What’s new in PyPy 2.0¶
Split rpython and pypy into seperate directories
Callbacks from C are now better JITted
Implement __lenght_hint__ according to PEP 424
Long double support for numpypy
Since r_longdouble support is missing, disable all longdouble and derivative dtypes using ENABLED_LONG_DOUBLE = False
Convert real, imag from ufuncs to views. This involves the beginning of view() functionality
Adds indexing by scalar, adds int conversion from scalar and single element array, fixes compress, indexing by an array with a smaller shape and the indexed object.
Allow concatenation of str and numeric arrays
Improved RPython typing
Rudimentary support for bytearray in RPython
Fix a bug which caused cffi to return the wrong result when calling a C function which calls a Python callback which forces the frames
JIT optimizations which make cffi calls even faster, by removing the need to allocate a temporary buffer where to store the arguments.
Improve documents and straighten out links
Inline the fast path of newarray in the assembler. Disabled on ARM until we fix issues.
Allow dynamic loading of a (Reflex) backend that implements the C-API needed to provide reflection information
Fixed the interaction between two internal tools for controlling the JIT.
Better optimized certain types of frame accesses in the JIT, particularly around exceptions that escape the function they were raised in.
Some missing attributes from ndarrays
Consolidated the lib_pypy/pypy_test and pypy/module/test_lib_pypy tests into one directory for reduced confusion and so they all run nightly.
Add “__pypy__.thread.signals_enabled”, a context manager. Can be used in a non-main thread to enable the processing of signal handlers in that thread.
Support enumerate() over rstr types.
Cleanup _numpypy and numpypy namespaces to more closely resemble numpy.
Random cleanups to hide FlowObjSpace from public view.
Moves optimized JIT frames from stack to heap. As a side effect it enables stackless to work well with the JIT on PyPy. Also removes a bunch of code from the GC which fixes cannot find gc roots.
Documentation fixes after going through the docs at PyCon 2013 sprint.
cffi implementation of _curses
cffi implementation of sqlite3