svn commit: r328492 - head/contrib/opie/libopie

[Pedro Giffuni]

On 01/27/18 20:42, Bruce Evans wrote:
> On Sat, 27 Jan 2018, Pedro Giffuni wrote:
>
>> On 01/27/18 18:21, Bruce Evans wrote:
>>> On Sat, 27 Jan 2018, Dimitry Andric wrote:
>>>
>>>> On 27 Jan 2018, at 23:20, Ed Schouten <ed at nuxi.nl> wrote:
>
>> * [... context lost to corruption of spaces which makes it unreadable]
>
>>>>>
>>>>> Wait... This may access utmp.ut_host one byte past the end and no
>>>>> longer guarantees that host is null-terminated, right?
>>>
>>>> No, strncpy "copies at most len characters from src into dst".  
>>>> However,
>>>
>>> No, the change breaks the length so 1 byte past the end is accessed
>>> in implementations where ut_host is not guaranteed to be NUL terminated
>>> and the current instance of ut_host is not NUL terminated.
>>>
>> The main change is in the sizeof(). Regularly you should use the size 
>> of destination not the source, and apparently GCC8 decided there was 
>> something to check there.
>
> That is the main breakage.  Using the size of the destination is very 
> wrong,
> since that size is intentionally 1 larger than the size of the source, to
> leave space for appending a NUL.
>
I am considering reverting the change. Looking at other ways to get rid 
of the warning, please be patient.

>> ...
>> Looking in detail, upstream (which appears to have disappeared) does 
>> have the explicit NULL termination in our last import. For 
>> consistency and given that we already have a strlcpy in that code, we 
>> should use strlcpy() there. Every modern OS out there has strlcpy(3) 
>> and if not they can figure out what to do.
>
> strlcpy() still seems to be intentionally left out of glibc.
>

glibc is not portable. I understand some systems that carry glibc also 
carry libbsd, or they can still use musl:

https://github.com/esmil/musl/blob/master/src/string/strlcpy.c


Pedro.