Re: [RFC net-next 1/2] selftests: drv-net: add ability to schedule cleanup with defer()

[Przemek Kitszel <przemyslaw.kitszel@intel.com>]

On 6/26/24 18:49, Jakub Kicinski wrote:
> On Wed, 26 Jun 2024 09:43:54 +0200 Przemek Kitszel wrote:
>>> As a nice safety all exceptions from defer()ed calls are captured,
>>> printed, and ignored (they do make the test fail, however).
>>> This addresses the common problem of exceptions in cleanup paths
>>> often being unhandled, leading to potential leaks.
>>
>> Nice! Please only make it so that cleanup-failure does not overwrite
>> happy-test-path-failure (IOW "ret = ret ? ret : cleanup_ret")
> 
> That should be what we end up doing. The ret is a boolean (pass / fail)
> so we have:
> 
> 	pass &= cleanup_pass
> 
> effectively.
> 
>>> +            ksft_pr("Exception while handling defer / cleanup!")
>>
>> please print current queue size, if only for convenience of test
>> developer to be able tell if they are moving forward in
>> fix-rerun-observe cycle
> 
> Hm... not a bad point, defer() cycles are possible.
> But then again, we don't guard against infinite loops
> in  tests either, and kselftest runner (the general one,
> outside our Python) has a timeout, so it will kill the script.

I mean the flow:
$EDITOR mytest.py
./mytest.py
# output: Exception while handling defer / cleanup (at 4 out of 13 cleanups)

then repeat with the hope that fix to cleanup procedure will move us
forward, say:
$EDITOR mytest.py; ./mytest.py
#output: ... (at 7 of 13 cleanups)

just name of failed cleanup method is not enough as those could be
added via loop

> 
>>> +            tb = traceback.format_exc()
>>> +            for line in tb.strip().split('\n'):
>>> +                ksft_pr("Defer Exception|", line)
>>> +            KSFT_RESULT = False
>>
>> I have no idea if this could be other error than just False, if so,
>> don't overwrite
> 
> Yup, just True / False. The other types (skip, xfail) are a pass
> (True) plus a comment, per KTAP spec.
> 


Thanks!