Writing tests for the
numpy.i SWIG interface file is a combinatorial headache. At present, 12 different data types are supported, each with 74 different argument signatures, for a total of 888 typemaps supported “out of the box”. Each of these typemaps, in turn, might require several unit tests in order to verify expected behavior for both proper and improper inputs. Currently, this results in more than 1,000 individual unit tests executed when
make test is run in the
To facilitate this many similar unit tests, some high-level programming techniques are employed, including C and SWIG macros, as well as Python inheritance. The purpose of this document is to describe the testing infrastructure employed to verify that the
numpy.i typemaps are working as expected.
There are three independent testing frameworks supported, for one-, two-, and three-dimensional arrays respectively. For one-dimensional arrays, there are two C++ files, a header and a source, named:
that contain prototypes and code for a variety of functions that have one-dimensional arrays as function arguments. The file:
is a SWIG interface file that defines a python module
Vector that wraps the functions in
Vector.h while utilizing the typemaps in
numpy.i to correctly handle the C arrays.
swig to generate
Vector_wrap.cxx, and also executes the
setup.py script that compiles
Vector_wrap.cxx and links together the extension module
_Vector.dylib, depending on the platform. This extension module and the proxy file
Vector.py are both placed in a subdirectory under the
The actual testing takes place with a Python script named:
that uses the standard Python library module
unittest, which performs several tests of each function defined in
Vector.h for each data type supported.
Two-dimensional arrays are tested in exactly the same manner. The above description applies, but with
Matrix substituted for
Vector. For three-dimensional tests, substitute
Vector. For four-dimensional tests, substitute
Vector. For flat in-place array tests, substitute
Vector. For the descriptions that follow, we will reference the
Vector tests, but the same information applies to
make test will ensure that all of the test software is built and then run all three test scripts.
Vector.h is a C++ header file that defines a C macro called
TEST_FUNC_PROTOS that takes two arguments:
TYPE, which is a data type name such as
unsigned int; and
SNAME, which is a short name for the same data type with no spaces, e.g.
uint. This macro defines several function prototypes that have the prefix
SNAME and have at least one argument that is an array of type
TYPE. Those functions that have return arguments return a
TEST_FUNC_PROTOS is then implemented for all of the data types supported by
unsigned long long
Vector.cxx is a C++ source file that implements compilable code for each of the function prototypes specified in
Vector.h. It defines a C macro
TEST_FUNCS that has the same arguments and works in the same way as
TEST_FUNC_PROTOS does in
TEST_FUNCS is implemented for each of the 12 data types as above.
Vector.i is a SWIG interface file that defines python module
Vector. It follows the conventions for using
numpy.i as described in this chapter. It defines a SWIG macro
%apply_numpy_typemaps that has a single argument
TYPE. It uses the SWIG directive
%apply to apply the provided typemaps to the argument signatures found in
Vector.h. This macro is then implemented for all of the data types supported by
numpy.i. It then does a
%include "Vector.h" to wrap all of the function prototypes in
Vector.h using the typemaps in
make is used to build the testing extension modules,
testVector.py can be run to execute the tests. As with other scripts that use
unittest to facilitate unit testing,
testVector.py defines a class that inherits from
However, this class is not run directly. Rather, it serves as a base class to several other python classes, each one specific to a particular data type. The
VectorTestCase class stores two strings for typing information:
SNAMEprefixes used in
Vector.cxx. For example,
self.typeStr. For example, if
Each test defined by the
VectorTestCase class extracts the python function it is trying to test by accessing the
Vector module’s dictionary:
length = Vector.__dict__[self.typeStr + "Length"]
In the case of double precision tests, this will return the python function
We then define a new test case class for each supported data type with a short definition such as:
class doubleTestCase(VectorTestCase): def __init__(self, methodName="runTest"): VectorTestCase.__init__(self, methodName) self.typeStr = "double" self.typeCode = "d"
Each of these 12 classes is collected into a
unittest.TestSuite, which is then executed. Errors and failures are summed together and returned as the exit argument. Any non-zero result indicates that at least one test did not pass.
© 2005–2019 NumPy Developers
Licensed under the 3-clause BSD License.