Re: [PATCH nft] mergesort: avoid cloning value in expr_msort_cmp()

[Pablo Neira Ayuso <pablo@netfilter.org>]

Hi Thomas,

Thanks for your patch.

On Wed, Sep 27, 2023 at 08:59:34AM +0200, Thomas Haller wrote:
> If we have a plain EXPR_VALUE value, there is no need to copy
> it via mpz_set().
> 
> Signed-off-by: Thomas Haller <thaller@redhat.com>
> ---
>  src/mergesort.c | 34 ++++++++++++++++++----------------
>  1 file changed, 18 insertions(+), 16 deletions(-)
> 
> diff --git a/src/mergesort.c b/src/mergesort.c
> index 5965236af6b7..af7f163b2779 100644
> --- a/src/mergesort.c
> +++ b/src/mergesort.c
> @@ -12,8 +12,6 @@
>  #include <gmputil.h>
>  #include <list.h>
>  
> -static void expr_msort_value(const struct expr *expr, mpz_t value);
> -
>  static void concat_expr_msort_value(const struct expr *expr, mpz_t value)
>  {
>  	unsigned int len = 0, ilen;
> @@ -29,20 +27,20 @@ static void concat_expr_msort_value(const struct expr *expr, mpz_t value)
>  	mpz_import_data(value, data, BYTEORDER_HOST_ENDIAN, len);
>  }
>  
> -static void expr_msort_value(const struct expr *expr, mpz_t value)
> +static mpz_srcptr expr_msort_value(const struct expr *expr, mpz_t value)
>  {
> +recursive_again:

We are in userspace, recursion is possible. Any chance to avoid the
copy without this goto approach?