Re: problem with inline functions in gsweb and clang

[Wolfgang Lux]

David Chisnall wrote:

> On 27 May 2013, at 18:18, Ivan Vučica <ivucica@gmail.com> wrote:
> 
>> Somewhat off topic: I really like the following error; it's extremely 
>> professional and mature. Where is it coming from?
>> 
>> /usr/local/lib/libgcrypt.so.18.0: warning: stpcpy() is dangerous GNU crap; 
>> don't use it
> 
> It's a linker error that is triggered by the presence of an attribute on a 
> function.  The text is a reference to Ulrich Drepper's refusal to accept 
> patches to implement the strl* functions in glibc because they were 
> 'inefficient BSD crap'.  
> 
> It is 'dangerous', because it will NULL-terminate the string in some cases, 
> but not others.  This means that you must always check the lengths of both 
> strings before calling it, at which point you may as well use memcpy, which 
> will be faster.

Don't get things mixed up. :-)
stpcpy and strcpy are dangerous because these functions might overrun the 
buffer you are writing to. It is strpncpy and strncpy, which may fail to write 
a '\0' to the end of the string.

> Note, however, that the GNU man page explicitly claims not to have invented 
> this function, and it is now part of POSIX2008 (Austin has a habit of 
> standardising bad ideas if two or more libcs have implemented them), so the 
> 'GNU' in the warning is misleading.  

Wolfgang