Re: [PATCH] Check malloc result in getopt_long

[Pádraig Brady]

On 16/02/15 20:06, Tobias Stoeckmann wrote:
> In contrast to glibc's getopt implementation, this one uses malloc instead
> of alloca. While in general this is a good idea, malloc returns NULL in
> case of error, therefore the return value should be checked.
> 
> The fprintf statements are copied from the same source file; I am rather
> indifferent about the error message.
> 
> Using coreutils' ls (I verified that gnulib's getopt is used on my system):
> 
> tobias:~/git/coreutils$ ulimit -Sv 4226
> tobias:~/git/coreutils$ ./src/ls --t
> Segmentation fault

Yes we shouldn't segfault.
That change came with 
http://git.savannah.gnu.org/gitweb/?p=gnulib.git;a=commit;h=70aa91cc

> With this patch:
> 
> tobias:~/git/coreutils$ ./src/ls --t
> ./src/ls: out of memory, dropping option 'time' while resolving ambiguity
> ./src/ls: out of memory, dropping option 'time-style' while resolving 
> ambiguity
> ./src/ls: option '--tabsize' requires an argument
> Try './src/ls --help' for more information.

I'm not sure it's generally worth proceeding in OOM conditions
unless there are clear fall backs.  It would be nice to depend on xalloc
and just call xmalloc(). Could this suffice?

  #ifdef _LIBC
  struct option_list *newp = alloca (sizeof (*newp));
  #else
  struct option_list *newp = xmalloc (sizeof (*newp));
  #endif

thanks!
Pádraig.