Re: 70x speed-up for seq's common cases

[Bernhard Voelker]

On 09/14/2012 10:40 AM, Jim Meyering wrote:
> Bernhard Voelker wrote:
>> Shouldn't we free(buf), too?
> 
> Thanks.
> Yes, to placate leak-checking tools, but for the record, any time buf
> was allocated, we'd exit (successfully) right after return from that
> function.
> 
> Hence, I've added the free guarded by IF_LINT:
> 
> @@ -429,7 +429,7 @@ seq_fast (char const *a, char const *b)
>        if (buf < z)
>          fwrite (buf, z - buf, 1, stdout);
> 
> +      IF_LINT (free (buf));
>      }

The other 2 calls to free() could then also be IF_LINT()ed.

>>> @@ -412,6 +520,33 @@ main (int argc, char **argv)
>>
>>> [...]
>>
>>> +  if (format_str == NULL
>>> +      && all_digits_p (argv[1])
>>> +      && (n_args == 1 || all_digits_p (argv[2]))
>>> +      && (n_args < 3 || STREQ ("1", argv[3])))
>>> +    {
>>> +      char const *s1 = n_args == 1 ? "1" : argv[1];
>>> +      char const *s2 = n_args == 1 ? argv[1] : argv[2];
>>> +      if (seq_fast (s1, s2))
>>> +        exit (EXIT_SUCCESS);
>>> +
>>> +      /* Upon any failure, let the more general code deal with it.  */
>>> +    }
>>
>> seq_fast can only return ok.
> 
> True, but "ok" can be false ;-)
> I.e., with out of order arguments:
> 
>   seq 9 1

ah, right. I was tricked by the variable name ;-)

>> BTW:
>> src/seq.c: In function 'all_digits_p':
>> src/seq.c:440:1: error: function might be candidate for attribute
>> pure' if it is known to return normally
>> [-Werror=suggest-attribute=pure]
>>
>> Therefore:
>>   -static bool
>>   +static bool _GL_ATTRIBUTE_PURE
>>   all_digits_p (char const *s)
> 
> Oh!  Thank you.
> Why didn't gcc 4.8.0 20120912 tell me that one?

I don't know, mine is gcc 4.6.2.

Have a nice day,
Berny