Re: top on machines with more than 9 CPUs

[christos%astron.com@localhost (Christos Zoulas)]

In article <20100624145307.GM12982%beverly.kleinbus.org@localhost>,
Ignatios Souvatzis  <is%netbsd.org@localhost> wrote:
>Hi,
>
>one of my users at work, monitoring his computation on our 12-CPU amd64,
>noticed that the output of top is slightly distorted.
>
>I investigated, and found that this is caused by a formatting that only
>checks for 1 or more than 1 CPU, but assumes fixed column positions.
>
>Also, the width computation is off by one per column, leading to 
>unnecessary early switch to the abbreviated format.
>
>I found that changing top/dist/display.c to thus:
>
>diff -u -r1.7 display.c
>--- display.c   5 May 2009 18:52:13 -0000       1.7
>+++ display.c   24 Jun 2010 14:35:11 -0000
>@@ -854,7 +854,8 @@
>            *ip++ = cpustate_total_length;
>            if ((i = strlen(*pp++)) > 0)
>            {
>-               cpustate_total_length += i + 8;
>+               cpustate_total_length += i + 7;
>+               /* strlen(" 100% ") is 6, strlen(" 99.9% ") is 7. Never 8. */
>            }
>        }
>     }
>
>and to change cpustates_tag() to read (sorry, the diff is longer and
>unreadable).
>
>static char *
>cpustates_tag(int c)
> 
>{
>    unsigned width, u;
>    static char fmttag[100]; 
> 
>    const char *short_tag = !multi || ncpu <= 1 ? "CPU: " : "CPU%0*d";
>
>    const char *long_tag = !multi || ncpu <= 1 ?
>       "CPU states: " : "CPU%0*d states: ";
>
>    for (width=0, u=ncpu; u>0; u /= 10) {
>        ++width;
>    }
>    /* if length + strlen(long_tag) > screen_width, then we have to
>       use the shorter tag */
>
>    snprintf(fmttag, sizeof(fmttag), long_tag, width, c);
>
>    if (cpustate_total_length + (signed)strlen(fmttag)  > screen_width) {
>        snprintf(fmttag, sizeof(fmttag), short_tag, width, c);
>    }
>
>    /* set x_cpustates accordingly then return result */
>    x_cpustates = strlen(fmttag);
>    return(fmttag);
>}
>
>fixes both problems. I'll commit next week unless serious objections 
>arise. Of course I'll feed the change upstreams.

I thought it could be more than 100% like 204.3%

christos