Testing

Introduction

Libraries need pervasive testing for continuous improvement. Any problem found and behavior described must be written down as test so that it is assured that no new regressions will be added.

Running Tests

Running all tests in the build directory:

make run_all

And on the target (installed) system:

kdb run_all

To run memcheck tests run in the build directory:

make run_memcheck

They are supplementary, ideally you run all three.

Some tests write into system paths. To avoid running tests that write to the disk you can use:

make run_nokdbtests

Directly running ctest might cause problems: You need to set LD_LIBRARY_PATH as run_all and run_nokdbtests do.

If the access is denied, several tests will fail. You have some options to avoid running them as root:

1. To avoid running the problematic test cases (reduces the test coverage!) run ctest without tests that have the label kdbtests: ctest --output-on-failure -LE kdbtests (which is also what make run_nokdbtests does)
2. To give your user the permissions to the relevant paths, i.e. (once as root): ``` kdb mount-info chown -R whoami kdb get system/info/constants/cmake/CMAKE_INSTALL_PREFIX/kdb get system/info/constants/cmake/KDB_DB_SPEC chown -R whoami kdb get system/info/constants/cmake/KDB_DB_SYSTEM ``` After that all test cases should run successfully as described above.
3. Compile Elektra so that system paths are not actual system paths, e.g. to write everything into the home directory (~) use cmake options: -DKDB_DB_SYSTEM="~/.config/kdb/system" -DKDB_DB_SPEC="~/.config/kdb/spec" (for another example with ini see scripts/configure-home)
4. Use the XDG resolver (see scripts/configure-xdg) and set the environment variable XDG_CONFIG_DIRS, currently lacks spec namespaces, see #734.

Conventions

• All names of the test must start with test (needed by test driver for installed tests).
• No tests should run if ENABLE_TESTING is OFF.
• All tests that access system/spec namespaces (e.g. mount something):
  • should be tagged with kdbtests:

     set_property(TEST testname PROPERTY LABELS kdbtests)
    

  • should not run, if ENABLE_KDB_TESTING is OFF.
  • should only write below
    • /tests/<testname> (e.g. /tests/ruby) and
    • system/elektra (e.g. for mounts or globalplugins).
  • clean up everything they change (in KDB and temporary files)
• If your test has memory leaks, e.g. because the library used leaks and they cannot be fixed, give them the label memleak with the following command:

    set_property(TEST testname PROPERTY LABELS memleak)
  

Strategy

The testing must happen on every level of the software to achieve a maximum coverage with the available time. In the rest of the document we describe the different levels and where these tests are.

CFramework

This is basically a bunch of assertion macros and some output facilities. It is written in pure C and very lightweight.

It is located here.

ABI Tests

C ABI Tests are written in plain C with the help of cframework.

The main purpose of these tests are, that the binaries of old versions can be used against new versions as ABI tests.

So lets say we compile Elektra 0.8.8 (at this version dedicated ABI tests were introduced) in the -full variant. But when we run the tests, we use libelektra-full.so.0.8.9 (either by installing it or by setting LD_LIBRARY_PATH). You can check with ldd which version is used.

The tests are located here.

C Unit Tests

C Unit Tests are written in plain C with the help of cframework.

It is used to test internal data structures of libelektra that are not ABI relevant.

ABI tests can be done on theses tests, too. But by nature from time to time these tests will fail.

They are located here.

Internal Functions

According to src/libs/elektra/libelektra-symbols.map, all functions starting with:

• libelektra
• elektra
• kdb
• key
• ks

get exported. Functions not starting with this prefix are internal only and therefore not visible in the test cases. Test internal functionality by including the corresponding C file.

Module Tests

The modules, which are typically used as plugins in Elektra (but can also be available statically or in the -full variant), should have their own tests.

Use the CMake macro add_plugintest for adding these tests.

C++ Unit Tests

C++ Unit tests are done using the Google Test framework. See architectural decision.

Use the CMake macro add_gtest for adding these tests.

Script Tests

Tests which need scripts are done using shell recorder or directly with POSIX shell commands. See architectural decision.

The script tests have different purposes:

• End to End tests (usage of tools as an end user would do)
• External compilation tests (compile and run programs as a user would do)
• Conventions tests (do internal checks that check for common problems)
• Meta Test Suites (run other test suites)

See here.

Shell Recorder

The more elegant way to specify script tests are via the so called shell recorder using Markdown Syntax.

See here.

Fuzz Testing

Copy some import files to testcase_dir and run:

/usr/src/afl/afl-1.46b/./afl-fuzz -i testcase_dir -o findings_dir bin/kdb import user/tests

ASAN

To enable sanitize checks use ENABLE_ASAN via cmake.

Then, to use ASAN, run run_asan in the build directory, which simply does:

    ASAN_OPTIONS=symbolize=1 ASAN_SYMBOLIZER_PATH=$(shell which llvm-symbolizer) make run_all

It could also happen that you need to preload ASAN library, e.g.:

    LD_PRELOAD=/usr/lib/clang/3.8.0/lib/linux/libclang_rt.asan-x86_64.so run_asan

or on Debian:

    LD_PRELOAD=/usr/lib/llvm-3.8/lib/clang/3.8.1/lib/linux/libclang_rt.asan-x86_64.so run_asan

See also build server jobs:

• clang-asan
• gcc-asan

CBMC

For bounded model checking tests, see scripts/cbmc.

Static Code Checkers

There is a number of static code checkers available for all kind of programming languages. The following section show how the most common ones can be used with libelektra.

Cppcheck

Cppcheck can be used directly on a C or C++ source file by calling it with cppcheck --enable=all <sourcefile>. This way it might miss some header files though and thus doesn't detect all possible issues, but still gives useful hints in general.

To analyze the whole project, use it in conjunction with cmake by calling cmake with the parameter -DCMAKE_EXPORT_COMPILE_COMMANDS=ON. This way cmake creates a file called compile_commands.json in the build directory. Afterwards, call cppcheck with the cmake settings and store the output as xml:

cppcheck --project=compile_commands.json --enable=all -j 8 --xml-version=2 2> cppcheck_result.xml

Since the XML file is difficult to read directly, the best way is to convert it to an HTML report. Cppcheck already includes a tool for that, call it with the XML report:

cppcheck-htmlreport --file=cppcheck_result.xml --report-dir=cppcheck_report --source-dir=.

Now you can view the html report by opening index.html in the specified folder to get an overview of the issues found in the whole project.

OCLint

OCLint is a static code analyzer for C, C++ and Objective C. To use this tool enable the CMake option CMAKE_EXPORT_COMPILE_COMMANDS. The steps below show a step-by-step guide on how to analyze files with OCLint.

1. Create a build directory if you have not done so already and change the working path to this directory:

mkdir -p build
cd build

1. Run CMake with the option CMAKE_EXPORT_COMPILE_COMMANDS:

cmake .. -DCMAKE_EXPORT_COMPILE_COMMANDS=ON

1. Build Elektra

make

1. Run the oclint command specifying the files you want to analyze

cd ..
oclint -p build -no-analytics -enable-global-analysis -enable-clang-static-analyzer src/plugins/ini/*.c

scan-build

scan-build is a tool that is usually bundled along with LLVM/Clang and is also primarily intended for C and C++ code. On macOS you have to install the package llvm with homebrew, then you'll find the tool in the folder /usr/local/opt/llvm/bin/.

To use it, change the C compiler and the C++ compiler to the LLVM analyzer. To do this, you can configure the project from scratch and prefix the cmake command with scan-build. Alternatively, set the c compiler to ccc-analyzer and the C++ compiler to c++-analyzer (bundled with LLVM/Clang).

Then you can build the project with make like usual, prefixing the command with scan-build. The -o option specifies where the html results get stored. Ensure you build the project from scratch, otherwise the analyzation might be incomplete.

scan-build -o ./scanbuild_result make -j 4

Afterwards, the report can be viewed by using the tool scan-view, also found in the llvm folder. The report is created in the folder specified above, along with the current date of the analyzation, for instance:

scan-view <path specified above>/2017-06-18-171027-27108-1

Alternatively, you can also open the index.html file in the aforementioned folder, but using the tool the report is enriched with further information.

SonarLint

SonarLint is a static code checker primarily intended for Java. It is usually used by installing the corresponding plugin for the used IDE, then there is no further configuration required.

Randoop

For using the unit test generator randoop with the jna bindings, see scripts/randoop/randoop.

Code Coverage

Run:

    make coverage-start
    # now run all tests! E.g.:
    make run_all
    make coverage-stop
    make coverage-genhtml

The HTML files can be found in the build directory in the folder coverage.

See also the build server job:

• gcc-asan

See also

• COMPILE.
• INSTALL.