PyPy v7.2.0: release of 2.7, and 3.6

The PyPy team is proud to release the version 7.2.0 of PyPy, which includes two different interpreters:

  • PyPy2.7, which is an interpreter supporting the syntax and the features of Python 2.7 including the stdlib for CPython 2.7.13
  • PyPy3.6: which is an interpreter supporting the syntax and the features of Python 3.6, including the stdlib for CPython 3.6.9.

The interpreters are based on much the same codebase, thus the double release.

With the support of Arm Holdings Ltd. and Crossbar.io, this release supports the 64-bit aarch64 ARM architecture. More about the work and the performance data around this welcome development can be found in the blog post.

This release removes the “beta” tag from PyPy3.6. While there may still be some small corner-case incompatibilities (around the exact error messages in exceptions and the handling of faulty codec errorhandlers) we are happy with the quality of the 3.6 series and are looking forward to working on a Python 3.7 interpreter.

We updated our benchmark runner at https://speed.pypy.org to a more modern machine and updated the baseline python to CPython 2.7.11. Thanks to Baroque Software for maintaining the benchmark runner.

The CFFI-based _ssl module was backported to PyPy2.7 and updated to use cryptography version 2.7. Additionally the _hashlib, and crypt (or _crypt on Python3) modules were converted to CFFI. This has two consequences. End users and packagers can more easily update these libraries for their platform by executing (cd lib_pypy; ../bin/pypy _*_build.py). More significantly, since PyPy itself links to fewer system shared objects (DLLs), on platforms with a single runtime namespace like linux different CFFI and c-extension modules can load different versions of the same shared object into PyPy without collision (issue 2617).

Until downstream providers begin to distribute c-extension builds with PyPy, we have made packages for some common packages available as wheels.

The CFFI backend has been updated to version 1.13.0. We recommend using CFFI rather than c-extensions to interact with C, and cppyy for interacting with C++ code.

Thanks to Anvil, we revived the PyPy Sandbox, which allows total control over a python interpreter’s interactions with the external world.

We implemented a new JSON decoder that is much faster, uses less memory, and uses a JIT-friendly specialized dictionary.

As always, this release is 100% compatible with the previous one and fixed several issues and bugs raised by the growing community of PyPy users. We strongly recommend updating. Many of the fixes are the direct result of end-user bug reports, so please continue reporting issues as they crop up.

You can download the v7.2 releases here:

We would like to thank our donors for the continued support of the PyPy project. If PyPy is not quite good enough for your needs, we are available for direct consulting work.

We would also like to thank our contributors and encourage new people to join the project. PyPy has many layers and we need help with all of them: PyPy and RPython documentation improvements, tweaking popular modules to run on pypy, or general help with making RPython’s JIT even better. Since the previous release, we have accepted contributions from 27 new contributors, thanks for pitching in.

What is PyPy?

PyPy is a very compliant Python interpreter, almost a drop-in replacement for CPython 2.7, 3.6. It’s fast (PyPy and CPython 2.7.x performance comparison) due to its integrated tracing JIT compiler.

We also welcome developers of other dynamic languages to see what RPython can do for them.

This PyPy release supports:

  • x86 machines on most common operating systems (Linux 32/64 bits, Mac OS X 64 bits, Windows 32 bits, OpenBSD, FreeBSD)
  • big- and little-endian variants of PPC64 running Linux,
  • s390x running Linux
  • 64-bit ARM machines running Linux.

Unfortunately at the moment of writing our ARM buildbots are out of service, so for now we are not releasing any binary for the ARM architecture (32 bit), although PyPy does support ARM 32 bit processors.

Changelog

Changes released in v7.1.1

  • Improve performance of u''.append
  • Prevent a crash in zlib when flushing a closed stream
  • Fix a few corner cases when encountering unicode values above 0x110000
  • Teach the JIT how to handle very large constant lists, sets, or dicts
  • Fix building on ARM32 (issue 2984)
  • Fix a bug in register assignment in ARM32
  • Package windows DLLs needed by cffi modules next to the cffi c-extensions (issue 2988)
  • Cleanup and refactor JIT code to remove rpython.jit.metainterp.typesystem
  • Fix memoryviews of ctype structures with padding, (CPython issue 32780)

Changes to Python 3.6 released in v7.1.1

  • On win32, override some errno.E* values that were added to MSVC in v2010 so that errno.E* == errno.WSAE* as in CPython
  • Do the same optimization that CPython does for (1, 2, 3, *a) (but at the AST level)
  • Raise a TypeError when using buffers and unicode such as ''.strip(buffer) and 'a' < buffer
  • Support _overlapped and asyncio on win32
  • Fix an issue where ''.join(list_of_strings) would rarely confuse utf8 and bytes (issue 2997)
  • Fix io.IncrementalNewlineDecoder interaction with \r (issue 3012)

Changes shared across versions

  • Update cffi to 1.13.0
  • Add support for ARM aarch64
  • Many internal changes to the utf-8 processing code, since now unicode strings are stored internally as utf-8. A few corner cases were fixed, and performance bottlenecks were improved. Specifically, issues were fixed with maketrans, strip, comparison with bytearray, use in array.array, join, translate, forrmatting, __int__, str(<int>), startswith, endswith,
  • Reduce the probability of a deadlock when acquiring a semaphore by moving global state changes closer to the actual aquire (issue 2953)
  • Cleanup and refactor parts of the JIT code
  • Cleanup optimizeopt
  • Support the z15 variant of the s390x CPU.
  • Fixes to _ctypes handling of memoryviews
  • Fix a shadowstack overflow when using sys.setrecursionlimit (issue 2722)
  • Fix a bug that prevent memory-tracking in vmprof working on PyPy
  • Improve the speed and memory use of the _pypyjson JSON decoder. The resulting dictionaries that come out of the JSON decoder have faster lookups too
  • struct.unpack of a sliced bytearray exposed a subtle bug where the JIT’s gc_load family of calls must force some lazy code (issue 3014)
  • Remove copystrcontent and copyunicodecontent in the backends. Instead, replace it in rewrite.py with a direct call to memcpy() and a new basic operation, load_effective_address, which the backend can even decide not to implement.
  • Allow 2d indexing in memoryview.__setitem__ (issue 3028)
  • Speed up ‘bytearray += bytes’ and other similar combinations
  • Compute the greatest common divisor of two RPython rbigint instances using Lehmer’s algorithm and use it in the math module
  • Add RFile.closed to mirror standard file behaviour
  • Add a -D pytest option to run tests directly on the host python without any knowlege of PyPy internals. This allows using pypy3 pytest.py ... for a subset of tests (called app-level testing)
  • Accept arguments to subprocess.Popen that are not directly subscriptable (like iterators) (issue 3050)
  • Catch more low-level SocketError exceptions and turn them into app-level exceptions (issue 3049)
  • Fix formatting of a memoryview: b"<%s>" % memoryview(b"X")
  • Correctly wrap the I/O errors we can get when importing modules
  • Fix bad output from JSON with 'skipkeys=True' (issue 3052)
  • Fix compatibility with latest virtualenv HEAD
  • Avoid RuntimeError in repr() of recursive dictviews (CPython issue 18533)
  • Fix for printing after gc.get_objects() (issue 2979)
  • Optimize many fast-paths through utf-8 code when we know it is ascii or no surroagates are present
  • Check for a rare case of someone shrinking a buffer from another thread while using it in a read() variant. One of the issues discovered when reviewing the code for the sandbox.
  • Prevent segfault when slicing array.array with a large step size
  • Support building ncurses on Suse Linux
  • Update statically-linked _ssl OpenSSL to 1.1.0c on darwin
  • Optimize W_TextIOWrapper._readline and ByteBuffer.getslice
  • Fix possible race condition in threading Lock.release() (issue 3072)
  • Make CDLL(None) on win32 raise TypeError
  • Change sys.getfilesystemcodeerors() to 'strict' on win32
  • Update vendored version of pycparser to version 2.19
  • Implement a much faster JSON decoder (3x speedup for large json files, 2x less memory)

C-API (cpyext) and c-extensions

  • Add DateTime_FromTimestamp and Date_FromTimestamp to the PyDateTime_CAPI struct
  • Add constants and macros needed to build opencv2 with PyPy2.7
  • Add more constants to sysconfig`. Set MACOSX_DEPLOYMENT_TARGET for darwin (issue 2994)
  • fix CBuffer.buffer_attach
  • Add _PyDict_GetItemWithError (PyDict_GetItemWithError on Python3)

Python 3.6 only

  • Accept a, b = (*x, 2) (issue 2995)
  • Class methods with the signature def meth(*args, **kwargs) were not adding an implied self argument (issue 2996)
  • Fix handling of __fpath__ (issue 2985)
  • Disable assert when run with -O (issue 3000)
  • codecs.encode, codecs.decode can behave differently than ustr.encode, bytes.decode (issue 3001)
  • Putting pdb.set_trace call in a threaded program did not work (issue 3003)
  • Fix parsing for converting strings with underscore into ints
  • Add memoryview.obj which stores a reference, (issue 3016)
  • Fix datetime.fromtimestamp for win32 (CPython issue 29097)
  • Improve multiprocessing support on win32
  • Support negative offsets in lnotab (issue 2943)
  • Fix leak of file descriptor with _io.FileIO(‘dir/’)
  • Fix float.__round__(None) (issue 3033)
  • Fix for when we should use the Universal Newline mode on Windows for stdin/stdout/stderr (issue 3007)
  • Fix ImportError invalid arguments error wording
  • Ignore GeneratorExit when throwing into the aclose coroutine of an asynchronous generator (CPython issue 35409)
  • Improve the pure-python faulthander module
  • Properly raise an exception when a BlockingIOError exception escapes from W_BufferedReader.readline_w() (issue 3042)
  • Fix a code path only used in zipimport (issue 3034)
  • Update the stdlib to 3.6.9, fix many failing tests
  • Fix handling of __debug__, -O, and sys.flags.optimizeOptimize (CPython issue 27169)
  • Fix raising SystemExit in atexit
  • Fix case where int(<subint>) would go into infinite recursion
  • Don’t ignore fold parameter in (date,)time.replace()
  • Fix logic bug for memoryview.cast (when view.format is not 'B')
  • Implement retry-on-EINTR in fcntl module (CPython issue 35189)
  • Fix handling of 1st argument to hashlib.blake2{b,s}() (CPython issue 33729)
  • Prevent overflow in _hashlib digest() (CPython issue 34922)
  • IOBase.readlines() relies on the iterator protocol instead of calling readline() directly
  • Don’t inherit IS_ABSTRACT flag in classes
  • Reset raw_pos after unwinding the raw stream (CPython issue 32228)
  • Add existing options -b and -d to pypy3 --help text
  • Clean up _codecs error handling code
  • Add support for using stdlib as a zipfile
  • Check return type of __prepare__() (CPython issue 31588)
  • Fix logic in _curses.get_wch (issue 3064)
  • Match CPython exit code when failing to flush stdout/stderr at exit
  • Improve SyntaxError message output
  • Add range.__bool__
  • Add cursor validity check to _sqlite.Cursor.close
  • Improve message when mistakenly using print something in Python3
  • Handle generator exit in athrow() (CPython issue 33786)
  • Support unmarshalling TYPE_INT64 and turn OverflowErrors from marshal.loads into ValueErrors
  • Update _posixsubprocess.c to match CPython (CPython issue 32270)
  • Remove unused _posixsubprocess.cloexec_pipe()
  • Add missing constants to stat and kill _stat (issue 3073)
  • Fix argument handling in select.poll().poll()
  • Raise SyntaxError instead of DeprecationWarning when treating invalid escapes in bytes as errors (CPython issue 28691)
  • Handle locale in time.strftime(). (issue 3079)
  • Fix an issue when calling PyFrame.fset_f_lineno (issue 3066)

Python 3.6 c-API

  • Add PyStructSequence_InitType2, Py_RETURN_NOTIMPLEMENTED, PyGILState_Check, PyUnicode_AsUCS4, PyUnicode_AsUCS4Copy, PyErr_SetFromWindowsErr,
  • Sync the various Py**Flag constants with CPython
  • Allow PyTypeObject with tp_doc=="" (issue 3055)
  • Update pymacro.h to match CPython 3.6.9
  • Support more datetime C functions and definitions