Update C-API to add features needed for numarray compatibility. Output argument added for several functions and clipmode argument added for a few others.

1 files changed, 408 insertions, 55 deletions

diff --git a/numpy/numarray/functions.py b/numpy/numarray/functions.py
index d549f938f..1f358385c 100644
--- a/numpy/numarray/functions.py
+++ b/numpy/numarray/functions.py
@@ -1,71 +1,80 @@
 
 # missing Numarray defined names (in from numarray import *)
-##__all__ = ['ArrayType', 'CLIP', 'ClassicUnpickler', 'Complex32_fromtype',
-##           'Complex64_fromtype', 'ComplexArray', 'EarlyEOFError', 'Error',
-##           'FileSeekWarning', 'MAX_ALIGN', 'MAX_INT_SIZE', 'MAX_LINE_WIDTH',
-##           'MathDomainError', 'NDArray', 'NewArray', 'NewAxis', 'NumArray',
-##           'NumError', 'NumOverflowError', 'PRECISION', 'Py2NumType',
+##__all__ = ['ClassicUnpickler', 'Complex32_fromtype',
+##           'Complex64_fromtype', 'ComplexArray', 'Error',
+##           'MAX_ALIGN', 'MAX_INT_SIZE', 'MAX_LINE_WIDTH',
+##           'NDArray', 'NewArray', 'NumArray',
+##           'NumError', 'PRECISION', 'Py2NumType',
 ##           'PyINT_TYPES', 'PyLevel2Type', 'PyNUMERIC_TYPES', 'PyREAL_TYPES',
-##           'RAISE', 'SLOPPY', 'STRICT', 'SUPPRESS_SMALL', 'SizeMismatchError',
-##           'SizeMismatchWarning', 'SuitableBuffer', 'USING_BLAS',
-##           'UnderflowError', 'UsesOpPriority', 'WARN', 'WRAP', 'all',
-##           'allclose', 'alltrue', 'and_', 'any', 'arange', 'argmax',
-##           'argmin', 'argsort', 'around', 'array2list', 'array_equal',
-##           'array_equiv', 'array_repr', 'array_str', 'arrayprint',
-##           'arrayrange', 'average', 'choose', 'clip',
-##           'codegenerator', 'compress', 'concatenate', 'conjugate',
-##           'copy', 'copy_reg', 'diagonal', 'divide_remainder',
-##           'dotblas', 'e', 'explicit_type', 'flush_caches', 'fromfile',
-##           'fromfunction', 'fromlist', 'fromstring', 'generic',
-##           'genericCoercions', 'genericPromotionExclusions', 'genericTypeRank',
-##           'getShape', 'getTypeObject', 'handleError', 'identity', 'indices',
-##           'info', 'innerproduct', 'inputarray', 'isBigEndian',
-##           'kroneckerproduct', 'lexsort', 'libnumarray', 'libnumeric',
-##           'load', 'make_ufuncs', 'math', 'memory',
-##           'numarrayall', 'numarraycore', 'numerictypes', 'numinclude',
-##           'operator', 'os', 'outerproduct', 'pi', 'put', 'putmask',
-##           'pythonTypeMap', 'pythonTypeRank', 'rank', 'repeat',
-##           'reshape', 'resize', 'round', 'safethread', 'save', 'scalarTypeMap',
-##           'scalarTypes', 'searchsorted', 'session', 'shape', 'sign', 'size',
-##           'sometrue', 'sort', 'swapaxes', 'sys', 'take', 'tcode',
-##           'tensormultiply', 'tname', 'trace', 'transpose', 'typeDict',
-##           'typecode', 'typecodes', 'typeconv', 'types', 'ufunc',
-##           'ufuncFactory', 'value', 'ieeemask', 'cumproduct', 'cumsum',
-##           'nonzero']
-
+##           'SUPPRESS_SMALL',
+##           'SuitableBuffer', 'USING_BLAS',
+##           'UsesOpPriority', 
+##           'codegenerator', 'generic', 'libnumarray', 'libnumeric',
+##           'make_ufuncs', 'memory',
+##           'numarrayall', 'numarraycore', 'numinclude', 'safethread',
+##           'typecode', 'typecodes', 'typeconv', 'ufunc', 'ufuncFactory',
+##           'ieeemask']
 
 __all__ = ['asarray', 'ones', 'zeros', 'array', 'where']
 __all__ += ['vdot', 'dot', 'matrixmultiply', 'ravel', 'indices',
-            'arange', 'concatenate']
+            'arange', 'concatenate', 'all', 'allclose', 'alltrue', 'and_',
+            'any', 'argmax', 'argmin', 'argsort', 'around', 'array_equal',
+            'array_equiv', 'arrayrange', 'array_str', 'array_repr',
+            'array2list', 'average', 'choose', 'CLIP', 'RAISE', 'WRAP',
+            'clip', 'compress', 'concatenate', 'copy', 'copy_reg',
+            'diagonal', 'divide_remainder', 'e', 'explicit_type', 'pi',
+            'flush_caches', 'fromfile', 'os', 'sys', 'STRICT',
+            'SLOPPY', 'WARN', 'EarlyEOFError', 'SizeMismatchError',
+            'SizeMismatchWarning', 'FileSeekWarning', 'fromstring',
+            'fromfunction', 'fromlist', 'getShape', 'getTypeObject',
+            'identity', 'indices', 'info', 'innerproduct', 'inputarray',
+            'isBigEndian', 'kroneckerproduct', 'lexsort', 'math',
+            'operator', 'outerproduct', 'put', 'putmask', 'rank',
+            'repeat', 'reshape', 'resize', 'round', 'searchsorted',
+            'shape', 'size', 'sometrue', 'sort', 'swapaxes', 'take',
+            'tcode', 'tname', 'tensormultiply', 'trace', 'transpose',
+            'types', 'value', 'cumsum', 'cumproduct', 'nonzero'
+            ]
 
-from numpy import dot as matrixmultiply, dot, vdot, ravel, concatenate
+import copy, copy_reg, types
+import os, sys, math, operator
 
-def array(sequence=None, typecode=None, copy=1, savespace=0,
-          type=None, shape=None, dtype=None):
-    dtype = type2dtype(typecode, type, dtype)
-    if sequence is None:
-        if shape is None:
-            return None
-        if dtype is None:
-            dtype = 'l'
-        return N.empty(shape, dtype)
-    arr = N.array(sequence, dtype, copy=copy)
-    if shape is not None:
-        arr.shape = shape
-    return arr
+from numpy import dot as matrixmultiply, dot, vdot, ravel, concatenate, all,\
+     allclose, any, argmax, argmin, around, argsort, array_equal, array_equiv,\
+     array_str, array_repr, average, CLIP, RAISE, WRAP, clip, concatenate, \
+     diagonal, e, pi, fromfunction, indices, inner as innerproduct, nonzero, \
+     outer as outerproduct, kron as kroneckerproduct, lexsort, putmask, rank, \
+     resize, searchsorted, shape, size, sort, swapaxes, trace, transpose
+import numpy as N
 
-def asarray(seq, type=None, typecode=None, dtype=None):
-    if seq is None:
-        return None
-    dtype = type2dtype(typecode, type, dtype)
-    return N.array(seq, dtype, copy=0)
+from numerictypes import typefrom
+
+isBigEndian = sys.byteorder != 'little'
+value = tcode = 'f'
+tname = 'Float32'
 
+#  If dtype is not None, then it is used
+#  If type is not None, then it is used
+#  If typecode is not None then it is used
+#  If use_default is True, then the default
+#   data-type is returned if all are None
+def type2dtype(typecode, type, dtype, use_default=True):
+    if dtype is None:
+        if type is None:
+            if use_default or typecode is not None:
+                dtype = N.dtype(typecode)
+        else:
+            dtype = N.dtype(type)
+    if use_default and dtype is None:
+        dtype = N.dtype(None)
+    return dtype
+    
 def ones(shape, type=None, typecode=None, dtype=None):
-    dtype = type2dtype(typecode, type, dtype)
+    dtype = type2dtype(typecode, type, dtype, 1)
     return N.ones(shape, dtype)
 
 def zeros(shape, type=None, typecode=None, dtype=None):
-    dtype = type2dtype(typecode, type, dtype)
+    dtype = type2dtype(typecode, type, dtype, 1)
     return N.zeros(shape, dtype)
 
 def where(condition, x=None, y=None, out=None):
@@ -83,5 +92,349 @@ def indices(shape, type=None):
 
 def arange(a1, a2=None, stride=1, type=None, shape=None,
            typecode=None, dtype=None):
-    dtype = type2dtype(typecode, type, dtype)
+    dtype = type2dtype(typecode, type, dtype, 0)
     return N.arange(a1, a2, stride, dtype)
+
+arrayrange = arange
+
+def alltrue(x, axis=0):
+    return N.alltrue(x, axis)
+
+def and_(a, b):
+    """Same as a & b
+    """
+    return a & b
+
+def divide_remainder(a, b):
+    a, b = asarray(a), asarray(b)
+    return (a/b,a%b)
+
+def around(array, digits=0, output=None):
+    ret = N.around(array, digits, output)
+    if output is None:
+        return ret
+    return
+    
+def array2list(arr):
+    return arr.tolist()
+
+
+def choose(selector, population, outarr=None, clipmode=RAISE):
+    a = N.asarray(selector)
+    ret = a.choose(population, out=outarr, mode=clipmode)
+    if outarr is None:
+        return ret
+    return
+
+def compress(condition, a, axis=0):
+    return N.compress(condition, a, axis)
+
+# only returns a view
+def explicit_type(a):
+    x = a.view()
+    return x
+
+# stub
+def flush_caches():
+    pass
+
+
+class EarlyEOFError(Exception):
+    "Raised in fromfile() if EOF unexpectedly occurs."
+    pass
+
+class SizeMismatchError(Exception):
+    "Raised in fromfile() if file size does not match shape."
+    pass
+
+class SizeMismatchWarning(Warning):
+    "Issued in fromfile() if file size does not match shape."
+    pass
+
+class FileSeekWarning(Warning):
+    "Issued in fromfile() if there is unused data and seek() fails"
+    pass
+
+
+STRICT, SLOPPY, WARN = range(3)
+
+_BLOCKSIZE=1024
+
+# taken and adapted directly from numarray
+def fromfile(infile, type=None, shape=None, sizing=STRICT,
+             typecode=None, dtype=None):
+    if isinstance(infile, (str, unicode)):
+        infile = open(infile, 'rb')
+    dtype = type2dtype(typecode, type, dtype, True)
+    if shape is None:
+        shape = (-1,)
+    if not isinstance(shape, tuple):
+        shape = (shape,)
+        
+    if (list(shape).count(-1)>1):
+        raise ValueError("At most one unspecified dimension in shape")
+
+    if -1 not in shape:
+        if sizing != STRICT:
+            raise ValueError("sizing must be STRICT if size complete")
+        arr = N.empty(shape, dtype)
+        bytesleft=arr.nbytes
+        bytesread=0
+        while(bytesleft > _BLOCKSIZE):
+            data = infile.read(_BLOCKSIZE)
+            if len(data) != _BLOCKSIZE:
+                raise EarlyEOFError("Unexpected EOF reading data for size complete array")
+            arr.data[bytesread:bytesread+_BLOCKSIZE]=data
+            bytesread += _BLOCKSIZE
+            bytesleft -= _BLOCKSIZE
+        if bytesleft > 0:
+            data = infile.read(bytesleft)
+            if len(data) != bytesleft:
+                raise EarlyEOFError("Unexpected EOF reading data for size complete array")
+            arr.data[bytesread:bytesread+bytesleft]=data
+        return arr
+
+
+    ##shape is incompletely specified
+    ##read until EOF
+    ##implementation 1: naively use memory blocks
+    ##problematic because memory allocation can be double what is
+    ##necessary (!)
+
+    ##the most common case, namely reading in data from an unchanging
+    ##file whose size may be determined before allocation, should be
+    ##quick -- only one allocation will be needed.
+    
+    recsize = dtype.itemsize * N.product([i for i in shape if i != -1])
+    blocksize = max(_BLOCKSIZE/recsize, 1)*recsize
+
+    ##try to estimate file size
+    try:
+        curpos=infile.tell()
+        infile.seek(0,2)
+        endpos=infile.tell()
+        infile.seek(curpos)
+    except (AttributeError, IOError):
+        initsize=blocksize
+    else:
+        initsize=max(1,(endpos-curpos)/recsize)*recsize
+
+    buf = N.newbuffer(initsize)
+
+    bytesread=0
+    while 1:
+        data=infile.read(blocksize)
+        if len(data) != blocksize: ##eof
+            break
+        ##do we have space?
+        if len(buf) < bytesread+blocksize:
+            buf=_resizebuf(buf,len(buf)+blocksize)
+            ## or rather a=resizebuf(a,2*len(a)) ?
+        assert len(buf) >= bytesread+blocksize
+        buf[bytesread:bytesread+blocksize]=data
+        bytesread += blocksize
+
+    if len(data) % recsize != 0:
+        if sizing == STRICT:
+            raise SizeMismatchError("Filesize does not match specified shape")
+        if sizing == WARN:
+            _warnings.warn("Filesize does not match specified shape",
+                           SizeMismatchWarning)
+        try:
+            infile.seek(-(len(data) % recsize),1)
+        except AttributeError:
+            _warnings.warn("Could not rewind (no seek support)",
+                           FileSeekWarning)
+        except IOError:
+            _warnings.warn("Could not rewind (IOError in seek)",
+                           FileSeekWarning)
+    datasize = (len(data)/recsize) * recsize
+    if len(buf) != bytesread+datasize:
+        buf=_resizebuf(buf,bytesread+datasize)
+    buf[bytesread:bytesread+datasize]=data[:datasize]
+    ##deduce shape from len(buf)
+    shape = list(shape)
+    uidx = shape.index(-1)
+    shape[uidx]=len(buf) / recsize
+
+    a = N.ndarray(shape=shape, dtype=type, buffer=buf)
+    if a.dtype.char == '?':
+        N.not_equal(a, 0, a)
+    return a
+    
+def fromstring(datastring, type=None, shape=None, typecode=None, dtype=None):
+    dtype = type2dtype(typecode, type, dtype, True)
+    if shape is None:
+        count = -1
+    else:
+        count = N.product(shape)*dtype.itemsize
+    res = N.fromstring(datastring, count=count)
+    if shape is not None:
+        res.shape = shape
+    return res                
+
+
+# check_overflow is ignored
+def fromlist(seq, type=None, shape=None, check_overflow=0, typecode=None, dtype=None):
+    dtype = type2dtype(typecode, type, dtype, False)
+    return N.array(seq, dtype)
+
+def array(sequence=None, typecode=None, copy=1, savespace=0,
+          type=None, shape=None, dtype=None):
+    dtype = type2dtype(typecode, type, dtype, 0)
+    if sequence is None:
+        if shape is None:
+            return None
+        if dtype is None:
+            dtype = 'l'
+        return N.empty(shape, dtype)
+    if isinstance(sequence, file):
+        return fromfile(sequence, dtype=dtype, shape=shape)
+    if isinstance(sequence, str):
+        return fromstring(sequence, dtype=dtype, shape=shape)
+    if isinstance(sequence, buffer):
+        arr = N.frombuffer(sequence, dtype=dtype)
+    else:
+        arr = N.array(sequence, dtype, copy=copy)
+    if shape is not None:
+        arr.shape = shape
+    return arr
+
+def asarray(seq, type=None, typecode=None, dtype=None):
+    if isinstance(seq, N.ndarray) and type is None and \
+           typecode is None and dtype is None:
+        return seq
+    return array(seq, type=type, typecode=typecode, copy=0, dtype=dtype)
+
+inputarray = asarray
+
+
+def getTypeObject(sequence, type):
+    if type is not None:
+        return type
+    try:
+        return typefrom(N.array(sequence))
+    except:
+        raise TypeError("Can't determine a reasonable type from sequence")
+
+def getShape(shape, *args):
+    try:
+        if shape is () and not args:
+            return ()
+        if len(args) > 0:
+            shape = (shape, ) + args
+        else:
+            shape = tuple(shape)
+        dummy = N.array(shape)
+        if not issubclass(dummy.dtype.type, N.integer):
+            raise TypeError
+        if len(dummy) > N.MAXDIMS:
+            raise TypeError        
+    except:
+        raise TypeError("Shape must be a sequence of integers")
+    return shape
+
+    
+def identity(n, type=None, typecode=None, dtype=None):
+    dtype = type2dtype(typecode, type, dtype, True)
+    return N.identity(n, dtype)
+
+def info(obj):
+    print "class: ", type(obj)
+    print "shape: ", obj.shape
+    print "strides: ", obj.strides
+    print "byteoffset: 0"
+    print "bytestride: ", obj.strides[0]
+    print "itemsize: ", obj.itemsize
+    print "aligned: ", obj.flags.isaligned
+    print "contiguous: ", obj.flags.contiguous
+    print "buffer: ", obj.data
+    print "data pointer:", obj._as_paramater_, "(DEBUG ONLY)"
+    print "byteorder: ",
+    endian = obj.dtype.byteorder
+    if endian in ['|','=']:
+        print sys.byteorder
+    elif endian == '>':
+        print "big"
+    else:
+        print "little"
+    print "byteswap: ", not obj.dtype.isnative
+    print "type: ", typefrom(obj)
+
+#clipmode is ignored if axis is not 0 and array is not 1d
+def put(array, indices, values, axis=0, clipmode=RAISE):
+    if not isinstance(array, N.ndarray):
+        raise TypeError("put only works on subclass of ndarray")
+    work = asarray(array)
+    if axis == 0:
+        if array.ndim == 1:
+            work.put(indices, values, clipmode)
+        else:
+            work[indices] = values
+    elif isinstance(axis, (int, long, N.integer)):
+        work = work.swapaxes(0, axis)
+        work[indices] = values
+        work = work.swapaxes(0, axis)
+    else:
+        def_axes = range(work.ndim)
+        for x in axis:
+            def_axes.remove(x)
+        axis = list(axis)+def_axes
+        work = work.transpose(axis)
+        work[indices] = values
+        work = work.transpose(axis)
+
+def repeat(array, repeats, axis=0):
+    return N.repeat(array, repeats, axis)
+
+
+def reshape(array, shape, *args):
+    if len(args) > 0:
+        shape = (shape,) + args
+    return N.reshape(array, shape)
+
+
+import warnings as _warnings
+def round(*args, **keys):
+    _warnings.warn("round() is deprecated. Switch to around()",
+                   DeprecationWarning)
+    return around(*args, **keys)
+
+def sometrue(array, axis=0):
+    return N.sometrue(array, axis)
+
+#clipmode is ignored if axis is not an integer
+def take(array, indices, axis=0, outarr=None, clipmode=RAISE):
+    array = N.asarray(array)
+    if isinstance(axis, (int, long, N.integer)):
+        res = array.take(indices, axis, outarr, clipmode)
+        if outarr is None:
+            return res
+        return
+    else:
+        def_axes = range(array.ndim)
+        for x in axis:
+            def_axes.remove(x)
+        axis = list(axis) + def_axes
+        work = array.transpose(axis)
+        res = work[indices]
+        if outarr is None:
+            return res
+        out[...] = res
+        return 
+    
+def tensormultiply(a1, a2):
+    a1, a2 = N.asarray(a1), N.asarray(a2)
+    if (a1.shape[-1] != a2.shape[0]):
+        raise ValueError("Unmatched dimensions")
+    shape = a1.shape[:-1] + a2.shape[1:]
+    return N.reshape(dot(N.reshape(a1, (-1, a1.shape[-1])),
+                         N.reshape(a2, (a2.shape[0],-1))),
+                     shape)
+
+def cumsum(a1, axis=0, out=None, type=None, dim=0):
+    return N.asarray(a1).cumsum(axis,dtype=type,out=out)
+
+def cumproduct(a1, axis=0, out=None, type=None, dim=0):
+    return N.asarray(a1).cumprod(axis,dtype=type,out=out)
+