Libelektra: ASAN: Some Tests Fail on macOS with Enabled Leak Detection

[sanssecours]

Steps to Reproduce the Problem

Enter the commands below at the root directory of the repository.

brew install llvm
export CC=/usr/local/opt/llvm/bin/clang
export CXX=/usr/local/opt/llvm/bin/clang++
export ASAN_OPTIONS=detect_leaks=1
export PKG_CONFIG_PATH=/usr/local/opt/openssl/lib/pkgconfig

mkdir build
cd build
cmake .. -GNinja -DENABLE_ASAN=ON -DBUILD_TESTING=ON -DTOOLS='kdb' -DBINDINGS='cpp' -DPLUGINS='ALL'
ninja
ninja test

Expected Result

All tests finish successfully.

Actual Result

The following tests fail:

• testtool_backendbuilder
• testtool_plugindatabase
• testmod_augeas
• testmod_dbus
• testmod_type
• testscr_check_kdb_internal_check
• testkdb_allplugins

.

System Information

• Elektra Version: master
• Operating System: macOS 10.13.6

Further Log Files and Output

Travis: Build with Enabled Leak Detection
Log Files:

• testkdb_allplugins.log.txt
• testmod_augeas.log.txt
• testmod_dbus.log.txt
• testmod_type.log.txt
• testscr_check_kdb_internal_check.log.txt
• testtool_backendbuilder.log.txt
• testtool_plugindatabase.log.txt

[markus2330]

Thank you for reporting this problem! What did change so that these errors occurred? Or were they always present and simply never reported before?

[sanssecours]

What did change so that these errors occurred?

The environment variable ASAN_OPTIONS.

Or were they always present and simply never reported before?

The tests failed before too. I was just to lazy to report the issue. Since the build system now excludes all problematic tests on Linux, I thought it would make sense to do the same on macOS.

[markus2330]

Yes, it makes sense to automatically exclude tests that cannot work because of the platform/compiler/...

Let us discuss this tomorrow.

[markus2330]

Is now augeas the only test case that fails or do still all of them fail?

[sanssecours]

All of them fail. The tests testscr_check_kdb_internal_check, testkdb_allplugins, and testtool_plugindatabase do fail because of the augeas plugin though.

[sanssecours]

The reason why testmod_dbus failed on my machine was that I did not start dbus beforehand:

brew services start dbus

. I still think it would make sense to handle this situation more gracefully, if it does not require that much work.

[markus2330]

@waht Can you maybe take a look at the dbus issue?

[waht]

I'm on it. IMHO testmod_dbus is handling the unavailable dbus daemon rather gracefully (i.e. the test prints a warning and returns 0 errors) but the dbus library is producing a memory leak at least on macOS.
On Linux (stretch container) there is no leak without a dbus daemon.
API use seems fine (i.e. the error is freed before return) but I'll check if a call to dbus_shutdown() will fix this.

[waht]

It seems like dbus_shutdown() does the job: Travis Build Log
@sanssecours Do you want add the quite trivial fix to a PR of yours of should I create one?

[sanssecours]

@sanssecours Do you want add the quite trivial fix to a PR of yours of should I create one?

Thanks for taking care of the problem. I cherry-picked your commit in PR #2201.