Skip to end of metadata
  • Created by Former user, last modified by Former user on Feb 11, 2011
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 4 Current »

The CppUnit framework is used for unittesting C/C++ source code for all sipXecs related projects. The CppUnit Cookbook is the general guide to creating unittests but specific guidelines and information about extensions made to the CppUnit framework are described in this document.

For any given source file named src/foo/MyClass.cpp, place unittests in a file named src/test/foo/MyClassTest.cpp.

Use the source file TestTemplate.cpp as a template for creating all new unittests. (hint: Copy TestTemplate.cpp to a new file called MyClassTest.cpp then search and replace TestTemplate with MyClassTest.)

Add your unittest to the testsuite that is run as part of the build by listing it in the src/test/Makefile.am under the testsuite_SOURCES make variable. (TIP: To temporarily isolate just your unittest, remove all other source file listings from make variable except the files in the sipxunit directory. Alternatively you can create a separate target program with just your unittest and source files from the sipxunit directory, just be sure NOT to check it in that way.)

Logging is encouraged in unittests. All log levels are set to DEBUG during execution of the unittest, and the output of each test can be found in $BUILD/<project>/src/test with the appropriate test name.

If any one unittest fails, it will stop the build from succeeding. In scenarios where you want to check in a failed unittest to communicate a bug beyond your control to fix, you can use the macros KNOWN_BUG and KNOWN_FATAL_BUG to signify such issues. This will not break the build, but will report errors to stdout. See TestUtilities.h header file for more information.

libtool is used to build the testsuite, consequently you must use libtool to invoke a debugger on the testsuite. Example:

  libtool --mode=execute gdb .../testsuite

where the last argument is the path to the testsuite file in the directory containing the testsuite-....o files generated from the tests. Since the tests are built into shared libraries, you might not be able to set breakpoints in them at the moment gdb starts, since the shared libraries may not have yet been loaded. If so, use the usual trick of break main then run to force the executable preamble code to link in all the shared libraries.

You can also execute a subset of the tests by using the "sandbox" feature. To use it, edit Makefile.am in the directory containing the .o files. The sandbox_SOURCE variable lists the tests that will be included in the subset. Rebuild the directory (including the autoconf and configure steps), then run them under the debugger by doing:

  libtool --mode=execute gdb .../sandbox

It is recommended each unittest be run through tools like Valgrind, Electric Fence and gcov to test for memory management and threading issues and to determine unittest code coverage. These tools will be used by automated, stress-testing machines so it's best if you use these tools ahead of time.

Valgrind

Valgrind can detect many memory management and threading bugs. See Valgrind HOWTO for more information. To run unittests under valgrind:

 libtool --mode=execute valgrind --leak-check=yes -v --show-reachable=yes src/test/testsuite

Deciphering output printed to console will take some practice. You may find section "4.2.4 How to Suppress Errors" of Valgrind HOWTO helpful. Please email suggestions to dev list for ways to integrate expected leaks and errors into project.

Electric Fence

Running the unittests under Electric Fence is still experimental, but it has been known to find some issues. Install
Electric Fence according to documentation and simply link to memory management library like so:

  ./configure LDFLAGS=-lefence

Best documentation is man page man efence. If you see Electric Fence's initialization header printed to console when running unittests then Electric Fence is compiled in and working. Errors will surface as segmentation faults closer to points of bad memory management. Run unittests thru gdb or similar debugger to further diagnose issues.

gcov

It's important to analyze the code coverage of your unittests on the targeted source code. To compile a unittest with code
coverage analysis built in, configure the project like so:

  ./configure CXXFLAGS="-fprofile-arcs -ftest-coverage"
  make clean all check

whenever the unittests are run, coverage will be captured. To inspect coverage:

  cd src
  gcov {target}-MyClassName
  cat MyClass.cpp.gcov

Where {target} is your library or binary name. (hint: run ls in src directory.) Files ending in *.gcov will be annotated listing of the source code with the number of times each line has been executed or ##### for lines that have not been executed.

  • No labels