Increase testing coverage of rcutils to 95% #258
Conversation
brawner
force-pushed
the
brawner/rcutils-tests
branch
3 times, most recently
from
52d56a5 to
62ed0ee
Compare
|
Separate PR to fix Windows bug: |
Signed-off-by: Stephen Brawner <[email protected]>
Signed-off-by: Stephen Brawner <[email protected]>
brawner
force-pushed
the
brawner/rcutils-tests
branch
from
938b03f to
1aef73c
Compare
Signed-off-by: Stephen Brawner <[email protected]>
|
After merging #259, I've rebased this PR. Also for API considerations I moved the time_bomb allocator into the test directory. We can leave that discussion for post-foxy discussions. |
|
Retesting:
|
|
Also without #257 this shows up as only 94%. At 1666/1763, it just needs one more line of coverage or one less line of uncovered code 🤷♂️ |
Signed-off-by: Stephen Brawner <[email protected]>
|
Signed-off-by: Stephen Brawner <[email protected]>
Signed-off-by: Stephen Brawner <[email protected]>
|
|
||
| // There are no outputs of the handler function, and the only result are fprintf() calls. | ||
| // This is just a smoke test to check that the code can handle simple inputs cleanly. | ||
| TEST(TestLoggingConsoleOutputHandler, typical_inputs) { |
There was a problem hiding this comment.
@brawner meta: thinking again about this test, you could mock stdout and/or stderr for the duration of the test. Won't block on it, but it is doable.
There was a problem hiding this comment.
I tried this out and couldn't get it to work. Assigning a new file pointer to stdout/stderr left me with blank files. Alternatively, using freopen() worked but it wouldn't let me reset stdout/stderr back in an easily portable way.
I like the idea of checking expected input on stdout/stderr, but I'll need a better solution, which may have to come later.
There was a problem hiding this comment.
Hmm, interesting. Oh well, it can be done in a follow-up PR.
|
|
||
| // There are no outputs of the handler function, and the only result are fprintf() calls. | ||
| // This is just a smoke test to check that the code can handle bad inputs cleanly. | ||
| TEST(TestLoggingConsoleOutputHandler, bad_inputs) { |
|
Testing once more before merging:
|
I've added a bit more test cases to the unit tests in rcutils to boost the coverage to 95% (just barely).
The big question about this PR is the desired location of
time_bomb_allocator. @hidmic had suggested putting it into a centralized location so it can be used in other packages. I'm thinking I might want to do that for the allocator_testing_utils.h as well. But I don't think rcutils/include is the desired end location.The remaining 5% of untested logic pretty much consists of failure checks that won't fail, generally because bad parameters have already been sanitized by the time the function under test calls a function that is checked for failure. In a few cases it might be because the system call cannot be made to fail reliably or even at all in a simple unit test.
Signed-off-by: Stephen Brawner [email protected]