Dump the wisdom into a file on disk (or otherwise file-like object). wisdomFile should support the :code:`io.IOBase` or similar int

Called by `setup.py`, it produces the generated code required by the virual machine. Generated files are indicated by the ``*_GENERATED.cpp

This is the equivalent of prepareThreads but for serial mode, where the original self and iter aren't copied.

Initialize threads and allocate space for arrays in global_state

String representation of software versions that NumExpr imports and CPU information.

Run all the tests in the test suite.

Packs a NumReg into a tuple capable of being parsed by the virtual machine function :code:`NumExpr_init()`. Note that numpy.

Return the first location of non-empty NEEDLE within HAYSTACK, or NULL. HAYSTACK_LEN is the minimum known length of HAYSTACK. This method is o

Return the first location of non-empty NEEDLE within HAYSTACK, or NULL. HAYSTACK_LEN is the minimum known length of HAYSTACK. This method is o

Parallel iterator version of VM engine This function fills out the global state and then unlocks the mutexes used to control the threads in module.cpp

A convenience shortcut for :code:`NumExpr.run()`. Keyword arguments are identical.

Pickling magic method. Array references are removed as one can't pickle a weakref, and we don't want to pass full arrays.

Preserves NumExpr object via :code:`pickledBytes = pickle.dumps(neObj)` For pickling, we have to remove the local_dict attribute as

Evaluate a mutli-line expression element-wise, using NumPy broadcasting rules:: neObj = NumExpr('c=2*a+3*b') # Builds a

Restores NumExpr object via `neObj = pickle.loads(pickledBytes)`

Warning! This function has the potential to crash the Python runtime. Do not call it directly. Use the _get_cpu_info_from_cpuid function instead.

AST function handler for an intermediate assignment in a statement block. The convention is that intermediate targets are named targets if t