[U-Boot] [PATCH] test/py: hush_if_test: Add tests to cover octal/hex values

[Stephen Warren]

On 10/21/19 4:53 PM, Simon Glass wrote:
> Hi Michal,
> 
> On Tue, 15 Oct 2019 at 00:09, Michal Simek <michal.simek at xilinx.com> wrote:
>>
>> Hi Simon,
>>
>> On 11. 10. 19 17:53, Simon Glass wrote:
>>> Hi Michal,
>>>
>>> On Fri, 11 Oct 2019 at 01:50, Michal Simek <michal.simek at xilinx.com> wrote:
>>>>
>>>> On 10. 10. 19 19:06, Simon Glass wrote:
>>>>> Hi Michal,
>>>>>
>>>>> On Thu, 10 Oct 2019 at 05:44, Michal Simek <michal.simek at xilinx.com> wrote:
>>>>>>
>>>>>> Extend test suite to cover also automatic octal/hex converstions which
>>>>>> haven't been implemented in past.
>>>>>>
>>>>>> Signed-off-by: Michal Simek <michal.simek at xilinx.com>
>>>>>> ---
>>>>>>
>>>>>> Depends on https://lists.denx.de/pipermail/u-boot/2019-September/383309.html
>>>>>>
>>>>>> There are of course other tests which we can run but not sure if make sense
>>>>>> to have there all combinations. The most interesting are mixed tests which
>>>>>> are failing before patch above is applied.
>>>>>> Definitely please let me know if you want to add any other test.
>>>>>> ---
>>>>>>   test/py/tests/test_hush_if_test.py | 27 +++++++++++++++++++++++++++
>>>>>>   1 file changed, 27 insertions(+)
>>>>>>
>>>>>
>>>>> I worry that these tests might be very slow since it requires a lot of
>>>>> interaction with U-Boot over a pipe. Is it possible to put them in C
>>>>> code instead, e.g. cmd_ut?
>>>>
>>>> I have of course running it on my HW and it is quite fast. It is just 16
>>>> more simple tests. And if this breaks gitlab/travis CI loops then we
>>>> have bigger problem.
>>>
>>> I mean running these tests on sandbox. The interactions with the
>>> sandbox command line are quite slow I think.
>>
>>
>> I am not sharing this concern.
>>
>> Before:
>> [u-boot]$ time ./test/py/test.py --bd sandbox -s -k hush >/dev/null
>>
>> real    0m2,403s
>> user    0m1,263s
>> sys     0m0,299s
>>
>> After
>> [u-boot]$ time ./test/py/test.py --bd sandbox -s -k hush >/dev/null
>>
>> real    0m2,864s
>> user    0m1,563s
>> sys     0m0,305s
>>
>> And if 0.4s on testing will cause issues somewhere else we have
>> different kind of problem.
> 
> +Stephen Warren
> 
> I originally mentioned this concern to Stephen we the test setup was
> created. At present even 'make qcheck' takes over a minute. Adding
> half a second to this every time we add a new test is not going to
> lead to a good place.
> 
> Stephen made some improvements to speed things up, and suggested that
> the problem would not bear out. The alternative was presumably to
> build U-Boot into a Python module to avoid the comms overhead. But we
> didn't go that path.
> 
> So I think we should only use Python when the tests cannot be written in C.

I don't really see any concern with the addition of a couple extra 
seconds of test. Clearly I'd rather see the test written in Python and 
using external interfaces (i.e. the shell) where they test features 
accessible through those interfaces, since that allows them to be 
validated on all platforms, rather than only in sandbox. I feel that 
sandbox is good for unit tests and providing host-based smoke tests, but 
nothing beats actually being able to test real platforms.