Following <https://trac.webkit.org/changeset/230998> (bug #181043), run-api-tests now prepends the binary name of the test executable that contains the test when it emits the name of the test. Knowing the binary name that contains a test does not seem very meaningful because it is an implementation detail of how we chose to bundle and run tests (by splitting tests across more than one binary). As far as I can tell, any advantage claimed by prepending the binary name to the test to avoid collisions seem superfluous. It would be good to know if people ever ran into such collisions. Otherwise, it seems sufficient to revert to our behavior before <https://trac.webkit.org/changeset/230998> and rely on Google unit test's default namespace behavior where each test is identified by its test suite and test name. This problem may be easier to see visually. The following is the output of a passing test and failing test run with run-api-test before and after r230998. == Passing test == Before r230998: [[ $ Tools/Scripts/old-run-api-tests --debug --no-build --verbose WTF.StringViewStripLeadingAndTrailingMatchedCharacters PASS WTF.StringViewStripLeadingAndTrailingMatchedCharacters ]] After r230998: [[ $ Tools/Scripts/run-api-tests --debug --no-build --verbose WTF.StringViewStripLeadingAndTrailingMatchedCharacters Checking build ... "perl Tools/Scripts/webkit-build-directory --configuration --debug --mac" took 0.41s Collecting tests ... "/Volumes/Data/WebKitDev/OpenSource/WebKitBuild/Debug/TestWTF --gtest_list_tests" took 0.16s "/Volumes/Data/WebKitDev/OpenSource/WebKitBuild/Debug/TestWebKitAPI --gtest_list_tests" took 0.23s Found 1 tests Running tests Sharding tests ... worker/0 starting TestWTF.WTF.StringViewStripLeadingAndTrailingMatchedCharacters Passed worker/0 exiting Ran 1 tests of 1 with 1 successful ------------------------------ All tests successfully passed! Testing completed, Exit status: 0 ]] == Failing test == Before r230998: [[ $ Tools/Scripts/old-run-api-tests --debug --no-build --verbose WTF.StringViewStripLeadingAndTrailingMatchedCharacters FAIL WTF.StringViewStripLeadingAndTrailingMatchedCharacters /Volumes/Data/WebKitDev/OpenSource/Tools/TestWebKitAPI/Tests/WTF/StringView.cpp:958 Value of: stringViewFromLiteral("AAABBAAA").stripLeadingAndTrailingMatchedCharacters(isA) == stringViewFromLiteral("BBB") Actual: false Expected: true Tests that failed: WTF.StringViewStripLeadingAndTrailingMatchedCharacters ]] After r230998: [[ $ Tools/Scripts/run-api-tests --debug --no-build --verbose WTF.StringViewStripLeadingAndTrailingMatchedCharacters Checking build ... "perl Tools/Scripts/webkit-build-directory --configuration --debug --mac" took 0.39s Collecting tests ... "/Volumes/Data/WebKitDev/OpenSource/WebKitBuild/Debug/TestWTF --gtest_list_tests" took 0.17s "/Volumes/Data/WebKitDev/OpenSource/WebKitBuild/Debug/TestWebKitAPI --gtest_list_tests" took 0.20s Found 1 tests Running tests Sharding tests ... worker/0 starting /Volumes/Data/WebKitDev/OpenSource/Tools/TestWebKitAPI/Tests/WTF/StringView.cpp:958 Value of: stringViewFromLiteral("AAABBAAA").stripLeadingAndTrailingMatchedCharacters(isA) == stringViewFromLiteral("BBB") Actual: false Expected: true TestWTF.WTF.StringViewStripLeadingAndTrailingMatchedCharacters Failed worker/0 exiting Ran 1 tests of 1 with 0 successful ------------------------------ Test suite failed Failed TestWTF.WTF.StringViewStripLeadingAndTrailingMatchedCharacters /Volumes/Data/WebKitDev/OpenSource/Tools/TestWebKitAPI/Tests/WTF/StringView.cpp:958 Value of: stringViewFromLiteral("AAABBAAA").stripLeadingAndTrailingMatchedCharacters(isA) == stringViewFromLiteral("BBB") Actual: false Expected: true Testing completed, Exit status: 3 ]] As you can see from the above output we did not prepend the binary name (TestWTF) to the test name before r230998.
I very strongly disagree with the premise of this bug. Prepending the binary name ensures that we won't have collisions, and it really doesn't do any harm. This bug would have more merit if the test harness didn't recognize a test by <suite>.<name>, but it does. Additionally, if we start tracking API test results (which is something that's been requested) colliding suite or test names would be a real problem.