Re: [PATCH] trace-cmd: Optimize pretty_print() function

[Tzvetomir Stoyanov <tz.stoyanov@gmail.com>]

On Fri, May 15, 2020 at 4:14 AM Steven Rostedt <rostedt@goodmis.org> wrote:
>
[...]
> > +static int parse_arg_format(struct tep_print_parse **parse,
> > +                         struct tep_event *event,
> > +                         const char *format, struct tep_print_arg
> > **arg) +{
> > +     struct tep_print_arg *len_arg = NULL;
> > +     char print_format[32];
> > +     const char *start = format;
> > +     int ret = 0;
> > +     int ls = 0;
> > +     int res;
> > +     int len;
> > +
> > +     if (*format != '%')
> > +             return ret;
> > +     format++;
> > +
> > +     /* %% */
> > +     if (*format == '%')
> > +             return ret;
>
> The above two checks for '%' isn't needed as it is checked by the
> caller. If for some reason that the caller didn't catch it, and we
> return here,  it would return zero and the calling loop would never
> proceed. It's not protecting anything, I would just remove these two
> checks.
>
> > +     ret++;
> > +
> > +     for (; *format; format++) {
> > +             switch (*format) {
> > +             case '#':
> > +                     /* FIXME: need to handle properly */
> > +                     break;
> > +             case 'h':
> > +                     ls--;
> > +                     break;
> > +             case 'l':
> > +                     ls++;
> > +                     break;
> > +             case 'L':
> > +                     ls = 2;
> > +                     break;
> > +             case '.':
> > +             case 'z':
> > +             case 'Z':
> > +             case '0' ... '9':
> > +             case '-':
> > +                     break;
> > +             case '*':
> > +                     /* The argument is the length. */
> > +                     if (!arg) {
> > +                             do_warning_event(event, "no argument match");
> > +                             event->flags |= TEP_EVENT_FL_FAILED;
> > +                             goto out_failed;
> > +                     }
>
> Since all cases to exit this loop do the same check, I think we can
> move this check before the loop. The old code didn't do this check
> properly for %p, but that was a bug in the old code.
>

We still have to check for a valid argument inside the switch for almost all
exit cases, as the '*' case can consume the argument that would be verified
before the loop. And there is one exit that does not require a valid argument -
the default case. The old code just prints the unknown character, without
consuming an argument.

>
> > +                     len_arg = *arg;
>
[...]
> -- Steve

Thanks, Steve!

-- 
Tzvetomir (Ceco) Stoyanov
VMware Open Source Technology Center