Re: [PATCH] qtest: Fix bad printf format specifiers

[Thomas Huth]

On 06/11/2020 15.18, Philippe Mathieu-Daudé wrote:
> On 11/6/20 7:33 AM, Markus Armbruster wrote:
>> Thomas Huth <thuth@redhat.com> writes:
>>
>>> On 05/11/2020 06.14, AlexChen wrote:
>>>> On 2020/11/4 18:44, Thomas Huth wrote:
>>>>> On 04/11/2020 11.23, AlexChen wrote:
>>>>>> We should use printf format specifier "%u" instead of "%d" for
>>>>>> argument of type "unsigned int".
>>>>>>
>>>>>> Reported-by: Euler Robot <euler.robot@huawei.com>
>>>>>> Signed-off-by: Alex Chen <alex.chen@huawei.com>
>>>>>> ---
>>>>>>  tests/qtest/arm-cpu-features.c | 8 ++++----
>>>>>>  1 file changed, 4 insertions(+), 4 deletions(-)
>>>>>>
>>>>>> diff --git a/tests/qtest/arm-cpu-features.c 
>>>>>> b/tests/qtest/arm-cpu-features.c
>>>>>> index d20094d5a7..bc681a95d5 100644
>>>>>> --- a/tests/qtest/arm-cpu-features.c
>>>>>> +++ b/tests/qtest/arm-cpu-features.c
>>>>>> @@ -536,7 +536,7 @@ static void test_query_cpu_model_expansion_kvm(const 
>>>>>> void *data)
>>>>>>          if (kvm_supports_sve) {
>>>>>>              g_assert(vls != 0);
>>>>>>              max_vq = 64 - __builtin_clzll(vls);
>>>>>> -            sprintf(max_name, "sve%d", max_vq * 128);
>>>>>> +            sprintf(max_name, "sve%u", max_vq * 128);
>>>>>>
>>>>>>              /* Enabling a supported length is of course fine. */
>>>>>>              assert_sve_vls(qts, "host", vls, "{ %s: true }", max_name);
>>>>>> @@ -556,7 +556,7 @@ static void test_query_cpu_model_expansion_kvm(const 
>>>>>> void *data)
>>>>>>                   * unless all larger, supported vector lengths are also
>>>>>>                   * disabled.
>>>>>>                   */
>>>>>> -                sprintf(name, "sve%d", vq * 128);
>>>>>> +                sprintf(name, "sve%u", vq * 128);
>>>>>>                  error = g_strdup_printf("cannot disable %s", name);
>>>>>>                  assert_error(qts, "host", error,
>>>>>>                               "{ %s: true, %s: false }",
>>>>>> @@ -569,7 +569,7 @@ static void test_query_cpu_model_expansion_kvm(const 
>>>>>> void *data)
>>>>>>               * we need at least one vector length enabled.
>>>>>>               */
>>>>>>              vq = __builtin_ffsll(vls);
>>>>>> -            sprintf(name, "sve%d", vq * 128);
>>>>>> +            sprintf(name, "sve%u", vq * 128);
>>>>>>              error = g_strdup_printf("cannot disable %s", name);
>>>>>>              assert_error(qts, "host", error, "{ %s: false }", name);
>>>>>>              g_free(error);
>>>>>> @@ -581,7 +581,7 @@ static void test_query_cpu_model_expansion_kvm(const 
>>>>>> void *data)
>>>>>>                  }
>>>>>>              }
>>>>>>              if (vq <= SVE_MAX_VQ) {
>>>>>> -                sprintf(name, "sve%d", vq * 128);
>>>>>> +                sprintf(name, "sve%u", vq * 128);
>>>>>>                  error = g_strdup_printf("cannot enable %s", name);
>>>>>>                  assert_error(qts, "host", error, "{ %s: true }", name);
>>>>>>                  g_free(error);
>>>>>>
>>>>>
>>>>> max_vq and vq are both "uint32_t" and not "unsigned int" ... so if you 
>>>>> want
>>>>> to fix this really really correctly, please use PRIu32 from inttypes.h 
>>>>> instead.
>>>>>
>>>>
>>>> Hi Thomas,
>>>> Thanks for your review.
>>>> According to the definition of the macro PRIu32(# define PRIu32         
>>>> "u"),
>>>> using PRIu32 works the same as using %u to print, and using PRIu32 to print
>>>> is relatively rare in QEMU(%u 720, PRIu32 only 120). Can we continue to 
>>>> use %u to
>>>> print max_vq and vq in this patch.
>>>> Of course, this is just my small small suggestion. If you think it is 
>>>> better to use
>>>> PRIu32 for printing, I will send patch V2.
>>>
>>> Well, %u happens to work since "int" is 32-bit with all current compilers
>>> that we support.
>>
>> Yes, it works.
>>
>>>                  But if there is ever a compiler where the size of int is
>>> different, you'll get a compiler warning here again.
>>
>> No, we won't.
>>
>> If we ever use a compiler where int is narrower than 32 bits, then the
>> type of the argument is actually uint32_t[1].  We can forget about this
>> case, because "int narrower than 32 bits" is not going to fly with our
>> code base.

Agreed.

>> If we ever use a compiler where int is wider than 32 bits, then the type
>> of the argument is *not* uint32_t[2].  PRIu32 will work anyway, because
>> it will actually retrieve an unsigned int argument, *not* an uint32_t
>> argument[3].

I can hardly believe that this can be true. Sure, it's true for such cases
like this one here, where you multiply with an "int". But if you just try to
print a plain uint32_t variable?

I've seen compiler warning in cases one tries to print a 16-bit (i.e. short)
variable in the past if you use %d instead of the proper PRId16 (or %hd)
format specifier - maybe not on x86, but certainly on other architectures.
If you're statement was right, that should not have happened, should it?

>> In other words "%" PRIu32 is just a less legible alias for "%u" in all
>> cases that matter.
> 
> Can we add a checkpatch rule to avoid using 'PRI[dux]32' format,
> so it is clear for everyone?

I don't think that this is a good idea. Using PRI*32 is IMHO perfectly fine
if you really have a variable of the type [u]int32_t ... it's just not if
you do some arithmetics with it first.

 Thomas


>>
>>
>> [1] Because promotion does nothing either argument, and the usual
>> arithmetic conversions convert to uint32_t.  See my first reply.
>>
>> [2] Because uint32_t gets promoted to unsigned int.  See my first reply.
>>
>> [3] Because variable arguments undergo default argument promotion (§
>> 6.5.2.2 Function calls), which promotes uint32_t to unsigned int.
>>
>>
> 
>