[cfe-dev] scan-build constraints

[Paulo Matos via cfe-dev]

On 07/06/2019 06:38, Artem Dergachev wrote:
> You can use assert() to communicate your knowledge about the code to the
> static analyzer. E.g., assert that the pointer is non-null before
> dereferencing. It also saves you from undefined behavior when the static
> analyzer turns out to be right.
>

In that case, is there any way to determine through cpp macros if we are
running under scan_build?


> That said, also please feel free to send us a bug report about these
> false positives through llvm bugzilla (with either a full html report
> or, even better, a small reproducer) - i'll see what i can do.
> 
> On 06.06.2019 12:37, Paulo Matos via cfe-dev wrote:
>> Hi,
>>
>> scan-build is generating null pointer dereference complaints because it
>> does not understand that there's a relation between the arguments to a
>> function.
>>
>> The function looks like:
>> void *array_func(void **arr, int arrl) ...
>>
>> where arrl is equal to the number of elements in array l so any access
>> to arr[0] - arr[arrl-1] is fine. Is there a way to specify to scan-build
>> these relationships?
>>
>> Thanks,
>>

-- 
Paulo Matos