6.2 (2020-06-16)
|
||
+ Features
|
||
* It is now possible to build and use the embedded libuv on a
|
||
Cygwin platform.
|
||
* Note that Cygwin is not an officially supported platform of
|
||
upstream libuv and is not tested by gevent, so the actual
|
||
working status is unknown, and this may bitrot in future
|
||
releases.
|
||
* Thanks to berkakinci for the patch. See :issue:`issue1645`.
|
||
+ Bugfixes
|
||
* Relax the version constraint for psutil on PyPy.
|
||
* Previously it was pinned to 5.6.3 for PyPy2, except for on
|
||
Windows, where it was excluded. It is now treated the same as
|
||
CPython again. See :issue:`issue1643`.
|
||
|
||
- Update to version 20.6.1 (2020-06-10)
|
||
+ Features
|
||
* gevent's CI is now tested on Ubuntu 18.04 (Bionic), an
|
||
upgrade from 16.04 (Xenial). See :issue:`1623`.
|
||
+ Bugfixes
|
||
* On Python 2, the dnspython resolver can be used without
|
||
having selectors2 installed. Previously, an ImportError would
|
||
be raised. See :issue:`issue1641`.
|
||
* Python 3 gevent.ssl.SSLSocket objects no longer attempt to
|
||
catch ConnectionResetError and treat it the same as an
|
||
SSLError with SSL_ERROR_EOF (typically by suppressing it).
|
||
* This was a difference from the way the standard library
|
||
behaved (which is to raise the exception). It was added to
|
||
gevent during early testing of OpenSSL 1.1 and TLS 1.3. See
|
||
:issue:`1637`.
|
||
|
||
- Update to version 20.6.0 (2020-06-06)
|
||
+ Features
|
||
* Add gevent.selectors containing GeventSelector. This selector
|
||
implementation uses gevent details to attempt to reduce
|
||
overhead when polling many file descriptors, only some of
|
||
which become ready at any given time.
|
||
* This is monkey-patched as selectors.DefaultSelector by
|
||
default.
|
||
* This is available on Python 2 if the selectors2 backport is
|
||
installed. (This backport is installed automatically using
|
||
the recommended extra.) When monkey-patching, selectors is
|
||
made available as an alias to this module. See :issue:`1532`.
|
||
* Depend on greenlet >= 0.4.16. This is required for CPython
|
||
3.9 and 3.10a0. See :issue:`1627`.
|
||
* Add support for Python 3.9.
|
||
* No binary wheels are available yet, however. See
|
||
:issue:`1628`.
|
||
+ Bugfixes
|
||
* gevent.socket.create_connection and
|
||
gevent.socket.socket.connect no longer ignore IPv6 scope IDs.
|
||
* Any IP address (IPv4 or IPv6) is no longer subject to an
|
||
extra call to getaddrinfo. Depending on the resolver in use,
|
||
this is likely to change the number and order of greenlet
|
||
switches. (On Windows, in particular test cases when there
|
||
are no other greenlets running, it has been observed to lead
|
||
to LoopExit in scenarios that didn't produce that before.)
|
||
See :issue:`1634`.
|
||
|
||
- Update to version 20.5.2 (2020-05-28)
|
||
+ Bugfixes
|
||
* Forking a process that had use the threadpool to run tasks
|
||
that created their own hub would fail to clean up the
|
||
threadpool by raising greenlet.error. See :issue:`1631`.
|
||
|
||
- Update to version 20.5.1 (2020-05-26)
|
||
+ Features
|
||
* Waiters on Event and Semaphore objects that call wait() or
|
||
acquire(), respectively, that find the Event already set, or
|
||
the Semaphore available, no longer "cut in line" and run
|
||
before any previously scheduled greenlets. They now run in
|
||
the order in which they arrived, just as waiters that had to
|
||
block in those methods do. See :issue:`1520`.
|
||
* Update tested PyPy version from 7.3.0 to 7.3.1 on Linux. See
|
||
:issue:`1569`.
|
||
* Make zope.interface, zope.event and (by extension) setuptools
|
||
required dependencies. The events install extra now does
|
||
nothing and will be removed in 2021. See :issue:`1619`.
|
||
* Update bundled libuv from 1.36.0 to 1.38.0. See
|
||
:issue:`1621`.
|
||
* Update bundled c-ares from 1.16.0 to 1.16.1.
|
||
* On macOS, stop trying to adjust c-ares headers to make them
|
||
universal. See :issue:`1624`.
|
||
+ Bugfixes
|
||
* Make gevent locks that are monkey-patched usually work across
|
||
native threads as well as across greenlets within a single
|
||
thread. Locks that are only used in a single thread do not
|
||
take a performance hit. While cross-thread locking is
|
||
relatively expensive, and not a recommended programming
|
||
pattern, it can happen unwittingly, for example when using
|
||
the threadpool and logging.
|
||
* Before, cross-thread lock uses might succeed, or, if the lock
|
||
was contended, raise greenlet.error. Now, in the contended
|
||
case, if the lock has been acquired by the main thread at
|
||
least once, it should correctly block in any thread,
|
||
cooperating with the event loop of both threads. In certain
|
||
(hopefully rare) cases, it might be possible for contended
|
||
case to raise LoopExit when previously it would have raised
|
||
greenlet.error; if these cases are a practical concern,
|
||
please open an issue.
|
||
* Also, the underlying Semaphore always behaves in an atomic
|
||
fashion (as if the GIL was not released) when PURE_PYTHON is
|
||
set. Previously, it only correctly did so on PyPy. See
|
||
:issue:`issue1437`.
|
||
* Rename gevent's C accelerator extension modules using a
|
||
prefix to avoid clashing with other C extensions. See
|
||
:issue:`1480`.
|
||
* Using gevent.wait on an Event more than once, when that Event
|
||
is already set, could previously raise an AssertionError.
|
||
* As part of this, exceptions raised in the main greenlet will
|
||
now include a more complete traceback from the failing
|
||
greenlet. See :issue:`1540`.
|
||
* Avoid closing the same Python libuv watcher IO object twice.
|
||
Under some circumstances (only seen on Windows), that could
|
||
lead to program crashes. See :issue:`1587`.
|
||
* gevent can now be built using Cython 3.0a5 and newer. The
|
||
PyPI distribution uses this version.
|
||
* The libev extension was incompatible with this. As part of
|
||
this, certain internal, undocumented names have been changed.
|
||
* (Technically, gevent can be built with Cython 3.0a2 and
|
||
above. However, up through 3.0a4 compiling with Cython 3
|
||
results in gevent's test for memory leaks failing. See this
|
||
Cython issue.) See :issue:`1599`.
|
||
* Destroying a hub after joining it didn't necessarily clean up
|
||
all resources associated with the hub, especially if the hub
|
||
had been created in a secondary thread that was exiting. The
|
||
hub and its parent greenlet could be kept alive.
|
||
* Now, destroying a hub drops the reference to the hub and
|
||
ensures it cannot be switched to again. (Though using a new
|
||
blocking API call may still create a new hub.)
|
||
* Joining a hub also cleans up some (small) memory resources
|
||
that might have stuck around for longer before as well. See
|
||
:issue:`1601`.
|
||
* Fix some potential crashes under libuv when using
|
||
gevent.signal_handler. The crashes were seen running the test
|
||
suite and were non-deterministic. See :issue:`1606`.
|
||
|
||
- Update to version 20.5.0 (2020-05-01)
|
||
+ Features
|
||
* Update bundled c-ares to version 1.16.0. Changes. See
|
||
:issue:`1588`.
|
||
* Update all the bundled config.guess and config.sub scripts.
|
||
See :issue:`1589`.
|
||
* Update bundled libuv from 1.34.0 to 1.36.0. See
|
||
:issue:`1597`.
|
||
+ Bugfixes
|
||
* Use ares_getaddrinfo instead of a manual lookup.
|
||
* This requires c-ares 1.16.0.
|
||
* Note that this may change the results, in particular their
|
||
order.
|
||
* As part of this, certain parts of the c-ares extension were
|
||
adapted to use modern Cython idioms.
|
||
* A few minor errors and discrepancies were fixed as well, such
|
||
as gethostbyaddr('localhost') working on Python 3 and failing
|
||
on Python 2. The DNSpython resolver now raises the expected
|
||
TypeError in more cases instead of an AttributeError. See
|
||
:issue:`1012`.
|
||
* The c-ares and DNSPython resolvers now raise exceptions much
|
||
more consistently with the standard resolver. Types and
|
||
errnos are substantially more likely to match what the
|
||
standard library produces.
|
||
* Depending on the system and configuration, results may not
|
||
match exactly, at least with DNSPython. There are still some
|
||
rare cases where the system resolver can raise herror but
|
||
DNSPython will raise gaierror or vice versa. There doesn't
|
||
seem to be a deterministic way to account for this. On PyPy,
|
||
getnameinfo can produce results when CPython raises
|
||
socket.error, and gevent's DNSPython resolver also raises
|
||
socket.error.
|
||
* In addition, several other small discrepancies were
|
||
addressed, including handling of localhost and broadcast host
|
||
names.
|
||
* Note
|
||
* This has been tested on Linux (CentOS and Ubuntu), macOS, and
|
||
Windows. It hasn't been tested on other platforms, so results
|
||
are unknown for them. The c-ares support, in particular, is
|
||
using some additional socket functions and defines. Please
|
||
let the maintainers know if this introduces issues.
|
||
* See :issue:`1459`.
|
||
|
||
- Update to version 20.04.0 (2020-04-22)
|
||
+ Features
|
||
* Let CI (Travis and Appveyor) build and upload release wheels
|
||
for Windows, macOS and manylinux. As part of this, (a subset
|
||
of) gevent's tests can run if the standard library's
|
||
test.support module has been stripped. See :issue:`1555`.
|
||
* Update tested PyPy version from 7.2.0 on Windows to 7.3.1.
|
||
See :issue:`1569`.
|
||
+ Bugfixes
|
||
* Fix a spurious warning about watchers and resource leaks on
|
||
libuv on Windows. Reported by Stéphane Rainville. See
|
||
:issue:`1564`.
|
||
* Make monkey-patching properly remove select.epoll and
|
||
select.kqueue. Reported by Kirill Smelkov. See :issue:`1570`.
|
||
* Make it possible to monkey-patch :mod:`contextvars` before
|
||
Python 3.7 if a non-standard backport that uses the same name
|
||
as the standard library does is installed. Previously this
|
||
would raise an error. Reported by Simon Davy. See
|
||
:issue:`1572`.
|
||
* Fix destroying the libuv default loop and then using the
|
||
default loop again. See :issue:`1580`.
|
||
* libuv loops that have watched children can now exit.
|
||
Previously, the SIGCHLD watcher kept the loop alive even if
|
||
there were no longer any watched children. See :issue:`1581`.
|
||
+ Deprecations and Removals
|
||
* PyPy no longer uses the Python allocation functions for libuv
|
||
and libev allocations. See :issue:`1569`.
|
||
|
||
- Use the system libev by default
|
||
- Remove fix-tests.patch
|
||
- Remove use-libev-cffi.patch
|
||
- Greatly reduce the list of non functional tests
|
||
- Add fix-no-return-in-nonvoid-function.patch, applied for 15.1 and
|
||
below in order to not fail the build
|
||
- Add missing runtime dependencies:
|
||
python-zope.event and python-zope.interface
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Jan 3 11:51:00 CET 2020 - Matej Cepl <mcepl@suse.com>
|
||
|
||
- Use bundled libev library to overcome the current
|
||
incompatibility with libev > 4.25. gh#gevent/gevent#1501
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Jan 2 14:09:44 CET 2020 - Matej Cepl <mcepl@suse.com>
|
||
|
||
- Upgrade to 1.5a3:
|
||
- The file objects (FileObjectPosix, FileObjectThread) now
|
||
consistently text and binary modes. If neither 'b' nor 't' is
|
||
given in the mode, they will read and write native strings.
|
||
If 't' is given, they will always work with unicode strings,
|
||
and 'b' will always work with byte strings. (FileObjectPosix
|
||
already worked this way.) See :issue:`1441`.
|
||
- The file objects accept encoding, errors and newline
|
||
arguments. On Python 2, these are only used if 't' is in the
|
||
mode.
|
||
- The default mode for FileObjectPosix changed from rb to
|
||
simply r, for consistency with the other file objects and the
|
||
standard open and io.open functions.
|
||
- Fix FileObjectPosix improperly being used from multiple
|
||
greenlets. Previously this was hidden by forcing buffering,
|
||
which raised RuntimeError.
|
||
- Fix using monkey-patched threading.Lock and threading.RLock
|
||
objects as spin locks by making them call sleep(0) if they
|
||
failed to acquire the lock in a non-blocking call. This lets
|
||
other callbacks run to release the lock, simulating
|
||
preemptive threading. Using spin locks is not recommended,
|
||
but may have been done in code written for threads,
|
||
especially on Python 3. See :issue:`1464`.
|
||
- Fix Semaphore (and monkey-patched threading locks) to be
|
||
fair. This eliminates the rare potential for starvation of
|
||
greenlets. As part of this change, the low-level method
|
||
rawlink of Semaphore, Event, and AsyncResult now always
|
||
remove the link object when calling it, so unlink can
|
||
sometimes be optimized out. See :issue:`1487`.
|
||
- Make gevent.pywsgi support Connection: keep-alive in
|
||
HTTP/1.0. Based on :pr:`1331` by tanchuhan.
|
||
- Fix a potential crash using gevent.idle() when using libuv.
|
||
See :issue:`1489`.
|
||
- Fix some potential crashes using libuv async watchers.
|
||
- Make ThreadPool consistently raise InvalidThreadUseError when
|
||
spawn is called from a thread different than the thread that
|
||
created the threadpool. This has never been allowed, but was
|
||
inconsistently enforced. On gevent 1.3 and before, this would
|
||
always raise "greenlet error: invalid thread switch," or
|
||
LoopExit. On gevent 1.4, it could raise LoopExit, depending
|
||
on the number of tasks, but still, calling it from
|
||
a different thread was likely to corrupt libev or libuv
|
||
internals.
|
||
- Remove some undocumented, deprecated functions from the
|
||
threadpool module.
|
||
- libuv: Fix a perceived slowness spawning many greenlets at
|
||
the same time without yielding to the event loop while having
|
||
no active IO watchers or timers. If the time spent launching
|
||
greenlets exceeded the switch interval and there were no
|
||
other active watchers, then the default IO poll time of about
|
||
.3s would elapse between spawning batches. This could
|
||
theoretically apply for any non-switching callbacks. This can
|
||
be produced in synthetic benchmarks and other special
|
||
circumstances, but real applications are unlikely to be
|
||
affected. See :issue:`1493`.
|
||
- Fix using the threadpool inside a script or module run with
|
||
python -m gevent.monkey. Previously it would use greenlets
|
||
instead of native threads. See :issue:`1484`.
|
||
- Fix potential crashes in the FFI backends if a watcher was
|
||
closed and stopped in the middle of a callback from the event
|
||
loop and then raised an exception. This could happen if the
|
||
hub's handle_error function was poorly customized, for
|
||
example. See :issue:`1482`
|
||
- Make gevent.killall stop greenlets from running that hadn't
|
||
been run yet. This make it consistent with Greenlet.kill().
|
||
See :issue:`1473` reported by kochelmonster.
|
||
- Make gevent.spawn_raw set the loop attribute on returned
|
||
greenlets. This lets them work with more gevent APIs, notably
|
||
gevent.killall(). They already had dictionaries, but this may
|
||
make them slightly larger, depending on platform (on CPython
|
||
2.7 through 3.6 there is no apparent difference for one
|
||
attribute but on CPython 3.7 and 3.8 dictionaries are
|
||
initially empty and only allocate space once an attribute is
|
||
added; they're still smaller than on earlier versions
|
||
though).
|
||
- Add support for CPython 3.8.0. (Windows wheels are not yet
|
||
available.)
|
||
- Add an --module option to gevent.monkey allowing to run
|
||
a Python module rather than a script. See :pr:`1440`.
|
||
- Improve the way joining the main thread works on Python 3.
|
||
- Implement SSLSocket.verify_client_post_handshake() when
|
||
available.
|
||
- Fix tests when TLS1.3 is supported.
|
||
- Disable Nagle's algorithm in the backdoor server. This can
|
||
improve interactive response time.
|
||
- Test on Python 3.7.4. There are important SSL test fixes.
|
||
- Python version updates: gevent is now tested with CPython
|
||
2.7.16, 3.5.6, 3.6.8, and 3.7.2. It is also tested with PyPy2
|
||
7.1 and PyPy 3.6 7.1 (PyPy 7.0 and 7.1 were not capable of
|
||
running SSL tests on Travis CI).
|
||
- Support for Python 3.4 has been removed, as that version is
|
||
no longer supported uptstream.
|
||
- gevent binary wheels are now manylinux2010 and include libuv
|
||
support. pip 19 is needed to install them. See :issue:`1346`.
|
||
- gevent is now compiled with Cython 0.29.6 and cffi 1.12.2.
|
||
- gevent sources include a pyproject.toml file, specifying the
|
||
build requirements and enabling build isolation. pip 18 or
|
||
above is needed to take advantage of this. See :issue:`1180`.
|
||
- libev-cffi: Let the compiler fill in the definition of
|
||
nlink_t for st_nlink in struct stat, instead of trying to
|
||
guess it ourself. Reported in :issue:`1372` by Andreas
|
||
Schwab.
|
||
- Remove the Makefile. Its most useful commands, make clean and
|
||
make distclean, can now be accomplished in a cross-platform
|
||
way using python setup.py clean and python setup.py clean -a,
|
||
respectively. The remainder of the Makefile contained Travis
|
||
CI commands that have been moved to .travis.yml.
|
||
- Deprecate the EMBED and LIBEV_EMBED, etc, build-time
|
||
environment variables. Instead, use GEVENTSETUP_EMBED and
|
||
GEVENTSETUP_EMBED_LIBEV. See :issue:`1402`.
|
||
- The CFFI backends now respect the embed build-time setting.
|
||
This allows building the libuv backend without embedding
|
||
libuv (except on Windows).
|
||
- Support test resources. This allows disabling tests that use
|
||
the network. See :ref:`limiting-test-resource-usage` for
|
||
more.
|
||
- Python 3.7 subprocess: Copy a STARTUPINFO passed as
|
||
a parameter. Contributed by AndCycle in :pr:`1352`.
|
||
- subprocess: WIFSTOPPED and SIGCHLD are now handled for
|
||
determining Popen.returncode. See
|
||
https://bugs.python.org/issue29335
|
||
- subprocess: No longer close redirected FDs if they are in
|
||
pass_fds. This is a bugfix from Python 3.7 applied to all
|
||
versions gevent runs on.
|
||
- Fix certain operations on a Greenlet in an invalid state
|
||
(with an invalid parent) to raise a TypeError sooner rather
|
||
than an AttributeError later. This is also slightly faster on
|
||
CPython with Cython. Inspired by :issue:`1363` as reported by
|
||
Carson Ip. This means that some extreme corner cases that
|
||
might have passed by replacing a Greenlet's parent with
|
||
something that's not a gevent hub now no longer will.
|
||
- Fix: The spawning_stack for Greenlets on CPython should now
|
||
have correct line numbers in more cases. See :pr:`1379`.
|
||
- The result of gevent.ssl.SSLSocket.makefile() can be used as
|
||
a context manager on Python 2.
|
||
- Python 2: If the backport of the _thread_ module from futures
|
||
has already been imported at monkey-patch time, also patch
|
||
this module to be consistent. The pkg_resources package
|
||
imports this, and pkg_resources is often imported early on
|
||
Python 2 for namespace packages, so if futures is installed
|
||
this will likely be the case.
|
||
- Python 2: Avoid a memory leak when an io.BufferedWriter is
|
||
wrapped around a socket. Reported by Damien Tournoud in
|
||
:issue:`1318`.
|
||
- Avoid unbounded memory usage when creating very deep spawn
|
||
trees. Reported in :issue:`1371` by dmrlawson.
|
||
- Win: Make examples/process.py do something useful. See
|
||
:pr:`1378` by Robert Iannucci.
|
||
- Spawning greenlets can be up to 10% faster. See :pr:`1379`.
|
||
- Removed remove-testCongestion.patch which was subsumed in the
|
||
upstream tarball.
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Apr 11 14:20:20 UTC 2019 - Matej Cepl <mcepl@suse.com>
|
||
|
||
- Switch off type_https test as it fails with new Python 2.7.16
|
||
- Clean up the SPEC file.
|
||
|
||
-------------------------------------------------------------------
|
||
Mon Feb 18 12:09:52 UTC 2019 - Tomáš Chvátal <tchvatal@suse.com>
|
||
|
||
- Skip the SSL tests as they just only triggers false positives
|
||
with hope upstream sorts it out someday
|
||
|
||
-------------------------------------------------------------------
|
||
Mon Feb 11 11:02:21 UTC 2019 - Tomáš Chvátal <tchvatal@suse.com>
|
||
|
||
- Switch to pkgconfig requirements as c-ares was renamed between
|
||
SLE12 and SLE15
|
||
|
||
-------------------------------------------------------------------
|
||
Mon Feb 4 11:37:48 UTC 2019 - Antonio Larrosa <alarrosa@suse.com>
|
||
|
||
- Add patches to fix building the package:
|
||
* remove-testCongestion.patch to remove a test that is failing
|
||
due to a timeout
|
||
* fix-tests.patch to fix some tests
|
||
- ssl.OP_NO_COMPRESSION is set by default by ssl.
|
||
- thread_ident can be represented as a negative hex number now,
|
||
so replace the negative sign with the regex too, and not just the number.
|
||
* use-libev-cffi.patch, libev-cext seems to be broken on i586, so
|
||
use libev-cffi by default (also, the gevent documentation mentions
|
||
that upstream will make libev-cffi the default soon).
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Feb 1 10:35:13 UTC 2019 - Tomáš Chvátal <tchvatal@suse.com>
|
||
|
||
- Make sure to skip tests that need network access
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Jan 31 09:42:44 UTC 2019 - Tomáš Chvátal <tchvatal@suse.com>
|
||
|
||
- Version update to 1.4.0:
|
||
* generate with cython 0.29
|
||
* Refactored the gevent test runner and test suite to make them more reusable. In particular, the tests are now run with python -m gevent.tests. See issue #1293.
|
||
* Formatting run info no longer includes gevent.local.local objects that have no value in the greenlet. See issue #1275.
|
||
* Fixed negative length in pywsgi’s Input read functions for non chunked body. Reported in issue #1274 by tzickel.
|
||
* Fix opening files in text mode in CPython 2 on Windows by patching libuv. See issue #1282 reported by wiggin15.
|
||
* gevent now depends on greenlet 0.4.14 or above.
|
||
* gevent.local.local subclasses correctly supports @staticmethod functions. Reported by Brendan Powers in issue #1266.
|
||
- Do NOT bundle c-ares and libev
|
||
|
||
-------------------------------------------------------------------
|
||
Wed Jan 9 08:26:33 UTC 2019 - Tomáš Chvátal <tchvatal@suse.com>
|
||
|
||
- Switch the condition logic to match the previous changelog
|
||
|
||
-------------------------------------------------------------------
|
||
Wed Jan 9 00:10:41 UTC 2019 - Jonathan Brownell <jbrownell@suse.com>
|
||
|
||
- Use "Requires:" instead of "Recommends:" on older Red Hat platforms
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Aug 7 15:22:15 UTC 2018 - toddrme2178@gmail.com
|
||
|
||
- Update to 1.3.5
|
||
* Update the bundled libuv from 1.20.1 to 1.22.0.
|
||
* Test Python 3.7 on Appveyor. Fix the handling of Popen's
|
||
``close_fds`` argument on 3.7.
|
||
* Update Python versions tested on Travis, including PyPy to 6.0. See :issue:`1195`.
|
||
* :mod:`gevent.queue` imports ``_PySimpleQueue`` instead of
|
||
``SimpleQueue`` so that it doesn't block the event loop.
|
||
:func:`gevent.monkey.patch_all` makes this same substitution in
|
||
:mod:`queue`. This fixes issues with
|
||
:class:`concurrent.futures.ThreadPoolExecutor` as well. Reported in
|
||
:issue:`1248` by wwqgtxx and :issue:`1251` by pyld.
|
||
* :meth:`gevent.socket.socket.connect` doesn't pass the port (service)
|
||
to :func:`socket.getaddrinfo` when it resolves an ``AF_INET`` or
|
||
``AF_INET6`` address. (The standard library doesn't either.) This
|
||
fixes an issue on Solaris. Reported in :issue:`1252` by wiggin15.
|
||
* :meth:`gevent.socket.socket.connect` works with more address
|
||
families, notably AF_TIPC, AF_NETLINK, AF_BLUETOOTH, AF_ALG and AF_VSOCK.
|
||
- Update to 1.3.4
|
||
* Be more careful about issuing ``MonkeyPatchWarning`` for ssl
|
||
imports. Now, we only issue it if we detect the one specific
|
||
condition that is known to lead to RecursionError. This may produce
|
||
false negatives, but should reduce or eliminate false positives.
|
||
* Based on measurements and discussion in :issue:`1233`, adjust the
|
||
way :mod:`gevent.pywsgi` generates HTTP chunks. This is intended to
|
||
reduce network overhead, especially for smaller chunk sizes.
|
||
* Additional slight performance improvements in :mod:`gevent.pywsgi`.
|
||
See :pr:`1241`.
|
||
|
||
-------------------------------------------------------------------
|
||
Wed Jun 13 17:58:41 UTC 2018 - toddrme2178@gmail.com
|
||
|
||
- Update to 1.3.3
|
||
* :func:`gevent.sleep` updates the loop's notion of the current time
|
||
before sleeping so that sleep duration corresponds more closely to
|
||
elapsed (wall clock) time. :class:`gevent.Timeout` does the same.
|
||
Reported by champax and FoP in :issue:`1227`.
|
||
* Fix an ``UnboundLocalError`` in SSL servers when wrapping a socket
|
||
throws an error. Reported in :issue:`1236` by kochelmonster.
|
||
- Update to 1.3.2
|
||
* Allow weak refeneces to :class:`gevent.queue.Queue`. Reported in
|
||
:issue:`1217` by githrdw.
|
||
- Update to 1.3.1
|
||
* Allow weak references to :class:`gevent.event.Event`. Reported in
|
||
:issue:`1211` by Matias Guijarro.
|
||
* Fix embedded uses of :func:`gevent.Greenlet.spawn`, especially under
|
||
uwsgi. Reported in :issue:`1212` by Kunal Gangakhedkar.
|
||
* Fix :func:`gevent.os.nb_write` and :func:`gevent.os.nb_read` not
|
||
always closing the IO event they opened in the event of an
|
||
exception. This would be a problem especially for libuv.
|
||
- Update to 1.3.0
|
||
+ Dependencies
|
||
* Cython 0.28.2 is now used to build gevent from a source checkout.
|
||
* The bundled libuv is now 1.19.2, up from 1.18.0.
|
||
* On Windows, CFFI is now a dependency so that the libuv backend
|
||
really can be used by default.
|
||
* Cython 0.28b1 or later is now required to build gevent from a source
|
||
checkout (Cython is *not* required to build a source distribution
|
||
from PyPI).
|
||
* Update c-ares to 1.14.0. See :issue:`1105`.
|
||
* The bundled libuv is now 1.20.1, up from 1.19.2. See :issue:`1177`.
|
||
* gevent now **requires** the patched version of libuv it is
|
||
distributed with. Building gevent with a non-embedded libuv, while
|
||
not previously supported, is not possible now. See
|
||
:issue:`1126`.
|
||
* gevent is now built and tested with Cython 0.27. This is required
|
||
for Python 3.7 support.
|
||
* Update c-ares to 1.13.0. See :issue:`990`.
|
||
+ Platform Support
|
||
* Travis CI tests on Python 3.7.0b3.
|
||
* Windows now defaults to the libuv backend if CFFI is installed. See
|
||
:issue:`1163`.
|
||
* Python 3.7 passes the automated memory leak checks. See :issue:`1197`.
|
||
* Python 3.7.0b4 is now the tested and supported version of Python
|
||
3.7. PyPy 6.0 has been tested, although CI continues to use 5.10.
|
||
* Travis CI tests on Python 3.7.0b2 and PyPy 2.7 5.10.0 and PyPy 3.5
|
||
5.10.1.
|
||
* Add initial support for Python 3.7a3. It has the same level of
|
||
support as Python 3.6.
|
||
> Using unreleased Cython 0.28 and greenlet 0.4.13; requires Python 3.7a3.
|
||
> The ``async`` functions and classes have been renamed to
|
||
``async_`` due to ``async`` becoming a keyword in Python 3.7.
|
||
Aliases are still in place for older versions. See :issue:`1047`.
|
||
* gevent is now tested on Python 3.6.4. This includes the following
|
||
fixes and changes:
|
||
> Errors raised from :mod:`gevent.subprocess` will have a
|
||
``filename`` attribute set.
|
||
> The :class:`threading.Timer` class is now monkey-patched and can
|
||
be joined. Previously on Python 3.4 and above, joining a ``Timer``
|
||
would hang the process.
|
||
> :meth:`gevent.ssl.SSLSocket.unwrap` behaves more like the standard
|
||
library, including returning a SSLSocket and allowing certain
|
||
timeout-related SSL errors to propagate. The added standard
|
||
library tests ``test_ftplib.py`` now passes.
|
||
> :class:`gevent.subprocess.Popen` accepts a "path-like object" for
|
||
the *cwd* parameter on all platforms. Previously this only worked
|
||
on POSIX platforms under Python 3.6. Now it also works on Windows under
|
||
Python 3.6 (as expected) and is backported to all previous versions.
|
||
* Linux CI now tests on PyPy3 3.5-5.9.0, updated from PyPy3 3.5-5.7.1.
|
||
See :issue:`1001`. PyPy2 has been updated to 5.9.0 from 5.7.1,
|
||
Python 2.7 has been updated to 2.7.14 from 2.7.13, Python 3.4 is
|
||
updated to 3.4.7 from 3.4.5, Python 3.5 is now 3.5.4 from 3.5.3, and
|
||
Python 3.6 is now 3.6.4 from 3.6.0.
|
||
* Drop support for Python 3.3. The documentation has only claimed
|
||
support for 3.4+ since gevent 1.2 was released, and only 3.4+ has
|
||
been tested. This merely removes the supporting Trove classifier and
|
||
remaining test code. See :issue:`997`.
|
||
* PyPy is now known to run on Windows using the libuv backend, with
|
||
caveats. See the section on libuv for more information.
|
||
* Due to security concerns, official support for Python 2.7.8 and
|
||
earlier (without a modern SSL implementation) has been dropped.
|
||
These versions are no longer tested with gevent, but gevent can
|
||
still be installed on them. Supporting code will be removed in the
|
||
next major version of gevent. See :issue:`1073`.
|
||
* `gevent.subprocess.Popen` uses ``/proc/self/fd`` (on Linux) or
|
||
``/dev/fd`` (on BSD, including macOS) to find the file descriptors
|
||
to close when ``close_fds`` is true. This matches an optimization
|
||
added to Python 3 (and backports it to Python 2.7), making process
|
||
spawning up to 9 times faster. Also, on Python 3, since Python 3.3
|
||
is no longer supported, we can also optimize the case where
|
||
``close_fds`` is false (not the default), making process spawning up
|
||
to 38 times faster. Initially reported in :issue:`1172` by Ofer Koren.
|
||
+ Bug Fixes
|
||
* :class:`gevent.local.local` subclasses that mix-in ABCs can be instantiated.
|
||
Reported in :issue:`1201` by Bob Jordan.
|
||
* Fix a bug detecting whether we can use the memory monitoring
|
||
features when psutil is not installed.
|
||
* On Python 2, when monkey-patching `threading.Event`, also
|
||
monkey-patch the underlying class, ``threading._Event``. Some code
|
||
may be type-checking for that. See :issue:`1136`.
|
||
* Fix libuv io watchers polling for events that only stopped watchers
|
||
are interested in, reducing CPU usage. Reported in :issue:`1144` by
|
||
wwqgtxx.
|
||
* Fix calling ``shutdown`` on a closed socket. It was raising
|
||
``AttributeError``, now it once again raises the correct
|
||
``socket.error``. Reported in :issue:`1089` by André Cimander.
|
||
* Fix an interpreter crash that could happen if two or more ``loop``
|
||
objects referenced the default event loop and one of them was
|
||
destroyed and then the other one destroyed or (in the libev C
|
||
extension implementation only) deallocated (garbage collected). See
|
||
:issue:`1098`.
|
||
* Fix a race condition in libuv child callbacks. See :issue:`1104`.
|
||
* If a single greenlet created and destroyed many
|
||
:class:`gevent.local.local` objects without ever exiting, there
|
||
would be a leak of the function objects intended to clean up the
|
||
locals after the greenlet exited. Introduce a weak reference to
|
||
avoid that. Reported in :issue:`981` by Heungsub Lee.
|
||
* pywsgi also catches and ignores by default
|
||
:const:`errno.WSAECONNABORTED` on Windows. Initial patch in
|
||
:pr:`999` by Jan van Valburg.
|
||
* :meth:`gevent.subprocess.Popen.communicate` returns the correct type
|
||
of str (not bytes) in universal newline mode under Python 3, or when
|
||
an encoding has been specified. Initial patch in :pr:`939` by
|
||
William Grzybowski.
|
||
* :meth:`gevent.subprocess.Popen.communicate` (and in general,
|
||
accessing ``Popen.stdout`` and ``Popen.stderr``) returns the correct
|
||
type of str (bytes) in universal newline mode under Python 2.
|
||
Previously it always returned unicode strings. Reported in
|
||
:issue:`1039` by Michal Petrucha.
|
||
* :class:`gevent.fileobject.FileObjectPosix` returns native strings in
|
||
universal newline mode on Python 2. This is consistent with what
|
||
:class:`.FileObjectThread` does. See :issue:`1039`.
|
||
* ``socket.send()`` now catches ``EPROTOTYPE`` on macOS to handle a race
|
||
condition during shutdown. Fixed in :pr:`1035` by Jay Oster.
|
||
* :func:`gevent.socket.create_connection` now properly cleans up open
|
||
sockets if connecting or binding raises a :exc:`BaseException` like
|
||
:exc:`KeyboardInterrupt`, :exc:`greenlet.GreenletExit` or
|
||
:exc:`gevent.timeout.Timeout`. Reported in :issue:`1044` by
|
||
kochelmonster.
|
||
+ Enhancements
|
||
* Add additional optimizations for spawning greenlets, making it
|
||
faster than 1.3a2.
|
||
* Use strongly typed watcher callbacks in the libuv CFFI extensions.
|
||
This prevents dozens of compiler warnings.
|
||
* When gevent prints a timestamp as part of an error message, it is
|
||
now in UTC format as specified by RFC3339.
|
||
* Threadpool threads that exit now always destroy their hub (if one
|
||
was created). This prevents some forms of resource leaks (notably
|
||
visible as blocking functions reported by the new monitoring abilities).
|
||
* Hub objects now include the value of their ``name`` attribute in
|
||
their repr.
|
||
* Pools for greenlets and threads have lower overhead, especially for
|
||
``map``. See :pr:`1153`.
|
||
* The undocumented, internal implementation classes ``IMap`` and
|
||
``IMapUnordered`` classes are now compiled with Cython, further
|
||
reducing the overhead of ``[Thread]Pool.imap``.
|
||
* The classes `gevent.event.Event` and `gevent.event.AsyncResult`
|
||
are compiled with Cython for improved performance, as is the
|
||
``gevent.queue`` module and ``gevent.hub.Waiter`` and certain
|
||
time-sensitive parts of the hub itself. Please report any
|
||
compatibility issues.
|
||
* ``python -m gevent.monkey <script>`` accepts more values for
|
||
``<script>``, including paths to packages or compiled bytecode.
|
||
Reported in :issue:`1157` by Eddie Linder.
|
||
* Add a simple event framework for decoupled communication. It uses
|
||
:mod:`zope.event` if that is installed.
|
||
* :mod:`gevent.monkey` has support for plugins in the form of event
|
||
subscribers and setuptools entry points. See :pr:`1158` and
|
||
:issue:`1162`. setuptools must be installed at runtime for its entry
|
||
points to function.
|
||
* Add the ``dnspython`` resolver as a lightweight alternative to
|
||
c-ares. It is generally faster than c-ares and is supported on PyPy.
|
||
c-ares may be deprecated in the future. See :pr:`1088` and
|
||
:issue:`1103`.
|
||
* Add the module :mod:`gevent.time` that can be imported instead of
|
||
:mod:`time`, much like :mod:`gevent.socket` can be imported instead
|
||
of :mod:`socket`. It contains ``gevent.sleep``. This aids
|
||
monkey-patching.
|
||
* Simple subclasses of `gevent.local.local` now have the same
|
||
(substantially improved) performance characteristics of plain
|
||
`gevent.local.local` itself, making them 2 to 3 times faster than
|
||
before. See :pr:`1117`. If there are any compatibility
|
||
problems, please open issues.
|
||
* Add `gevent.util.assert_switches` to build on the monitoring
|
||
functions. Fixes :issue:`1182`.
|
||
* A started monitor thread for the active hub now survives a fork. See
|
||
:issue:`1185`.
|
||
* The greenlet tracer functions used for the various monitoring
|
||
capabilities are now compiled with Cython for substantially lower
|
||
overhead. See :pr:`1190`.
|
||
* libuv now collects all pending watchers and runs their callbacks at
|
||
the end of the loop iteration using UV_RUN_ONCE. This eliminates the
|
||
need to patch libuv to be greenlet-safe. It also means that
|
||
zero-duration timer watchers are actual timer watchers again
|
||
(instead of being turned into check watchers); newly added
|
||
zero-duration timers cannot block the event loop because they won't
|
||
be run until a safe time.
|
||
+ Monitoring and Debugging
|
||
* Introduce the configuration variable
|
||
`gevent.config.track_greenlet_tree` (aka
|
||
``GEVENT_TRACK_GREENLET_TREE``) to allow disabling the greenlet tree
|
||
features for applications where greenlet spawning is performance
|
||
critical. This restores spawning performance to 1.2 levels.
|
||
* Add an optional monitoring thread for each hub. When enabled, this
|
||
thread (by default) looks for greenlets that block the event loop
|
||
for more than 0.1s. You can add your own periodic monitoring
|
||
functions to this thread. Set ``GEVENT_MONITOR_THREAD_ENABLE`` to
|
||
use it, and ``GEVENT_MAX_BLOCKING_TIME`` to configure the blocking
|
||
interval.
|
||
* The monitoring thread emits events when it detects certain
|
||
conditions, like loop blocked or memory limits exceeded.
|
||
* Add settings for monitoring memory usage and emitting events when a
|
||
threshold is exceeded and then corrected. gevent currently supplies
|
||
no policy for what to do when memory exceeds the configured limit.
|
||
``psutil`` must be installed to use this. See :pr:`1150`.
|
||
* Greenlet objects now keep track of their spawning parent greenlet
|
||
and the code location that spawned them, in addition to maintaining
|
||
a "spawn tree local" mapping. This adds some runtime overhead in
|
||
relative terms, but absolute numbers are still relatively small.
|
||
Based on a proposal from PayPal and comments by Mahmoud Hashemi and
|
||
Kurt Rose. See :issue:`755` and :pr:`1115`. As always, feedback is
|
||
appreciated.
|
||
* Greenlet objects now have a `minimal_ident
|
||
<gevent.Greenlet.minimal_ident>` property. It functions
|
||
similarly to ``Thread.ident`` or ``id`` by uniquely identifying the
|
||
greenlet object while it remains alive, and it can be reused after
|
||
the greenlet object is dead. It is different in that it is small and
|
||
sequential. Based on a proposal from PayPal and comments by Mahmoud
|
||
Hashemi and Kurt Rose. See :issue:`755`. As always, feedback is
|
||
appreciated.
|
||
* `gevent.Greenlet` objects now have a `gevent.Greenlet.name`
|
||
attribute that is included in the default repr.
|
||
* Include the values of `gevent.local.local` objects associated with
|
||
each greenlet in `gevent.util.format_run_info`.
|
||
* Add `gevent.util.GreenletTree` to visualize the greenlet tree. This
|
||
is used by `gevent.util.format_run_info`.
|
||
+ Build Changes
|
||
* Fix building from a source distribution (PyPI) without Cython
|
||
installed.
|
||
* When building gevent from a source checkout (*not* a distributed
|
||
source distribution), ``make`` is no longer required and the
|
||
``Makefile`` is not used. Neither is an external ``cython`` command.
|
||
Instead, the ``cythonize`` function is used, as recommended by
|
||
Cython. (The external commands were never required by source
|
||
distributions.) See :issue:`1076`.
|
||
* :class:`gevent.local.local` is compiled with Cython on CPython.
|
||
* The Cython ares 'channel' class is no longer declared to be publicly
|
||
accessible from a named C structure. Doing so caused a conflict with
|
||
the c-ares header files.
|
||
* Update autoconf's config.guess and config.sub to the latest versions
|
||
for c-ares and libev.
|
||
+ Subprocess
|
||
* Make :class:`gevnt.subprocess.Popen` accept the ``restore_signals``
|
||
keyword argument on all versions of Python, and on Python 2 have it
|
||
default to false. It previously defaulted to true on all versions;
|
||
now it only defaults to true on Python 3. The standard library in
|
||
Python 2 does not have this argument and its behaviour with regards
|
||
to signals is undocumented, but there is code known to rely on
|
||
signals not being restored under Python 2. Initial report and patch
|
||
in :pr:`1063` by Florian Margaine.
|
||
* Allow :class:`gevent.subprocess.Popen` to accept the keyword
|
||
arguments ``pass_fds`` and ``start_new_session`` under Python 2.
|
||
They have always had the same default as Python 3, namely an empty
|
||
tuple and false, but now are accessible to Python 2.
|
||
* Support the ``capture_output`` argument added to Python 3.7 in
|
||
:func:`gevent.subprocess.run`.
|
||
+ Configuration
|
||
* Centralize all gevent configuration in an object at
|
||
``gevent.config``, allowing for gevent to be configured through code
|
||
and not *necessarily* environment variables, and also provide a
|
||
centralized place for documentation. See :issue:`1090`.
|
||
> The new ``GEVENT_CORE_CFFI_ONLY`` environment variable has been
|
||
replaced with the pre-existing ``GEVENT_LOOP`` environment
|
||
variable. That variable may take the values ``libev-cext``,
|
||
``libev-cffi``, or ``libuv-cffi``, (or be a list in preference
|
||
order, or be a dotted name; it may also be assigned to an
|
||
object in Python code at ``gevent.config.loop``).
|
||
> The ``GEVENTARES_SERVERS`` environment variable is deprecated in
|
||
favor of ``GEVENT_RESOLVER_SERVERS``. See :issue:`1103`.
|
||
+ Other Changes
|
||
* The internal, undocumented module ``gevent._threading`` has been
|
||
simplified.
|
||
* The internal, undocumented class ``gevent._socket3._fileobject`` has
|
||
been removed. See :issue:`1084`.
|
||
* Simplify handling of the libev default loop and the ``destroy()``
|
||
method. The default loop, when destroyed, can again be requested and
|
||
it will regenerate itself. The default loop is the only one that can
|
||
receive child events.
|
||
* Make :meth:`gevent.socket.socket.sendall` up to ten times faster on
|
||
PyPy3, through the same change that was applied in gevent 1.1b3 for PyPy2.
|
||
* Be more careful about issuing a warning about patching SSL on
|
||
Python 2. See :issue:`1108`.
|
||
* Signal handling under PyPy with libuv is more reliable. See
|
||
:issue:`1112`.
|
||
* The :mod:`gevent.greenlet` module is now compiled with Cython to
|
||
offset any performance decrease due to :issue:`755`. Please open
|
||
issues for any compatibility concerns. See :pr:`1115` and :pr:`1120`.
|
||
* On CPython, allow the pure-Python implementations of
|
||
`gevent.Greenlet`, `gevent.local` and `gevent.lock` to be
|
||
used when the environment variable ``PURE_PYTHON`` is set. This is
|
||
not recommended except for debugging and testing. See :issue:`1118`.
|
||
* :meth:`gevent.select.poll.poll` now interprets a *timeout* of -1 the
|
||
same as a *timeout* of *None* as the standard requires. Previously,
|
||
on libuv this was interpreted the same as a *timeout* of 0. In
|
||
addition, all *timeout* values less than zero are interpreted like
|
||
*None* (as they always were under libev). See :issue:`1127`.
|
||
* Monkey-patching now defaults to patching ``threading.Event``.
|
||
* ``Pool.add`` now accepts ``blocking`` and ``timeout`` parameters,
|
||
which function similarly to their counterparts in ``Semaphore``.
|
||
See :pr:`1032` by Ron Rothman.
|
||
* Defer adjusting the stdlib's list of active threads until
|
||
``threading`` is monkey patched. Previously this was done when
|
||
:mod:`gevent.threading` was imported. That module is documented to
|
||
be used as a helper for monkey patching, so this should generally
|
||
function the same, but some applications ignore the documentation
|
||
and directly import that module anyway.
|
||
A positive consequence is that ``import gevent.threading, threading;
|
||
threading.current_thread()`` will no longer return a DummyThread
|
||
before monkey-patching. Another positive consequence is that PyPy
|
||
will no longer print a ``KeyError`` on exit if
|
||
:mod:`gevent.threading` was imported *without* monkey-patching.
|
||
See :issue:`984`.
|
||
* Specify the Requires-Python metadata for improved installation
|
||
support in certain tools (setuptools v24.2.1 or newer is required).
|
||
See :issue:`995`.
|
||
* Monkey-patching after the :mod:`ssl` module has been imported now
|
||
prints a warning because this can produce ``RecursionError``.
|
||
* :class:`gevent.local.local` objects are now approximately 3.5 times faster
|
||
getting, setting and deleting attributes on PyPy. This involved
|
||
implementing more of the attribute protocols directly. Please open
|
||
an issue if you have any compatibility problems. See :issue:`1020`.
|
||
* :class:`gevent.local.local` is compiled with Cython on CPython. It
|
||
was already 5 to 6 times faster due to the work on :issue:`1020`,
|
||
and compiling it with Cython makes it another 5 to 6 times faster,
|
||
for a total speed up of about 35 times. It is now in the same
|
||
ballpark as the native :class:`threading.local` class. It also uses
|
||
one pointer less memory per object, and one pointer less memory per
|
||
greenlet. See :pr:`1024`.
|
||
* More safely terminate subprocesses on Windows with
|
||
:meth:`gevent.subprocess.Popen.terminate`. Reported in :issue:`1023`
|
||
by Giacomo Debidda.
|
||
* gevent now uses cffi's "extern 'Python'" callbacks. These should be
|
||
faster and more stable. This requires at least cffi 1.4.0. See :issue:`1049`.
|
||
* gevent now approximately tries to stick to a scheduling interval
|
||
when running callbacks, instead of simply running a count of
|
||
callbacks. The interval is determined by
|
||
:func:`gevent.getswitchinterval`. On Python 3, this is the same as
|
||
the thread switch interval. On Python 2, this defaults to 0.005s and
|
||
can be changed with :func:`gevent.setswitchinterval`. This should
|
||
result in more fair "scheduling" of greenlets, especially when
|
||
``gevent.sleep(0)`` or other busy callbacks are in use. The interval
|
||
is checked every 50 callbacks to keep overhead low. See
|
||
:issue:`1072`. With thanks to Arcadiy Ivanov and Antonio Cuni.
|
||
* The long-deprecated and undocumented module ``gevent.wsgi`` was removed.
|
||
+ libuv
|
||
* Add initial *experimental* support for using libuv as a backend
|
||
instead of libev, controlled by setting the environment variable
|
||
``GEVENT_CORE_CFFI_ONLY=libuv`` before importing gevent. This
|
||
suffers a number of limitations compared to libev, notably:
|
||
> libuv support is not available in the manylinux wheels uploaded to
|
||
PyPI. The manylinux specification requires glibc 2.5, while libuv
|
||
requires glibc 2.12. Install from source to access libuv on Linux
|
||
(e.g., pip's ``--no-binary`` option).
|
||
> Timers (such as ``gevent.sleep`` and ``gevent.Timeout``) only
|
||
support a resolution of 1ms (in practice, it's closer to 1.5ms).
|
||
Attempting to use something smaller will automatically increase it
|
||
to 1ms and issue a warning. Because libuv only supports
|
||
millisecond resolution by rounding a higher-precision clock to an
|
||
integer number of milliseconds, timers apparently suffer from more
|
||
jitter.
|
||
> Using negative timeouts may behave differently from libev.
|
||
> libuv blocks delivery of all signals, so signals are handled using
|
||
an (arbitrary) 0.3 second timer. This means that signal handling
|
||
will be delayed by up to that amount, and that the longest the
|
||
event loop can sleep in the operating system's ``poll`` call is
|
||
that amount. Note that this is what gevent does for libev on
|
||
Windows too.
|
||
> libuv only supports one io watcher per file descriptor, whereas
|
||
libev and gevent have always supported many watchers using
|
||
different settings. The libev behaviour is emulated at the Python
|
||
level, but that adds overhead.
|
||
> Looping multiple times and expecting events for the same file
|
||
descriptor to be raised each time without any data being read or
|
||
written (as works with libev) does not appear to work correctly on
|
||
Linux when using ``gevent.select.poll`` or a monkey-patched
|
||
``selectors.PollSelector``.
|
||
> The build system does not support using a system libuv; the
|
||
embedded copy must be used. Using setuptools to compile libuv was
|
||
the most portable method found.
|
||
> If anything unexpected happens, libuv likes to ``abort()`` the
|
||
entire process instead of reporting an error. For example, closing
|
||
a file descriptor it is using in a watcher may cause the entire
|
||
process to be exited.
|
||
> There may be occasional otherwise unexplained and hard to
|
||
duplicate crashes. If you can duplicate a crash, **please** submit
|
||
an issue.
|
||
> This is the only backend that PyPy can use on Windows. As of this
|
||
alpha, there are many known issues with non-blocking sockets
|
||
(e.g., as used by :mod:`asyncore`; see ``test_ftplib.py``) and
|
||
sometimes sockets not getting closed in a timely fashion
|
||
(apparently; see ``test_httpservers.py``) and communicating with
|
||
subprocesses (it always hangs). Help tracking those down would be
|
||
appreciated. Only PyPy2 is tested.
|
||
Other differences include:
|
||
> The order in which timers and other callbacks are invoked may be
|
||
different than in libev. In particular, timers and IO callbacks
|
||
happen in a different order, and timers may easily be off by up to
|
||
half of the supposed 1ms resolution. See :issue:`1057`.
|
||
> Starting a ``timer`` watcher does not update the loop's time by
|
||
default. This is because, unlike libev, a timer callback can cause
|
||
other timer callbacks to be run if they expire because the loop's
|
||
time updated, without cycling the event loop. See :issue:`1057`.
|
||
libev has also been changed to follow this behaviour.
|
||
Also see :issue:`1072`.
|
||
> Timers of zero duration do not necessarily cause the event loop to
|
||
cycle, as they do in libev. Instead, they may be called
|
||
immediately. If zero duration timers are added from other zero
|
||
duration timer callbacks, this can lead the loop to appear to
|
||
hang, as no IO will actually be done.
|
||
To mitigate this issue, ``loop.timer()`` detects attempts to use
|
||
zero duration timers and turns them into a check watcher. check
|
||
watchers do not support the ``again`` method.
|
||
> All watchers (e.g., ``loop.io``) and the ``Timeout`` class have a
|
||
``close`` method that should be called when code is done using the
|
||
object (they also function as context managers and a ``with``
|
||
statement will automatically close them). gevent does this
|
||
internally for sockets, file objects and internal timeouts.
|
||
Neglecting to close an object may result in leaking native
|
||
resources. To debug this, set the environment variables
|
||
``GEVENT_DEBUG=debug`` and ``PYTHONTRACEMALLOC=n`` before starting
|
||
the process.
|
||
The traditional cython-based libev backend will not leak if
|
||
``close`` is not called and will not produce warnings. The
|
||
CFFI-based libev backend will not currently leak but will produce
|
||
warnings. The CFFI-based libuv backend may leak and will produce
|
||
warnings.
|
||
Again, this is extremely experimental and all of it is subject to
|
||
change.
|
||
See :issue:`790` for history and more in-depth discussion.
|
||
+ libev
|
||
* The C extension has been updated to use more modern Cython idioms
|
||
and generate less code for simplicity, faster compilation and better
|
||
cache usage. See :pr:`1077`.
|
||
> Watcher objects may be slightly larger. On a 64-bit platform, a
|
||
typical watcher may be 16 bytes (2 pointers) larger. This is
|
||
offset by slight performance gains.
|
||
> Cython is no longer preprocessed. Certain attributes that were
|
||
previously only defined in certain compilation modes (notably
|
||
LIBEV_EMBED) are now always defined, but will raise
|
||
``AttributeError`` or have a negative value when not available. In
|
||
general these attributes are not portable or documented and are
|
||
not implemented by libuv or the CFFI backend. See :issue:`1076`.
|
||
> Certain private helper functions (``gevent_handle_error``, and part of
|
||
``gevent_call``) are now implemented in Cython instead of C. This
|
||
reduces our reliance on internal undocumented implementation
|
||
details of Cython and Python that could change. See :pr:`1080`.
|
||
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Mar 6 15:26:44 UTC 2018 - aplanas@suse.com
|
||
|
||
- Allows Recommends and Suggest in Fedora
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Feb 27 17:32:29 UTC 2018 - aplanas@suse.com
|
||
|
||
- Recommends only for SUSE
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Jan 23 13:34:36 UTC 2018 - dimstar@opensuse.org
|
||
|
||
- Disable building of the documentation: the make html call fails
|
||
already, and with rpm 4.14, this aborts the entire build.
|
||
- Drop python-Sphinx buildrequires: not needed when not building
|
||
the docs.
|
||
|
||
-------------------------------------------------------------------
|
||
Wed Nov 22 15:29:26 UTC 2017 - jmatejek@suse.com
|
||
|
||
- add greenlet dependency to Python 3 as well (bsc#1055386)
|
||
|
||
-------------------------------------------------------------------
|
||
Sun Nov 12 16:56:47 UTC 2017 - arun@gmx.de
|
||
|
||
- specfile:
|
||
* added CHANGES.rst CONTRIBUTING.rst to %doc
|
||
|
||
- update to version 1.2.2:
|
||
* Testing on Python 3.5 now uses Python 3.5.3 due to SSL
|
||
changes. See :issue:`943`.
|
||
* Linux CI has been updated from Ubuntu 12.04 to Ubuntu 14.04 since
|
||
the former has reached EOL.
|
||
* Linux CI now tests on PyPy2 5.7.1, updated from PyPy2 5.6.0.
|
||
* Linux CI now tests on PyPy3 3.5-5.7.1-beta, updated from PyPy3
|
||
3.3-5.5-alpha.
|
||
* Python 2 sockets are compatible with the SOCK_CLOEXEC flag found
|
||
on Linux. They no longer pass the socket type or protocol to
|
||
getaddrinfo when connect is called. Reported in :issue:`944` by
|
||
Bernie Hackett.
|
||
* Replace optparse module with argparse. See :issue:`947`.
|
||
* Update to version 1.3.1 of tblib to fix :issue:`954`, reported by
|
||
ml31415.
|
||
* Fix the name of the type parameter to
|
||
:func:`gevent.socket.getaddrinfo` to be correct on Python 3. This
|
||
would cause callers using keyword arguments to raise a
|
||
:exc:`TypeError`. Reported in :issue:`960` by js6626069. Likewise,
|
||
correct the argument names for fromfd and socketpair on Python 2,
|
||
although they cannot be called with keyword arguments under
|
||
CPython.
|
||
Note:
|
||
The gethost* functions take different argument names under CPython
|
||
and PyPy. gevent follows the CPython convention, although these
|
||
functions cannot be called with keyword arguments on CPython.
|
||
* The previously-singleton exception objects FileObjectClosed and
|
||
cancel_wait_ex were converted to classes. On Python 3, an
|
||
exception object is stateful, including references to its context
|
||
and possibly traceback, which could lead to objects remaining
|
||
alive longer than intended.
|
||
* Make sure that python -m gevent.monkey <script> runs code in the
|
||
global scope, not the scope of the main function. Fixed in
|
||
:pr:`975` by Shawn Bohrer.
|
||
|
||
-------------------------------------------------------------------
|
||
Wed May 31 19:30:52 UTC 2017 - dmueller@suse.com
|
||
|
||
- adjust buildrequirements for singlespec building on SLE_12
|
||
|
||
-------------------------------------------------------------------
|
||
Wed Apr 19 20:55:27 UTC 2017 - toddrme2178@gmail.com
|
||
|
||
- Update to version 1.2.1
|
||
* CI services now test on 3.6.0.
|
||
* Windows: Provide ``socket.socketpair`` for all Python 3 versions.
|
||
This was added to Python 3.5, but tests were only added in 3.6.
|
||
(For versions older than 3.4 this is a gevent extension.) Previously
|
||
this was not supported on any Python 3 version.
|
||
* Windows: List ``subprocess.STARTUPINFO`` in ``subprocess.__all__``
|
||
for 3.6 compatibility.
|
||
* The ``_DummyThread`` objects created by calling
|
||
:func:`threading.current_thread` from inside a raw
|
||
:class:`greenlet.greenlet` in a system with monkey-patched
|
||
``threading`` now clean up after themselves when the
|
||
greenlet dies (:class:`gevent.Greenlet`-based ``_DummyThreads`` have
|
||
always cleaned up). This requires the use of a :class:`weakref.ref`
|
||
(and may not be timely on PyPy).
|
||
Reported in :issue:`918` by frozenoctobeer.
|
||
* Build OS X wheels with ``-D_DARWIN_FEATURE_CLOCK_GETTIME=0`` for
|
||
compatibility with OS X releases before 10.12 Sierra. Reported by
|
||
Ned Batchelder in :issue:`916`.
|
||
- Update to version 1.2.0
|
||
* The c-ares DNS resolver ignores bad flags to getnameinfo, like the
|
||
system resolver does. Discovered when cleaning up the DNS resolver
|
||
tests to produce more reliable results. See :issue:`774`.
|
||
- Update to version 1.2a2
|
||
* Update libev to version 4.23.
|
||
* Allow the ``MAKE`` environment variable to specify the make command
|
||
on non-Windows systems for ease of development on BSD systems where
|
||
``make`` is BSD make and ``gmake`` is GNU make (gevent requires GNU
|
||
make). See :issue:`888`.
|
||
* Let :class:`gevent.server.StreamServer` accept an ``SSLContext`` on
|
||
Python versions that support it. Added in :pr:`904` by Arcadiy Ivanov.
|
||
- Update to version 1.2a1
|
||
+ Incompatible Changes
|
||
* Support for Python 2.6 has been removed. See :pr:`766`.
|
||
* Remove module ``gevent.coros`` which was replaced by ``gevent.lock``
|
||
and has been deprecated since 1.0b2.
|
||
* The internal implementation modules ``gevent.corecext`` and
|
||
``gevent.corecffi`` have been moved. Please import from
|
||
``gevent.core`` instead; this has always been the only documented place to
|
||
import from.
|
||
+ Libraries and Installation
|
||
* Update libev to version 4.22 (was 4.20).
|
||
* Update tblib to 1.3.0.
|
||
* Update Cython to 0.25 (was 0.23.5).
|
||
* Update c-ares to version 1.12.0 (was 1.10.0) (`release notes <https://c-ares.haxx.se/changelog.html>`_).
|
||
* For the benefit of downstream package maintainers, gevent is now
|
||
tested with c-ares and libev linked dynamically and not embedded
|
||
(i.e., using the system libraries). However, only the versions
|
||
shipped with gevent are tested and known to work.
|
||
* The repository directory layout has been changed to make it easier
|
||
to include third-party dependencies. Likewise, the setup.py script
|
||
has been split to make it easier to build third-party dependencies.
|
||
* PyPy/CFFI: The corecffi native extension is now only built at
|
||
installation time. Previously, if it wasn't available, a build was
|
||
attempted at every import. This could lead to scattered "gevent"
|
||
directories and undependable results.
|
||
* setuptools is now required at build time on all platforms.
|
||
Previously it was only required for Windows and PyPy.
|
||
* POSIX: Don't hardcode ``/bin/sh`` into the configuration command
|
||
line, instead relying on ``sh`` being on the ``PATH``, as
|
||
recommended by `the standard <http://pubs.opengroup.org/onlinepubs/9699919799/utilities/sh.html>`_.
|
||
Fixed in :pr:`809` by Fredrix Fornwall.
|
||
+ Security
|
||
* :mod:`gevent.pywsgi` now checks that the values passed to
|
||
``start_response`` do not contain a carriage return or newline in
|
||
order to prevent HTTP response splitting (header injection), raising
|
||
a :exc:`ValueError` if they do. See :issue:`775`.
|
||
* Incoming headers containing an underscore are no longer placed in
|
||
the WSGI environ. See :issue:`819`.
|
||
* Errors logged by :class:`~gevent.pywsgi.WSGIHandler` no
|
||
longer print the entire WSGI environment by default. This avoids
|
||
possible information disclosure vulnerabilities. Applications can
|
||
also opt-in to a higher security level for the WSGI environment if they
|
||
choose and their frameworks support it. Originally reported
|
||
in :pr:`779` by sean-peters-au and changed in :pr:`781`.
|
||
+ Platforms
|
||
* As mentioned above, Python 2.6 is no longer supported.
|
||
* Python 3.6 is now tested on POSIX platforms. This includes a few
|
||
notable changes:
|
||
* SSLContext.wrap_socket accepts the ``session`` parameter, though
|
||
this parameter isn't useful prior to 3.6.
|
||
* SSLSocket.recv(0) or read(0) returns an empty byte string. This is
|
||
a fix for `Python bug #23804 <http://bugs.python.org/issue23804>`_
|
||
which has also been merged into Python 2.7 and Python 3.5.
|
||
* PyPy3 5.5.0 *alpha* (supporting Python 3.3.5) is now tested and passes the
|
||
test suite. Thanks to btegs for :issue:`866`, and Fabio Utzig for :pr:`826`.
|
||
Note that PyPy3 is not optimized for performance either by the PyPy
|
||
developers or under gevent, so it may be significantly slower than PyPy2.
|
||
+ Stdlib Compatibility
|
||
* The modules :mod:`gevent.os`, :mod:`gevent.signal` and
|
||
:mod:`gevent.select` export all the attributes from their
|
||
corresponding standard library counterpart.
|
||
* Python 2: ``reload(site)`` no longer fails with a ``TypeError`` if
|
||
gevent has been imported. Reported in :issue:`805` by Jake Hilton.
|
||
* Python 2: ``sendall`` on a non-blocking socket could spuriously fail
|
||
with a timeout.
|
||
+ select/poll
|
||
* If :func:`gevent.select.select` is given a negative *timeout*
|
||
argument, raise an exception like the standard library does.
|
||
* If :func:`gevent.select.select` is given closed or invalid
|
||
file descriptors in any of its lists, raise the appropriate
|
||
``EBADF`` exception like the standard library does. Previously,
|
||
libev would tend to return the descriptor as ready. In the worst
|
||
case, this adds an extra system call, but may also reduce latency if
|
||
descriptors are ready at the time of entry.
|
||
* :class:`selectors.SelectSelector` is properly monkey-patched
|
||
regardless of the order of imports. Reported in :issue:`835` by
|
||
Przemysław Węgrzyn.
|
||
* :meth:`gevent.select.poll.unregister` raises an exception if *fd* is not
|
||
registered, like the standard library.
|
||
* :meth:`gevent.select.poll.poll` returns an event with
|
||
``POLLNVAL`` for registered fds that are invalid. Previously it
|
||
would tend to report both read and write events.
|
||
+ File objects
|
||
* ``FileObjectPosix`` exposes the ``read1`` method when in read mode,
|
||
and generally only exposes methods appropriate to the mode it is in.
|
||
* ``FileObjectPosix`` supports a *bufsize* of 0 in binary write modes.
|
||
Reported in :issue:`840` by Mike Lang.
|
||
* Python 3: :meth:`gevent.socket.connect_ex` was letting
|
||
``BlockingIOError`` (and possibly others) get raised instead of
|
||
returning the errno due to the refactoring of the exception
|
||
hierarchy in Python 3.3. Now the errno is returned. Reported in
|
||
:issue:`841` by Dana Powers.
|
||
+ Other Changes
|
||
* :class:`~.Group` and :class:`~.Pool` now return whether
|
||
:meth:`~.Group.join` returned with an empty group. Suggested by Filippo Sironi in
|
||
:pr:`503`.
|
||
* Unhandled exception reports that kill a greenlet now include a
|
||
timestamp. See :issue:`137`.
|
||
* :class:`~.PriorityQueue` now ensures that an initial items list is a
|
||
valid heap. Fixed in :pr:`793` by X.C.Dong.
|
||
* :class:`gevent.hub.signal` (aka :func:`gevent.signal`) now verifies
|
||
that its `handler` argument is callable, raising a :exc:`TypeError`
|
||
if it isn't. Reported in :issue:`818` by Peter Renström.
|
||
* If ``sys.stderr`` has been monkey-patched (not recommended),
|
||
exceptions that the hub reports aren't lost and can still be caught.
|
||
Reported in :issue:`825` by Jelle Smet.
|
||
* The :func:`gevent.os.waitpid` function is cooperative in more
|
||
circumstances. Reported in :issue:`878` by Heungsub Lee.
|
||
* The various ``FileObject`` implementations are more consistent with
|
||
each other. **Note:** Writing to the *io* property of a FileObject should be
|
||
considered deprecated.
|
||
* Timeout exceptions (and other asynchronous exceptions) could cause
|
||
the BackdoorServer to fail to properly manage the
|
||
stdout/stderr/stdin values. Reported with a patch in :pr:`874` by
|
||
stefanmh.
|
||
* The BackDoorServer now tracks spawned greenlets (connections) and
|
||
kills them in its ``stop`` method.
|
||
+ Servers
|
||
* Default to AF_INET6 when binding to all addresses (e.g.,
|
||
""). This supports both IPv4 and IPv6 connections (except on
|
||
Windows). Original change in :pr:`495` by Felix Kaiser.
|
||
* pywsgi/performance: Chunks of data the application returns are no longer copied
|
||
before being sent to the socket when the transfer-encoding is
|
||
chunked, potentially reducing overhead for large responses.
|
||
+ Threads
|
||
* Add :class:`gevent.threadpool.ThreadPoolExecutor` (a
|
||
:class:`concurrent.futures.ThreadPoolExecutor` variant that always
|
||
uses native threads even when the system has been monkey-patched)
|
||
on platforms that have ``concurrent.futures``
|
||
available (Python 3 and Python 2 with the ``futures`` backport
|
||
installed). This is helpful for, e.g., grpc. Reported in
|
||
:issue:`786` by Markus Padourek.
|
||
* Native threads created before monkey-patching threading can now be
|
||
joined. Previously on Python < 3.4, doing so would raise a
|
||
``LoopExit`` error. Reported in :issue:`747` by Sergey Vasilyev.
|
||
+ SSL
|
||
* On Python 2.7.9 and above (more generally, when the SSL backport is
|
||
present in Python 2), :func:`gevent.ssl.get_server_certificate`
|
||
would raise a :exc:`ValueError` if the system wasn't monkey-patched.
|
||
Reported in :issue:`801` by Gleb Dubovik.
|
||
* On Python 2.7.9 and Python 3, closing an SSL socket in one greenlet
|
||
while it's being read from or written to in a different greenlet is
|
||
less likely to raise a :exc:`TypeError` instead of a
|
||
:exc:`ValueError`. Reported in :issue:`800` by Kevin Chen.
|
||
+ subprocess module
|
||
* Setting SIGCHLD to SIG_IGN or SIG_DFL after :mod:`gevent.subprocess`
|
||
had been used previously could not be reversed, causing
|
||
``Popen.wait`` and other calls to hang. Now, if SIGCHLD has been
|
||
ignored, the next time :mod:`gevent.subprocess` is used this will be
|
||
detected and corrected automatically. (This potentially leads to
|
||
issues with :func:`os.popen` on Python 2, but the signal can always
|
||
be reset again. Mixing the low-level process handling calls,
|
||
low-level signal management and high-level use of
|
||
:mod:`gevent.subprocess` is tricky.) Reported in :issue:`857` by
|
||
Chris Utz.
|
||
* ``Popen.kill`` and ``send_signal`` no longer attempt to send signals
|
||
to processes that are known to be exited.
|
||
+ Several backwards compatible updates to the subprocess module have
|
||
been backported from Python 3 to Python 2, making
|
||
:mod:`gevent.subprocess` smaller, easier to maintain and in some cases
|
||
safer.
|
||
* Popen objects can be used as context managers even on Python 2. The
|
||
high-level API functions (``call``, etc) use this for added safety.
|
||
* The :mod:`gevent.subprocess` module now provides the
|
||
:func:`gevent.subprocess.run` function in a cooperative way even
|
||
when the system is not monkey patched, on all supported versions of
|
||
Python. (It was added officially in Python 3.5.)
|
||
* Popen objects save their *args* attribute even on Python 2.
|
||
* :exc:`gevent.subprocess.TimeoutExpired` is defined even on Python 2,
|
||
where it is a subclass of the :exc:`gevent.timeout.Timeout`
|
||
exception; all instances where a ``Timeout`` exception would
|
||
previously be thrown under Python 2 will now throw a
|
||
``TimeoutExpired`` exception.
|
||
* :func:`gevent.subprocess.call` (and ``check_call``) accepts the
|
||
*timeout* keyword argument on Python 2. This is standard on Python
|
||
3, but a gevent extension on Python 2.
|
||
* :func:`gevent.subprocess.check_output` accepts the *timeout* and
|
||
*input* arguments on Python 2. This is standard on Python 3, but a
|
||
gevent extension on Python 2.
|
||
- Implement single-spec version
|
||
- Add rpmlintrc to deal with source files used by cffi.
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Dec 13 14:46:50 UTC 2016 - dmueller@suse.com
|
||
|
||
- update to 1.1.2:
|
||
* Python 2: ``sendall`` on a non-blocking socket could spuriously fail
|
||
with a timeout.
|
||
* If ``sys.stderr`` has been monkey-patched (not recommended),
|
||
exceptions that the hub reports aren't lost and can still be caught.
|
||
Reported in :issue:`825` by Jelle Smet.
|
||
* :class:`selectors.SelectSelector` is properly monkey-patched
|
||
regardless of the order of imports. Reported in :issue:`835` by
|
||
Przemysław Węgrzyn.
|
||
* Python 2: ``reload(site)`` no longer fails with a ``TypeError`` if
|
||
gevent has been imported. Reported in :issue:`805` by Jake Hilton.
|
||
* Nested callbacks that set and clear an Event no longer cause
|
||
``wait`` to return prematurely. Reported in :issue:`771` by Sergey
|
||
Vasilyev.
|
||
* Fix build on Solaris 10. Reported in :issue:`777` by wiggin15.
|
||
* The ``ref`` parameter to :func:`gevent.os.fork_and_watch` was being ignored.
|
||
* Python 3: :class:`gevent.queue.Channel` is now correctly iterable, instead of
|
||
raising a :exc:`TypeError`.
|
||
* Python 3: Add support for :meth:`socket.socket.sendmsg`,
|
||
:meth:`socket.socket.recvmsg` and :meth:`socket.socket.recvmsg_into`
|
||
on platforms where they are defined. Initial :pr:`773` by Jakub
|
||
Klama.
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Mar 10 20:20:38 UTC 2016 - toddrme2178@gmail.com
|
||
|
||
- Update to 1.1.0 (final):
|
||
* See changelog.rst for detailed informations
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Jan 29 09:14:54 UTC 2016 - tchvatal@suse.com
|
||
|
||
- Update to 1.1rc3:
|
||
* See changelog.rst for detailed informations
|
||
|
||
-------------------------------------------------------------------
|
||
Wed Jul 1 12:54:10 UTC 2015 - tchvatal@suse.com
|
||
|
||
- Bump once more to 1.1a1:
|
||
* See changelog.rst for detailed informations
|
||
* The py2.7 compat in 1.0.2 behaves weirdly on openSUSE,
|
||
this release seems way nicer on that front
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Jun 30 08:48:56 UTC 2015 - tchvatal@suse.com
|
||
|
||
- Version bump to 1.0.2:
|
||
- Fix LifoQueue.peek() to return correct element. PR #456. Patch by Christine
|
||
Spang.
|
||
- Upgrade to libev 4.19
|
||
- Remove SSL3 entirely as default TLS protocol
|
||
- Import socket on Windows (closes #459)
|
||
- Fix C90 syntax error (PR #449)
|
||
- Add compatibility with Python 2.7.9's SSL changes. Issue #477.
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Nov 18 14:00:29 UTC 2014 - toddrme2178@gmail.com
|
||
|
||
- Update to 1.0.1
|
||
- Upgrade libev to 4.15. This fixes #361: installation on armv5.
|
||
- Better support for win64. Patch by Alexey Borzenkov.
|
||
- Fix #423: Pool's imap/imap_unordered could hang forever. Based on patch and test by Jianfei Wang.
|
||
- Enable libev's check watchers.
|
||
- Add dummy Event._reset_internal_locks() method. Only relevant if patch_all(Event=True) is enabled (non-default).
|
||
- Backport new testrunner.py from master branch.
|
||
- Removed upstream included libev.patch
|
||
|
||
-------------------------------------------------------------------
|
||
Thu Jun 12 08:47:18 UTC 2014 - schwab@suse.de
|
||
|
||
- libev.patch: fix syntax error in preprocessor conditional
|
||
|
||
-------------------------------------------------------------------
|
||
Sun Dec 15 15:01:04 UTC 2013 - p.drouand@gmail.com
|
||
|
||
- Update to version 1.0
|
||
+ pywsgi: Pass copy of error list instead of direct reference
|
||
+ Ignore the autogenerated doc/gevent.*.rst files.
|
||
+ Fix cythonpp.py on Windows.
|
||
+ Remove gevent.run (use gevent.wait).
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Oct 26 16:23:38 UTC 2012 - douglarek@outlook.com
|
||
|
||
- Update to version 0.13.8:
|
||
|
||
* Fixed issue #80: gevent.httplib failed with RequestFailed errors because timeout was reset to 1s. Patch by Tomasz Prus.
|
||
* core: fix compilation with the latest Cython: remove emit_ifdef/emit_else/emit_endif.
|
||
* Fixed issue #132: gevent.socket.gethostbyname(<unicode>) now does ascii encoding and uses gevent's resolver rather than
|
||
calling built-in resolver. Patch by Alexey Borzenkov.
|
||
|
||
|
||
-------------------------------------------------------------------
|
||
Mon Jun 25 15:19:15 UTC 2012 - saschpe@suse.de
|
||
|
||
- Update to version 0.13.7:
|
||
+ Fixed #94: fallback to buffer if memoryview fails in _get_memory on
|
||
python 2.7.
|
||
+ Fixed #103: ``Queue(None).full()`` returns ``False`` now (previously
|
||
it returned ``True``).
|
||
+ Fixed #112: threading._sleep is not patched.
|
||
+ Fixed #115: _dummy gets unexpected Timeout arg.
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Mar 27 11:41:44 UTC 2012 - saschpe@suse.de
|
||
|
||
- Run testsuite
|
||
- Require python-greenlet
|
||
- Simplify macro usage
|
||
|
||
-------------------------------------------------------------------
|
||
Fri Nov 18 14:17:34 UTC 2011 - saschpe@suse.de
|
||
|
||
- Cython should not be needed, gevent is build as a binary module
|
||
|
||
-------------------------------------------------------------------
|
||
Tue Jul 19 12:23:20 UTC 2011 - saschpe@suse.de
|
||
|
||
- Initial version
|
||
|