diff options
| author | Zachary Turner <zturner@google.com> | 2015-03-26 16:43:25 +0000 |
|---|---|---|
| committer | Zachary Turner <zturner@google.com> | 2015-03-26 16:43:25 +0000 |
| commit | 65fe1eb5f83c9ec9c4505287cd0a9b1f1b093c17 (patch) | |
| tree | 86595ef03756b4dfe2cfa97f69a21356e1d26f88 /libcxx/test/std/utilities/meta | |
| parent | aa4dabfd857d7592d52e6f1510170ea56b4c4e25 (diff) | |
| download | bcm5719-llvm-65fe1eb5f83c9ec9c4505287cd0a9b1f1b093c17.tar.gz bcm5719-llvm-65fe1eb5f83c9ec9c4505287cd0a9b1f1b093c17.zip | |
Tear down tests in reverse order from setting them up.
Tests derive from TestBase, which derives from Base. In the
test setUp() methods, we always call TestBase.setUp() first and
then call implementation-specific setup. Tear down needs to do
the reverse.
This was causing over 20 failures on Windows, and was the culprit
behind about 80% of the files not being cleaned up after test run.
TestBase.tearDown() is responsible for deleting all targets created
during the test run and without this step, on Windows files will
be locked and cannot be deleted. But TestBase.tearDown() was
calling Base.tearDown() before its own cleanup (i.e. deleting the
targets) and in some cases one of the teardown hooks would be to
call make clean. So make clean would be run before the targets
had been deleted, and fail to remove the files, and subsequently
result in a failed test as well.
llvm-svn: 233284
Diffstat (limited to 'libcxx/test/std/utilities/meta')
0 files changed, 0 insertions, 0 deletions

