What’s new in PyPy 2.4+¶
Bytearray operations no longer copy the bytearray unnecessarily
Added support for
__len__ to RPython
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 programs.
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.
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).
Optimize the errno handling in the JIT, notably around external function calls. Linux-only.
Remove non-functioning ctypes.pyhonapi and ctypes.PyDLL, document this incompatibility with cpython. Recast sys.dllhandle to an int.
Fix performance regression on ufunc(<scalar>, <scalar>) in numpy.
Update our copies of py.test and pylib to versions 2.5.2 and 1.4.20, respectively.
Classes in the ast module are now distinct from structures used by the compiler.
Upgrades from 2.7.6 to 2.7.8
Fix issue #1861 - cpython compatability madness