Tim Day wrote:
If I grep my /usr/include/boost/test (I'm still on 1.35) for "depend" then I'm intrigued to see there appears to be some sort of support for declaring dependencies between tests (which would presumably influence the order)
No. Not at the moment. If you believe this needs t obe fixed please submit bug report.
e.g in unit_test_suite_impl.hpp test_unit has a depends_on method taking (a pointer to) another test_unit as an argument. I haven't tried doing anything with it though. Would be interested to know more... it doesn't seem to be documented.
Essentially it's single method in test_unit: void depends_on( test_unit* tu );
I can sympathise with wanting to run tests in a specific order. I generally like to run the quick (typically low level) ones first, and leave the slow ones until later. My preferred solution is to maintain two test executables; one with the tests quick enough to be run after every compile, the other (which might take some 10s of seconds) is better run on demand (after making big changes, before commits).
You can maintain one executable and select est cases to run with command line parameter. For example: test_module.exe --run=*/basic_test only executes test cases named basic_test in all test suites Gennadiy