Re: [PATCH 4/5] perf diff: Fix output ordering to honor next column

[Jiri Olsa <jolsa@redhat.com>]

On Sat, Dec 27, 2014 at 02:06:32PM +0900, Namhyung Kim wrote:

SNIP

>  	if (!pairs_left || !pairs_right)
>  		return pairs_left ? -1 : 1;
>  
> -	p_left  = get_pair_data(left,  &data__files[sort_compute]);
> -	p_right = get_pair_data(right, &data__files[sort_compute]);
> +	p_left  = get_pair_data(left,  &data__files[sort_idx]);
> +	p_right = get_pair_data(right, &data__files[sort_idx]);
>  
>  	if (!p_left && !p_right)
>  		return 0;
> @@ -555,8 +560,13 @@ hist_entry__cmp_nop(struct hist_entry *left __maybe_unused,
>  static int64_t
>  hist_entry__cmp_baseline(struct hist_entry *left, struct hist_entry *right)
>  {
> -	if (sort_compute)
> -		return 0;
> +	/*
> +	 * This function will be called first for each entry to resort
> +	 * output.  Next compare-functions use this idx to find their
> +	 * data and increase it for next data so we need to initialize
> +	 * it everytime.
> +	 */
> +	data_idx = 0;

hum, could we omit the global 'data_idx' variable usage by passing
'struct perf_hpp_fmt' into color fmt callbacks? (like hist_entry__cmp_delta..)

we could get the data_idx from 'struct diff_hpp_fmt'::idx

jirka