[PATCH] debug: Fix fortified realpath C++ ODR violation (BZ 30516)

[Adhemerval Zanella Netto]

On 03/07/23 13:03, Florian Weimer wrote:
> * Adhemerval Zanella via Libc-alpha:
> 
>> @@ -41,10 +35,6 @@ __NTH (realpath (const char *__restrict __name, char *__restrict __resolved))
>>    if (sz == (size_t) -1)
>>      return __realpath_alias (__name, __resolved);
>>  
>> -#if defined _LIBC_LIMITS_H_ && defined PATH_MAX
>> -  if (__glibc_unsafe_len (PATH_MAX, sizeof (char), sz))
>> -    return __realpath_chk_warn (__name, __resolved, sz);
>> -#endif
>>    return __realpath_chk (__name, __resolved, sz);
>>  }
> 
> The issue is tje _LIBC_LIMITS_H_ conditional, right?
> 
> I suggest we just define PATH_MAX to the correct value in a Linux header
> (in a namespace-clean fashion), and add a check of that constant against
> the Linux UAPI value.  This constant is ABI anyway, and it's good to
> know if the kernel decides to change it underneath us.
> 
> Regressing C diagnostics for experimental C++ module support does not
> seem the right trade-off, sorry.

Ok, this makes sense.   I will send an updated version.