On Thu, 31 Jan 2008, Reece Dunn wrote: [...]
Also, the security tests (the previous one) have two output lines, e.g.:
security: 18 tests executed (0 marked as todo, 0 failures), 0 skipped. security: 944 tests executed (0 marked as todo, 0 failures), 1 skipped.
Yes, this typically happens when one does tests in a child process. This is wrong because it breaks 'make xxx.ok'.
The reason is that the parent process does not know whether some tests failed in the child process or not, and thus it will return 0. Make will thus think that everything went well and so 'make xxx.ok' will succeed eventhough it should have failed.
Alexandre relies on 'make test' to give him a reliable result so this may/will cause him not to notice new test failures introduced by a patch if they happen in a child process.
In dlls/kernel32/tests/debugger.c I have code that passes data from the child process back to the parent so the tests can be done in the parent, specifically to avoid this problem. But maybe we should have a general mechanism to deal with this.