Re: [PATCH for-4.19] tools/xentop: fix cpu% sort order

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 14/05/2024 9:13 am, Leigh Brown wrote:
> Although using integer comparison to compare doubles kind of
> works, it's annoying to see domains slightly out of order when
> sorting by cpu%.
>
> Add a compare_dbl() function and update compare_cpu_pct() to
> call it.
>
> Signed-off-by: Leigh Brown <leigh@xxxxxxxxxxxxx>
> ---
>  tools/xentop/xentop.c | 13 ++++++++++++-
>  1 file changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/tools/xentop/xentop.c b/tools/xentop/xentop.c
> index 545bd5e96d..99199caec9 100644
> --- a/tools/xentop/xentop.c
> +++ b/tools/xentop/xentop.c
> @@ -85,6 +85,7 @@ static void set_delay(const char *value);
>  static void set_prompt(const char *new_prompt, void (*func)(const char *));
>  static int handle_key(int);
>  static int compare(unsigned long long, unsigned long long);
> +static int compare_dbl(double, double);
>  static int compare_domains(xenstat_domain **, xenstat_domain **);
>  static unsigned long long tot_net_bytes( xenstat_domain *, int);
>  static bool tot_vbd_reqs(xenstat_domain *, int, unsigned long long *);
> @@ -422,6 +423,16 @@ static int compare(unsigned long long i1, unsigned long 
> long i2)
>       return 0;
>  }
>  
> +/* Compares two double precision numbers, returning -1,0,1 for <,=,> */
> +static int compare_dbl(double d1, double d2)
> +{
> +     if(d1 < d2)
> +             return -1;
> +     if(d1 > d2)
> +             return 1;
> +     return 0;
> +}
> +
>  /* Comparison function for use with qsort.  Compares two domains using the
>   * current sort field. */
>  static int compare_domains(xenstat_domain **domain1, xenstat_domain 
> **domain2)
> @@ -523,7 +534,7 @@ static double get_cpu_pct(xenstat_domain *domain)
>  
>  static int compare_cpu_pct(xenstat_domain *domain1, xenstat_domain *domain2)
>  {
> -     return -compare(get_cpu_pct(domain1), get_cpu_pct(domain2));
> +     return -compare_dbl(get_cpu_pct(domain1), get_cpu_pct(domain2));

Oh, we were doing an implicit double->unsigned long long conversion. 
Over the range 0.0 to 100.0, that ought to work as expected.  What kind
of out-of-order are you seeing?

Nevertheless, this should comparison should clearly be done using
doubles.  AFACT, get_cpu_pct() shouldn't ever return a NaN, so I think
this simple form is fine.

Oleksii: This is another bugfix to xentop, and should be considered for
4.19 at this point.

~Andrew