BUG: Add type promotion support for eval() expressions with many properties

This commit modifies the call to numpy.result_type to get around the
NPY_MAXARGS limit, which at the moment is 32.  Instead of passing a
generator of all types involved in an expression, the type promotion
is done on a pair-wise basis with a call to reduce.

This fixes bugs for code such as the following:

    from numpy.random import randn
    from pandas import DataFrame

    d = DataFrame(randn(10, 2), columns=list('ab'))

    # Evaluates fine
    print(d.eval('*'.join(['a'] * 32)))

    # Fails to evaluate due to NumPy argument limits
    print(d.eval('*'.join(['a'] * 33)))

Author: xenoscopic

doc/source/release.rst

@@ -76,6 +76,7 @@ Bug Fixes
 - Indexing bugs with reordered indexes (:issue:`6252`, :issue:`6254`)
 - Bug in ``.xs`` with a Series multiindex (:issue:`6258`, :issue:`5684`)
 - Bug in conversion of a string types to a DatetimeIndex with a specified frequency (:issue:`6273`, :issue:`6274`)
+- Bug in ``eval`` where type-promotion failed for large expressions (:issue:`6205`)
 
 pandas 0.13.1
 -------------

pandas/computation/align.py

@@ -10,6 +10,7 @@
 import pandas as pd
 from pandas import compat
 import pandas.core.common as com
+from pandas.computation.common import _result_type_many
 
 
 def _align_core_single_unary_op(term):
@@ -85,11 +86,11 @@ def wrapper(terms):
         # only scalars or indexes
         if all(isinstance(term.value, pd.Index) or term.isscalar for term in
                terms):
-            return np.result_type(*term_values), None
+            return _result_type_many(*term_values), None
 
         # no pandas objects
         if not _any_pandas_objects(terms):
-            return np.result_type(*term_values), None
+            return _result_type_many(*term_values), None
 
         return f(terms)
     return wrapper
@@ -199,7 +200,7 @@ def _align(terms):
 
     # if all resolved variables are numeric scalars
     if all(term.isscalar for term in terms):
-        return np.result_type(*(term.value for term in terms)).type, None
+        return _result_type_many(*(term.value for term in terms)).type, None
 
     # perform the main alignment
     typ, axes = _align_core(terms)

pandas/computation/common.py

@@ -1,5 +1,6 @@
 import numpy as np
 import pandas as pd
+from pandas.compat import reduce
 
 
 def _ensure_decoded(s):
@@ -9,5 +10,18 @@ def _ensure_decoded(s):
     return s
 
 
+def _result_type_many(*arrays_and_dtypes):
+    """ wrapper around numpy.result_type which overcomes the NPY_MAXARGS (32)
+    argument limit """
+    try:
+        return np.result_type(*arrays_and_dtypes)
+    except ValueError:
+        # length 0 or length > NPY_MAXARGS both throw a ValueError, so check
+        # which one we're dealing with
+        if len(arrays_and_dtypes) == 0:
+            raise ValueError('at least one array or dtype is required')
+        return reduce(np.result_type, arrays_and_dtypes)
+
+
 class NameResolutionError(NameError):
     pass

pandas/computation/ops.py

@@ -13,7 +13,7 @@
 from pandas.compat import PY3, string_types, text_type
 import pandas.core.common as com
 from pandas.core.base import StringMixin
-from pandas.computation.common import _ensure_decoded
+from pandas.computation.common import _ensure_decoded, _result_type_many
 
 
 _reductions = 'sum', 'prod'
@@ -240,7 +240,7 @@ def return_type(self):
         # clobber types to bool if the op is a boolean operator
         if self.op in (_cmp_ops_syms + _bool_ops_syms):
             return np.bool_
-        return np.result_type(*(term.type for term in com.flatten(self)))
+        return _result_type_many(*(term.type for term in com.flatten(self)))
 
     @property
     def isscalar(self):

pandas/computation/tests/test_eval.py

@@ -1575,6 +1575,18 @@ def test_invalid_numexpr_version():
         yield check_invalid_numexpr_version, engine, parser
 
 
+def check_many_exprs(engine, parser):
+    a = 1
+    expr = ' * '.join('a' * 33)
+    expected = 1
+    res = pd.eval(expr, engine=engine, parser=parser)
+    tm.assert_equal(res, expected)
+
+def test_many_exprs():
+    for engine, parser in ENGINES_PARSERS:
+        yield check_many_exprs, engine, parser
+
+
 if __name__ == '__main__':
     nose.runmodule(argv=[__file__, '-vvs', '-x', '--pdb', '--pdb-failure'],
                    exit=False)