[PATCH] avoid GCC 10 warning -Wstringop-truncation

[Sebastian Huber]

On 30/07/2020 18:05, Gedare Bloom wrote:

> On Thu, Jul 30, 2020 at 9:55 AM Sebastian Huber
> <sebastian.huber at embedded-brains.de> wrote:
>> On 30/07/2020 17:53, Gedare Bloom wrote:
>>
>>> On Thu, Jul 30, 2020 at 9:44 AM Sebastian Huber
>>> <sebastian.huber at embedded-brains.de>  wrote:
>>>> On 30/07/2020 13:36, Aschref Ben-Thabet wrote:
>>>>
>>>>> diff --git a/testsuites/psxtests/psxndbm01/init.c b/testsuites/psxtests/psxndbm01/init.c
>>>>> index a13afa7315..b524aff0df 100644
>>>>> --- a/testsuites/psxtests/psxndbm01/init.c
>>>>> +++ b/testsuites/psxtests/psxndbm01/init.c
>>>>> @@ -218,7 +218,7 @@ rtems_task Init(rtems_task_argument ignored)
>>>>>
>>>>>       puts( "Fetch non-existing record and confirm error." );
>>>>>       test_strings = (char*)malloc(6);
>>>>> -  strncpy( test_strings, "Hello", 5 );
>>>>> +  memcpy( test_strings, "Hello", 5 );
>>>>>
>>>>>       test_strings[5] = '\0';
>>>> In the glibc devel list this approach was suggested for problems like this:
>>>>
>>>> *(char *) mempcpy( test_strings, "Hello", 5 ) = '\0';
>>>>
>>>> https://sourceware.org/legacy-ml/libc-alpha/2000-08/msg00061.html
>>>>
>>> This code is suspect/wrong. that should create a NUL at the start of
>>> the test_strings. I'd rather see the memcpy followed by appending the
>>> NUL. It is easy enough to understand I think.
>> Yes, this was also my impression then I did read this code snippet.
>> Please note that this is memPcpy(), a GNU extension.
> thanks, I had missed that.
>
> I don't think there is an easy generalized way to do the delimiting
> with the standard memcpy. You could do it in one line, if we really
> wanted something like:
>
> ((char *) memcpy (test_strings, "Hello", 5 ))[5] = '\0';
>
> We could even create a little helper to do it if it is a common pattern.

The mempcpy() is also available in Newlib.

The only thing I know is that strncpy() is a useless function. With the 
new GCC warnings it is nearly impossible to use correctly. This function 
is an historic accident.

There are several ways to fix the warnings and I think there is no clear 
direction. We basically have the option to use the OpenBSD invented 
strlcpy() and strlcat() functions or some sort of memcpy() and 
mempcpy(). I think the OpenBSD philosophy is that if you put something 
in the strl*() and then at least a C-string is created. This may help to 
contain software bugs a bit and hinder error propagation. The glibc 
developer philosophy is probably that they only care about correct code 
and if you make mistakes that you can go to hell.