whatsnew-head.rst 2.36 KB
Newer Older
Armin Rigo's avatar
Armin Rigo committed
What's new in PyPy 2.4+
Armin Rigo's avatar
Armin Rigo committed

.. this is a revision shortly after release-2.3.x
mattip's avatar
mattip committed
.. startrev: ca9b7cf02cf4

Tyler Wade's avatar
Tyler Wade committed
8 9
.. branch: fix-bytearray-complexity
Bytearray operations no longer copy the bytearray unnecessarily

Tyler Wade's avatar
Tyler Wade committed
11 12
Added support for ``__getitem__``, ``__setitem__``, ``__getslice__``,
``__setslice__``,  and ``__len__`` to RPython
Armin Rigo's avatar
Armin Rigo committed
13 14 15 16 17 18 19 20 21 22 23 24

.. branch: stringbuilder2-perf
Give the StringBuilder a more flexible internal structure, with a
chained list of strings instead of just one string. This make it
more efficient when building large strings, e.g. with cStringIO().

Also, use systematically jit.conditional_call() instead of regular
branches. This lets the JIT make more linear code, at the cost of
forcing a bit more data (to be passed as arguments to
conditional_calls). I would expect the net result to be a slight
slow-down on some simple benchmarks and a speed-up on bigger
Armin Rigo's avatar
Armin Rigo committed
25 26 27 28 29 30 31 32

.. branch: ec-threadlocal
Change the executioncontext's lookup to be done by reading a thread-
local variable (which is implemented in C using '__thread' if
possible, and pthread_getspecific() otherwise). On Linux x86 and
x86-64, the JIT backend has a special optimization that lets it emit
directly a single MOV from a %gs- or %fs-based address. It seems
actually to give a good boost in performance.
Armin Rigo's avatar
Armin Rigo committed
33 34 35 36 37 38 39 40 41

.. branch: fast-gil
A faster way to handle the GIL, particularly in JIT code. The GIL is
now a composite of two concepts: a global number (it's just set from
1 to 0 and back around CALL_RELEASE_GIL), and a real mutex. If there
are threads waiting to acquire the GIL, one of them is actively
checking the global number every 0.1 ms to 1 ms.  Overall, JIT loops
full of external function calls now run a bit faster (if no thread was
started yet), or a *lot* faster (if threads were started already).
Armin Rigo's avatar
Armin Rigo committed
42 43 44 45

.. branch: jit-get-errno
Optimize the errno handling in the JIT, notably around external
function calls. Linux-only.
mattip's avatar
mattip committed
46 47 48

.. branch: disable_pythonapi
Remove non-functioning ctypes.pyhonapi and ctypes.PyDLL, document this
Armin Rigo's avatar
Armin Rigo committed
incompatibility with cpython. Recast sys.dllhandle to an int.
Ronan Lamy's avatar
Ronan Lamy committed
50 51 52

.. branch: scalar-operations
Fix performance regression on ufunc(<scalar>, <scalar>) in numpy.
53 54 55 56

.. branch: pytest-25
Update our copies of py.test and pylib to versions 2.5.2 and 1.4.20, 
57 58 59

.. branch: split-ast-classes
Classes in the ast module are now distinct from structures used by the compiler.