Re: tar - covscan issues

[Paul Eggert]

All these GNU Tar Coverity reports appear to be false alarms. Here are 
some details:

On 4/7/21 1:24 AM, Ondrej Dubaj wrote:

> -      return copy != NULL ? copy : data;
> +      if (copy != NULL)
> +      {
> +        data = NULL;
> +        return copy;
> +      }
> +      else
> +        return data;

This patch doesn't do anything since 'data' is a local variable, which 
means setting it to NULL has no effect. If the patch pacifies Coverity, 
it's a bug in Coverity.

>       }
>   }
>   libc_hidden_def (__libc_scratch_buffer_dupfree)
> diff --git a/lib/wordsplit.c b/lib/wordsplit.c
> index 661a4f8..6ccaa2a 100644
> --- a/lib/wordsplit.c
> +++ b/lib/wordsplit.c
> @@ -615,7 +615,6 @@ coalesce_segment (struct wordsplit *wsp, struct
> wordsplit_node *node)
>            node->flags |= p->flags & _WSNF_QUOTE;
>            wsnode_remove (wsp, p);
>            stop = p == end;
> -         wsnode_free (p);
>          }
>         p = next;
>       }

This patch would cause a memory leak, no? Also, the calling code doesn't 
use p->next in this case, so this appears to be a Coverity false alarm.

> There is a compiler warning about issues in utimens.c, which I find as
> false positives.

Although report is also a false positive, it's an understandable one 
since the code uses the wrong declaration (albeit technically correct). 
I attempted to pacify Coverity by installing the following patch into 
Gnulib; please give it a try.

https://git.savannah.gnu.org/cgit/gnulib.git/commit/?id=a3a946f670718d0dee5a7425ad5ac0a29fb46ea1